Coordination in Service Value Networks : A Mechanism Design Approach

Blau, Benjamin Sebastian

2011

KIT Scientific Publishing

The fundamental paradigm shift from traditional value chains to agile service value networks (SVN) implies new economic and organizational challenges. This work provides an auction-based coordination mechanism that enables the allocation and pricing of service compositions in SVNs. The mechanism is multidimensional incentive compatible and implements an ex-post service level enforcement. Further extensions of the mechanism are evaluated following analytical and numerical research methods.

Business

Management

English

978-3-86644-724-0 (print)

Studies on eOrganisation and Market Engineering 13

Benjamin Sebastian Blau

Coordination in Service
Value Networks
A Mechanism Design Approach

Benjamin Sebastian Blau
Coordination in Service Value Networks
A Mechanism Design Approach

Studies on eOrganisation and Market Engineering
Karlsruher Institut für Technologie
Herausgeber:
Prof. Dr. Christof Weinhardt
Prof. Dr. Thomas Dreier
Prof. Dr. Rudi Studer

13

Coordination in Service Value Networks
A Mechanism Design Approach

by
Benjamin Sebastian Blau

Dissertation, Karlsruher Institut für Technologie
Fakultät für Wirtschaftswissenschaften, 2009
Referenten: Prof. Dr. Christof Weinhardt, Prof. Dr. Rudi Studer

Impressum
Karlsruher Institut für Technologie (KIT)
KIT Scientific Publishing
Straße am Forum 2
D-76131 Karlsruhe
www.ksp.kit.edu
KIT – Universität des Landes Baden-Württemberg und nationales
Forschungszentrum in der Helmholtz-Gemeinschaft

Diese Veröffentlichung ist im Internet unter folgender Creative Commons-Lizenz
publiziert: http://creativecommons.org/licenses/by-nc-nd/3.0/de/

KIT Scientific Publishing 2011
Print on Demand
ISSN 1862-8893
ISBN 978-3-86644-724-0

Coordination in Service Value
Networks
A Mechanism Design Approach

Zur Erlangung des akademischen Grades eines
Doktors der Wirtschaftswissenschaften
(Dr. rer. pol.)
von der Fakultät für
Wirtschaftswissenschaften
der Universität Karlsruhe (TH)
genehmigte
Dissertation

von

Dipl.-Inform.Wirt Benjamin Sebastian Blau

Tag der mündlichen Prüfung: 31.07.2009
Referent: Prof. Dr. Christof Weinhardt
Korreferent: Prof. Dr. Rudi Studer
Prüfer: Prof. Dr. Oliver Stein

2009 Karlsruhe

Abstract
The fundamental paradigm shift from traditional value chains to agile service
value networks (SVN) implies new economic and organizational challenges. In
service value networks, a multitude of participants co-create complex services
that create added value for customers by providing highly specialized service
components and by leveraging lightweight paradigms such as RESTful architectures and mashup technologies. Addressing the challenge of coordinating distributed activities in order to achieve a desired outcome, auctions have proven to
perform quite well in situations where intangible and heterogeneous economic
entities are traded [Smi89, LR00].
Nevertheless, traditional approaches in the area of multidimensional combinatorial auctions [BK05, Sch07] are not quite suitable to enable the trade of composite services. A flawless service execution and therefore the requester’s valuation highly depends on the accurate sequence of the functional parts of the
composition, meaning that in contrary to service bundles, composite services
only generate value through a valid order of their components. From a technical
perspective, service composition research [ZBD+ 03] traditionally assumes complete information about QoS characteristics and prices and does not account for
self-interested service owners that intent to maximize their utility and therefore
behave strategically.
Addressing these challenges, in the work at hand, the complex service auction
(CSA) is developed following a mechanism design approach. The auction mechanism facilitates the allocation of multidimensional service offers within service
value networks, enables service level enforcement and determines prices for complex services. The mechanism and the bidding language support various types
of QoS characteristics and their individual aggregation by incorporating semantic
information. Compliant with state of the art standards such as WS-Coordination,
a possible implementation of the complex service auction in distributed environments is presented and a computational tractable algorithm to solve the winner
determination problem is introduced.

ii
Leveraging analytical and numerical research methods, the mechanism’s
properties are evaluated comprehensively. It is analytically shown that the social
choice implemented by the complex service auction is incentive compatible with
respect to all dimensions of the service offer (quality and price), i.e. although
service providers act strategic, it is a weakly dominant strategy to report their
multidimensional type truthfully to the auctioneer. Counteracting the absence of
budget balance, a payment scheme is presented which is robust to manipulation
and at the same time incentivizes service providers to increase their services’ degree of interoperability which is shown by means of an agent-based simulation.
To leverage synergies and to reduce costs, it is beneficial for service providers under certain circumstances to offer bundled services. Depending on how service
providers are situated within a service value network, bundling and unbundling
strategies are analyzed following a simulation approach.

Acknowledgements
This work would not have been possible without the guidance and support of
many people. I would like to thank my advisor Professor Dr. Christof Weinhardt
for giving me the great opportunity to do this work and for his constant support
and innovative ideas. He granted me the freedom and the help necessary and
encouraged me during in times.
Additionally, I would like to thank my co-advisor Professor Dr. Rudi Studer
for his guidance and fruitful discussions that improved and enriched especially
the technical elements of my work. Thanks also to the other members of the committee, Professor Dr. Oliver Stein and Professor Dr. Stefan Tai who in particular
sensitized me to additional technical aspects to round up this work.
I would like to thank the outstanding team of the research group on Information and Market Engineering at the Institute of Information Systems and Management (IISM) and the colleagues of the Karlsruhe Service Research Institute (KSRI).
Their inspiration and valuable comments significantly improved my work and
helped me to solve initially “unsolvable” problems. I would also like to thank
Professor Dr. Dirk Neumann for his support in the early stage of this research
and his seminal ideas. In particular I am grateful to my friends Tobias Conte
and Jochen Stößer for proof reading major parts of this work and especially for
providing me with critical and constructive questions and comments.
Above all, I am indebted to my parents, Thomas Blau and Heide Blau, to my
sister Alexandra Blau, and to my fiancée Katharina Gofron. This work would not
have been possible without their constant support and their caring encouragement.

Benjamin Blau

Contents
I Foundations
1 Introduction
1.1 Motivation . . . . . . . . . . . . . . . .
1.2 Research Outline . . . . . . . . . . . .
1.3 Structure . . . . . . . . . . . . . . . . .
1.4 Publications & Research Development

1
.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

2 Preliminaries & Related Work
2.1 Service Concepts, Definitions, and Technologies . . . . . . . .
2.1.1 Tangibles, Intangibles, and Services . . . . . . . . . . .
2.1.1.1 Tangible and Intangible Goods . . . . . . . . .
2.1.1.2 Services . . . . . . . . . . . . . . . . . . . . . .
2.1.1.3 E-Services . . . . . . . . . . . . . . . . . . . . .
2.1.2 Service Decomposition Model . . . . . . . . . . . . . . .
2.1.2.1 Utility Services . . . . . . . . . . . . . . . . . .
2.1.2.2 Elementary Services . . . . . . . . . . . . . . .
2.1.2.3 Complex Services . . . . . . . . . . . . . . . . .
2.1.3 Service-Oriented Architectures . . . . . . . . . . . . . .
2.1.3.1 Basic Concepts . . . . . . . . . . . . . . . . . .
2.1.3.2 Web Services . . . . . . . . . . . . . . . . . . .
2.1.3.3 Quality of Service (QoS) . . . . . . . . . . . . .
2.1.3.4 Web Service Coordination . . . . . . . . . . . .
2.1.4 Service Value Networks and Situational Applications .
2.1.4.1 Networks as a Type of Governance Form . . .
2.1.4.2 Service Value Networks . . . . . . . . . . . . .
2.1.4.3 Situational Applications and Service Mashups
2.2 Markets in a Service World . . . . . . . . . . . . . . . . . . . . .
2.2.1 Why Auctions for Complex Services? . . . . . . . . . .
2.2.2 Electronic Markets and Market Engineering . . . . . . .
2.2.2.1 Environmental Analysis . . . . . . . . . . . . .
2.2.2.2 Design and Implementation . . . . . . . . . .
2.2.2.3 Testing and Evaluation . . . . . . . . . . . . .
2.2.2.4 Introduction . . . . . . . . . . . . . . . . . . . .

.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

3
3
6
10
12

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

15
16
17
18
19
22
25
25
26
27
32
32
37
46
48
53
54
55
62
66
67
69
71
72
73
73

vi

CONTENTS
2.2.3

2.3

Mechanism Design . . . . . . . . . . . . . . . . . . . . . . . .
2.2.3.1 Social Choice . . . . . . . . . . . . . . . . . . . . . .
2.2.3.2 Properties of Social Choice and Mechanism Implementations . . . . . . . . . . . . . . . . . . . . . . .
2.2.3.3 Possibility Results . . . . . . . . . . . . . . . . . . .
2.2.3.4 Impossibility Results . . . . . . . . . . . . . . . . . .
2.2.3.5 Algorithmic Mechanism Design . . . . . . . . . . .
2.2.4 Environmental Analysis and Related Work . . . . . . . . . .
2.2.4.1 Requirements . . . . . . . . . . . . . . . . . . . . . .
2.2.4.2 Related Work . . . . . . . . . . . . . . . . . . . . . .
Research Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.1 Theoretical Analysis . . . . . . . . . . . . . . . . . . . . . . .
2.3.2 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . .

73
74
77
79
82
83
83
83
86
88
89
89

II Design & Implementation

91

3 Complex Service Auction (CSA)
3.1 Service Value Network Model . . . . .
3.2 Bidding Language . . . . . . . . . . . .
3.2.1 Scoring Function . . . . . . . .
3.2.2 Service Requests . . . . . . . . .
3.2.3 Service Offers . . . . . . . . . .
3.3 Mechanism Implementation . . . . . .
3.3.1 Allocation . . . . . . . . . . . .
3.3.2 Transfer . . . . . . . . . . . . . .
3.3.3 Summary . . . . . . . . . . . . .
3.4 Related Work . . . . . . . . . . . . . . .
3.5 Auction Process Model & Architecture
3.6 Realization & Implementation . . . . .

.
.
.
.
.
.
.
.
.
.
.
.

93
95
98
99
103
104
106
107
108
109
110
112
115

.
.
.
.
.
.
.
.
.
.
.

123
124
124
125
128
130
130
133
134
134
136
136

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

4 Applicability Extensions
4.1 Verification and Service Level Enforcement . . . .
4.1.1 Related Work . . . . . . . . . . . . . . . . .
4.1.2 Compensation . . . . . . . . . . . . . . . . .
4.2 Achieving Budget Balance . . . . . . . . . . . . . .
4.2.1 Related Work . . . . . . . . . . . . . . . . .
4.2.2 Interoperability Transfer . . . . . . . . . . .
4.2.3 Finding the Optimal Threshold Parameter .
4.2.4 Summary . . . . . . . . . . . . . . . . . . . .
4.3 Managing Service Quality . . . . . . . . . . . . . .
4.3.1 Knowledge Representation Formalisms . .
4.3.2 Semantic QoS Management . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

CONTENTS

vii

III Evaluation

141

5 Analytical Results
5.1 Incentive Compatibility & Individual Rationality . .
5.1.1 One-Dimensional Bids in the Basic CSA . . .
5.1.2 Multidimensional Bids in the Extended CSA
5.1.3 Results & Implications . . . . . . . . . . . . .
5.2 Cooperation within the Value Chain . . . . . . . . .
5.2.1 Related Work . . . . . . . . . . . . . . . . . .
5.2.2 A Model of Cooperation . . . . . . . . . . . .

.
.
.
.
.
.
.

143
143
144
146
149
150
150
150

.
.
.
.
.
.
.
.
.
.
.
.
.

155
155
156
158
165
167
168
171
175
176
179
182
183
191

6 Numerical Results
6.1 Manipulation Robustness of the ITF Extension
6.1.1 Simulation Model . . . . . . . . . . . . .
6.1.2 Results . . . . . . . . . . . . . . . . . . .
6.1.3 Implications . . . . . . . . . . . . . . . .
6.2 Incentivizing Interoperability Endeavors . . . .
6.2.1 Simulation Model . . . . . . . . . . . . .
6.2.2 Results . . . . . . . . . . . . . . . . . . .
6.2.3 Implications . . . . . . . . . . . . . . . .
6.3 Bundling Strategies of Service Providers . . . .
6.3.1 Simulation Model . . . . . . . . . . . . .
6.3.2 Simulation Settings . . . . . . . . . . . .
6.3.3 Results & Implications . . . . . . . . . .
6.3.4 Strategic Recommendations . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

IV Finale

193

7 Conclusion & Outlook
7.1 Contribution . . . . . . . .
7.2 Open Questions . . . . . .
7.3 Complementary Research
7.4 Final Remarks . . . . . . .

.
.
.
.

195
195
200
202
205

.
.
.
.
.

207
207
208
209
210
218

A Appendix
A.1 Formal Notation . . . . . .
A.2 Incentive Compatibility . .
A.3 Allocative Efficiency . . .
A.4 Manipulation Robustness
A.5 Bundling Strategies . . . .
References

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

218

List of Figures
1.1

Structure of this work. . . . . . . . . . . . . . . . . . . . . . . . . . .

11

2.1
2.2
2.3
2.4
2.5
2.6

Service lifecycle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Service decomposition model. . . . . . . . . . . . . . . . . . . . . . .
Business scenario integrating a payment processing service. . . . .
Payment processing service (static view). . . . . . . . . . . . . . . .
Payment processing service (dynamic view). . . . . . . . . . . . . .
Business scenario “Service Request and Order Management”
(SROM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Roles and primary operations in service-oriented architectures. . .
SOA layers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Web service technology stack. . . . . . . . . . . . . . . . . . . . . . .
Service orchestration versus service choreography. . . . . . . . . . .
WS-Coordination sequence diagram. . . . . . . . . . . . . . . . . . .
Mapping of a reverse auction to a coordination model. . . . . . . . .
Service value network model. . . . . . . . . . . . . . . . . . . . . . .
Example of a service value network realizing a CRM complex service.
Situational applications address the long tail of business. . . . . . .
Blueprint of a translation and tagging service mashup. . . . . . . .
Characteristics of products and services affect forms of organization.
Stages of the market engineering process. . . . . . . . . . . . . . . .
Triangle relation of mechanism implementation and social choice. .

20
26
28
29
30

2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14
2.15
2.16
2.17
2.18
2.19
3.1
3.2
3.3
3.4
3.5

Framework for the design of mechanisms. . . . . . . . . . . . . . . .
Statechart formalization. . . . . . . . . . . . . . . . . . . . . . . . . .
Context-dependent cost structures of service providers. . . . . . . .
Service value network model. . . . . . . . . . . . . . . . . . . . . . .
Service value network with service offers and corresponding configurations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6 Requester utility for different attribute types. . . . . . . . . . . . . .
3.7 Service value network with service offers and internal costs. . . . .
3.8 Critical value and individual contribution. . . . . . . . . . . . . . . .
3.9 Triangle relation of the CSA mechanism implementation and social
choice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10 Process model of the CSA. . . . . . . . . . . . . . . . . . . . . . . . .
3.11 Architectural overview of the CSA. . . . . . . . . . . . . . . . . . . .

31
34
36
40
43
49
53
57
61
63
65
70
71
76
95
96
97
99
102
103
105
108
110
112
114

LIST OF FIGURES

ix

3.12 Performance analysis of the ComputeAllocation algorithm. . . . . . 119
3.13 Service value network with service offers exposing memorydependent attribute types. . . . . . . . . . . . . . . . . . . . . . . . . 120
4.1
4.2
4.3
4.4
5.1
5.2

Service value network with service offers characterized
rate quality attributes. . . . . . . . . . . . . . . . . . . . .
Non-budget-balanced outcome of the CSA. . . . . . . . .
Service value network with semantic QoS characteristics.
Security encryption ontology. . . . . . . . . . . . . . . . .

by
. .
. .
. .
. .

error
. . . .
. . . .
. . . .
. . . .

127
129
137
138

Cost dependency between service provider sy and sz . . . . . . . . . 151
Cooperation within the value chain of a payment processing complex service. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

6.1

Simulation model for the evaluation of manipulation robustness
using the ITF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
6.2 Decision tree of service providers. . . . . . . . . . . . . . . . . . . . . 159
6.3 Utility for a single manipulating service provider in different competition scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.4 Simulation model for the evaluation of interoperability incentives
using the ITF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.5 Interoperability degrees (ID) for 20 service offers in 4 candidate pools.173
6.6 Beneficial bundling strategy (ex-ante case). . . . . . . . . . . . . . . 177
6.7 Beneficial bundling strategy (ex-post case) . . . . . . . . . . . . . . . 178
6.8 Simulation model for the evaluation of bundling and unbundling
strategies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.9 Relative frequencies and expected payoffs of bundling and unbundling strategies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
6.10 Strategy fitness in different cost reduction scenarios with 20 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . . 189
6.11 Strategy fitness in different cost reduction scenarios with 28 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . . 190
7.1

Multi-layered market for complex services and resources. . . . . . . 203

A.1 Strategy fitness in different cost reduction scenarios with 32 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . . 219

List of Tables
2.1

2.3

Differentiation criteria of tangibles, intangibles, services, and eservices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
SaaS providers for CRM, SCM and FIN components of the business
scenario SROM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Requirements satisfaction degree of related approaches. . . . . . . .

3.1
3.2

Aggregation operations for different attribute types. . . . . . . . . . 100
Allocation computation stepwise procedure example. . . . . . . . . 121

5.1

Cooperation decision as a normal form game. . . . . . . . . . . . . . 152

6.1

Utility for a single manipulating service provider with 12 service
offers in 4 candidate pools (condensed). . . . . . . . . . . . . . . . . 160
Utility for a single manipulating service provider with 16 service
offers in 4 candidate pools (condensed). . . . . . . . . . . . . . . . . 161
Utility for a single manipulating service provider with 16 service
offers in 4 candidate pools (condensed). . . . . . . . . . . . . . . . . 162
Utility for a single manipulating service provider with 20 service
offers in 4 candidate pools (condensed). . . . . . . . . . . . . . . . . 162
Utility for a single manipulating service provider with 20 service
offers in 4 candidate pools (condensed). . . . . . . . . . . . . . . . . 163
Utility for a single manipulating service provider with 28 service
offers in 4 candidate pools (condensed). . . . . . . . . . . . . . . . . 163
Utility for a single manipulating service provider with 28 service
offers in 4 candidate pools (condensed). . . . . . . . . . . . . . . . . 164
Interoperability degrees (ID) for 20 service offers in 4 candidate pools.171
Interoperability degrees (ID) for 20 service offers in 4 candidate pools.172
Interoperability degrees (ID) for 32 service offers in 4 candidate pools.174
Analyzed events for the evaluation of bundling and unbundling
strategies of service providers. . . . . . . . . . . . . . . . . . . . . . . 182
Simulation settings for the evaluation of bundling and unbundling
strategies of service providers. . . . . . . . . . . . . . . . . . . . . . . 183
Evaluation of bundling and unbundling strategies of service
providers with 20 service offers in 4 candidate pools and 0% cost
reduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

2.2

6.2
6.2
6.3
6.3
6.4
6.4
6.5
6.5
6.6
6.7
6.8
6.9

25
31
88

LIST OF TABLES

xi

6.10 Evaluation of bundling and unbundling strategies of service
providers with 20 service offers in 4 candidate pools and 50% cost
reduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
6.11 Evaluation of bundling and unbundling strategies of service
providers with 28 service offers in 4 candidate pools and 0% cost
reduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
6.12 Evaluation of bundling and unbundling strategies of service
providers with 28 service offers in 4 candidate pools and 50% cost
reduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
A.1 Notation of abstract model and mechanism implementation. . . . .
A.1 Notation of abstract model and mechanism implementation. . . . .
A.2 Utility for a single manipulating service provider with 12 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.2 Utility for a single manipulating service provider with 12 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.2 Utility for a single manipulating service provider with 12 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.3 Utility for a single manipulating service provider with 16 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.3 Utility for a single manipulating service provider with 16 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.3 Utility for a single manipulating service provider with 16 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.4 Utility for a single manipulating service provider with 20 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.4 Utility for a single manipulating service provider with 20 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.4 Utility for a single manipulating service provider with 20 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.5 Utility for a single manipulating service provider with 28 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.5 Utility for a single manipulating service provider with 28 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .
A.5 Utility for a single manipulating service provider with 28 service
offers in 4 candidate pools. . . . . . . . . . . . . . . . . . . . . . . . .

207
208
210
211
212
212
213
214
214
215
216
216
217
218

List of Abbreviations
ACID . . . . . . . . . . .
B2B . . . . . . . . . . . . .
BN . . . . . . . . . . . . . .
BPEL . . . . . . . . . . . .
CRM . . . . . . . . . . . .
CTF . . . . . . . . . . . . .
FIN . . . . . . . . . . . . .
FOL . . . . . . . . . . . . .
FTP . . . . . . . . . . . . .
GXL . . . . . . . . . . . . .
HTML . . . . . . . . . . .
HTTP . . . . . . . . . . .
ICT . . . . . . . . . . . . . .
IT . . . . . . . . . . . . . . .
JSON . . . . . . . . . . . .
QoS . . . . . . . . . . . . .
RDF . . . . . . . . . . . . .
REST . . . . . . . . . . . .
RPC . . . . . . . . . . . . .
RSS . . . . . . . . . . . . .
SaaS . . . . . . . . . . . . .
SBN . . . . . . . . . . . . .
SCM . . . . . . . . . . . .
SemPIT . . . . . . . . . .
SLA . . . . . . . . . . . . .
SMTP . . . . . . . . . . .
SOA . . . . . . . . . . . . .
SOAP . . . . . . . . . . .
SROM . . . . . . . . . . .
SVN . . . . . . . . . . . . .
SVNP . . . . . . . . . . .
UDDI . . . . . . . . . . .
UML . . . . . . . . . . . .
URI . . . . . . . . . . . . .

Atomicity, Consistency, Isolation, Durability
Business-to-Business
Business Network
Business Process Execution Language
Customer Relationship Management
Compatibility Transfer Function
Finance
First-Order Logic
File Transfer Protocol
Graph eXchange Language
Hypertext Markup Language
Hypertext Transfer Protocol
Information and Communication Technology
Information Technology
JavaScript Object Notation
Quality of Service
Resource Description Framework
Representational State Transfer
Remote Procedure Call
Rich Site Summary
Software-as-a-Service
Smart Business Network
Supply Chain Management
Semantic and Policy-Based IT Management and Provisioning
Service Level Agreement
Simple Mail Transfer Protocol
Service-oriented Architecture
Simple Object Access Protocol
Service Request and Order Management
Service Value Network
Service Value Network Planner
Universal Description, Discovery, and Integration
Unified Modeling Language
Uniform Resource Identifier

xiv
VCG . . . . . . . . . . . .
VO . . . . . . . . . . . . . .
W3C . . . . . . . . . . . .
WADL . . . . . . . . . .
WSDL . . . . . . . . . . .
XML . . . . . . . . . . . .

LIST OF TABLES
Vickrey-Clarke-Groves
Virtual Organization
World Wide Web Consortium
Web Application Description Language
Web Service Description Language
eXtensible Markup Language

Part I
Foundations

Chapter 1
Introduction
The principle of utility neither requires nor admits of any other regulator than itself.
[Ben38]

his chapter firstly motivates the work at hand in Section 1.1 and elaborates
arguments that support the necessity and relevance of the addressed research questions. Section 1.2 describes the research outline and the research questions underlying this work. Based on the construction of the research outline,
Section 1.3 briefly introduces the main structure followed by an illustration of the
research development with respect to publications and presentations of different
parts of this work.

T

1.1 Motivation
Businesses are undergoing a paradigm shift from developing and distributing
goods to providing services as their core business [VL04]. As the focus on service
customization increases in order to provide tailored-solutions to customers, companies gain competitive advantage through the provision of highly specialized
services [VL04, LVO07]. In recent years the service sector has become a rapidly
growing sector in world economies. In Brazil, Russia, Japan, and Germany, services account for 50 percent of the labor force and 75 percent of the labor force
in the United Kingdom and the United States [OEC05]. The Bureau of Economic
Analysis (BEA) reported that in the United States, the private service-producing
sector continued to lead overall GDP growth in 2006, increasing by 4.2 percent,

4

CHAPTER 1. INTRODUCTION

whereas growth in the private goods-producing sector decreased down to 0.8
percent [BEA08].
A renaissance of HTTP appreciation through e.g. the RESTful architectural
style [Fie00, RR07] drives simplicity of service descriptions and interfaces and
enables service consumers to participate in the so called programmable Web. A
primer example for this trend is Amazon’s Simple Storage Service (S3)1 that is
fully accessible and manageable through basic HTTP methods following a RESTful architectural style2 . Programmatic access to services with lightweight APIs
can be used by consumers without in-depth technical knowledge. In January
2008, Amazon announced that the Amazon Web Services3 consume more bandwidth than the entire global network of Amazon.com retail sites [Ama08]. This reflects the shift from the production and consumption of statically presented information to ”living“ information services. Knowledge and information is more and
more intensively shared by building situational services (e.g. service mashups, intelligent document mashups, situational applications) instead of statically predefined information goods (e.g. blog posts, information on static Web sites). Driven
by simplicity and easy-of-use, this trend also implies a strong involvement of the
service consumer in the production process of services. The process of consuming
and contributing to service artifacts is no longer separable which results in a new
role called the service prosumer who co-creates value proactively [TW06]. As the
provision and consumption of services blurs, the number of co-created services
increases rapidly.
Due to growing modularization and simplicity, services are composable in a
plug-and-play fashion [VvHPP05, ZBD+ 03] in order to be rearranged into valueadded complex services. The process of composing and rearranging existing and
newly created service components enables agile innovation processes [BC00]. All
these trends foster a rapid growth of so called service value networks. Service
value networks are constituted by loosely-coupled formations of companies that
provide modularized services while concentrating on their core competencies.
These Web-enabled services expose standardized interfaces and foster an ad-hoc
composition in order to jointly generate added value for customers in an ondemand fashion.
Service composition enabled through modularization and simplicity leverages the power of business in the long tail [And06]. Flexible combining cus1 http://aws.amazon.com/s3/
2A

detailed introduction to the Amazon S3 architecture and the programmatic management
can be found in [RR07]
3 http://aws.amazon.com/

1.1. MOTIVATION

5

tomized service components increases variety and individuality which leverages
the power of mass-customization [DSBF01]. Traditionally, most of the individual
demand for specialized services could not be satisfied by off-the-shelf solutions.
By enabling the opportunity to co-create solutions and building nearly unlimited versions through innovating and recomposing loosely-coupled services into
value-added complex services, demand is nearly generated by customers themselves.
Nevertheless, current leading service providers traditionally offer their services charging static prices (e.g. pay-per-use or flat fees). However, such static
pricing models do not reflect the agility and distributed nature of service value
networks and situational applications from an economic perspective. Multiple
distributed self-interested providers that contribute to a value-added complex
service have different preferences for different outcomes which are private information. Static pricing schemes ignore such preferences and additional information that is inherent in the market. Although service providers like Amazon start
to incorporate economies of scale in their pricing models [BBT09] these pricing
schemes are still static and are not capable of balancing supply and demand. A
primer example for dynamic pricing models in the context of electronic services
is Google’s AdWords4 and Yahoo! Search Marketing5 . Google for example provides a generalized second price auction to allocate and price keywords and corresponding search rankings [EOS07, Var09]. In the first quarter of 2009, 67 percent
of Google’s revenues are realized by the AdWords campaign and further 30 percent through the complementary AdSense program reflecting Google’s partner
network6 . In total, Google’s revenue is predominantly generated (97 percent)
through its advertisement programs that are based on an auction pricing model
[EOS07].
Auctions have proven to perform quite well in situations where intangible
and heterogenous entities are traded [Smi89]. Furthermore, valuations are hard
to determine for single and especially value-added complex services as the value
of the service’s outcome highly depends on the customer’s preferences for which
current pricing models do not account. Auctions are predestinated to aggregate
information from distributed parties which results in an aggregated valuation
[PS00, Jac03]. Without prior knowledge about the valuations of each participant, auctions can provide suitable incentives to make truth-revelation an equi-

4 http://adwords.google.com/
5 http://searchmarketing.yahoo.com/
6 http://investor.google.com/releases/2009Q1_google_earnings.html

6

CHAPTER 1. INTRODUCTION

librium strategy and therefore automatically aggregate necessary information from
self-interested participants to determine adequate prices for complex services.

1.2

Research Outline

The overall question underlying this work is how an adequate auction mechanism can be designed which enables the trade of complex (composite) services
in distributed environments such as service value networks. A suitable mechanism must satisfy economic and applicability requirements and must at the
same time be theoretically sound. A well-known result from Market Engineering states that there is no such thing as an omnipotent mechanism that is suitable
and applicable in any domain and any setting [WHN03]. Thus, a mechanism
design for the allocation and pricing of complex services depends on economic
and technical characteristics of typical service offers in service value networks
(e.g. utility and elementary services with different QoS characteristics), different requesters’ preferences for various QoS characteristics of complex services
[ZBD+ 03] and the overall goals of the mechanism designer (e.g. revenue vs. welfare maximization) [Rot02, Neu04]. Addressing these challenges and satisfying
detailed requirements derived from an environmental analysis, the work at hand
extends the body of research on mechanisms for trading combinatorial entities
with special focus on sequential compositions of service components in service
value networks.
The first research question deals with the properties of service value networks
and complex services which embody the final outcome that is provisioned to service requesters. As an initial step, this question lays the groundwork for the
design of an adequate mechanism that enables the trade of service compositions
in service value networks. Hence, the first research question is stated as follows:

Research Question 1 ≺ E NVIRONMENTAL A NALYSIS ≻ . What are
the characteristics of service value networks and complex services, and
what are resulting economic and applicability requirements upon a mechanism to coordinate value creation?
The question is addressed by (i) defining traditional services, e-service, software
services and Web services and analyzing their key characteristics, (ii) providing a
clear understanding of service value networks by defining their characteristics, their

1.2. RESEARCH OUTLINE

7

structure, and their components and filling the lack of definitions in current related literature (iii) analyzing the concept of a complex services as a final outcome
created by a service value network through the realization of a sequence of modularized service offers. Finally, based on these results, economic and applicability
requirements upon an adequate mechanism for coordinating value creation in
service value networks are derived. In summary, the environmental analysis and
resulting requirement analysis serve as a starting point for the further development of the work at hand.
Targeting the core contribution of this work, the second research question addresses the challenge of how to design an adequate multidimensional and scalable auction mechanism which enables the allocation and pricing of complex services in service value networks.

Research Question 2 ≺ M ECHANISM D ESIGN ≻ . How can a scalable,
multidimensional auction mechanism for allocating and pricing of complex services in service value networks be designed that limits strategic
behavior of service providers?
The question is addressed by (i) providing an abstract model of service value networks that captures the key characteristics and components in a comprehensive
manner, (ii) designing a bidding language that enables the specification of multidimensional service offers and service requests, (iii) specifying a scoring function to
capture the service requester’s preferences for different QoS characteristics and
prices of complex services and (iv) designing an auction mechanism – the Complex
Service Auction (CSA) – consisting of an allocation and transfer function that
implements an allocative efficient, individual rational and incentive compatible
social choice with respect to all dimensions of the providers’ bids. Focusing on
a computational tractable implementation of the auction mechanism, (v) an algorithm is presented that solves the winner determination problem in polynomial
time regarding the number of service offers and feasible service compositions.
While traditional service composition approaches assume complete information about the service components and their providers [ZBD+ 03], service value
networks are characterized by self-interested service providers that try to maximize their individual utility. Pursuing individual goals, service providers act
strategically and have private information about their preferences for different
outcomes [NR01, Par01] (e.g. information about true valuations and QoS char-

8

CHAPTER 1. INTRODUCTION

acteristics of their services is private an cannot be assumed to be truthfully reported). Bridging this information gap, the approach of mechanism design targets the implementation of incentives (e.g. by means of an auction mechanism)
that make truth-revelation a dominant strategy equilibrium and consequently allows for computing a system-wide solution. Nevertheless, traditional combinatorial auctions [BK05, Sch07] and especially corresponding bidding languages are
not quite suitable to enable the trade of complex services. A flawless service execution and the requester’s valuation for the outcome highly depends on the accurate sequence of the functional parts of the composition, meaning that in contrary
to service bundles, complex services only generate value through a valid order of
their components.
In order to enable the mechanism’s application to the domain of service value
networks and the coordination of distributed service activities, the following research question states the challenges regarding necessary applicability extensions
to be addressed by this work:

Research Question 3 ≺ A PPLICABILITY E XTENSIONS ≻ . How can an
auction mechanism be extended to support complex QoS characteristics
and service level enforcement? How can the pricing scheme be modified in
order to achieve budget balance and incentivize interoperability endeavors
of service providers?
Providing highly specialized services, providers shift from price to quality
competition [Pap08]. Addressing the long tail of business, service providers tend
to offer various customized versions of their services at different QoS levels in order to satisfy varying idiosyncratic demands. Consequently, a mechanism must
account for complex QoS characteristics, that on the one hand are expressed
by service providers and on the other hand are incorporated in the requester’s
preferences. The challenge is to provide a common conceptualization of quality attributes and enable their description, aggregation and enforcement from
an economic and technical perspective. Addressing this question, the auction
mechanism is extended in order to support complex QoS characteristics by means of
rule-based semantic concepts and a toolbox of adequate aggregation operations.
Furthermore, the mechanism is extended by a a compensation function which incorporates ex-post information about each services’ performance in order to impose penalties if necessary. The compensation function is designed to implement

1.2. RESEARCH OUTLINE

9

a truth-telling equilibrium with respect to all dimensions of service providers’
bids, i.e. truthful reporting of QoS attributes is a weakly dominant strategy for all
service providers.
It is well-known in mechanism design research that based on strong theoretic
results certain combinations of economic desiderata are impossible to achieve
at the same time [GL78, Wal80, HW90, MS83]. There exist interdependencies
between the properties of a mechanism and implemented social choice. Thus,
mechanism design goals often result in a trade-off between different properties.
Budget balance is an important property for a mechanism in order to be sustainable in the long-run as continuous external subsidization is neither reasonable nor profitable for e.g. a platform provider. Addressing the second part of
Research Question 3, an extended transfer function – the Interoperability Transfer
Function (ITF) – is developed which restores budget balance by sacrificing incentive
compatibility to a certain extent and at the same time incentivizes service providers
to increase their services’ degree of interoperability, i.e. to increase the capability of
their offered services to communicate and function with other services within the
service value network.
The challenge of how a mechanism’s properties can be evaluated by means of
analytical and numerical methodologies is stated in the following research question:

Research Question 4 ≺ E VALUATION ≻ . How can an auction mechanism be analytically and numerically evaluated regarding its economic
properties as well as cooperation and bundling strategies of service
providers?
Research Question 4 is firstly addressed by an analytical evaluation of the
mechanism’s properties which shows that the complex service auction implements a social choice that is allocative efficient and incentive compatible with respect
to all dimensions of service providers’ bids, i.e. truth-revelation of private QoS
attributes and valuations of offered services is an equilibrium in dominant strategies. Furthermore it is analytically shown that there exist ex-ante agreements
between service providers about a form of cooperation to reduce internal costs that
are mutually beneficial.
By means of simulation-based analysis, the extended budget-balanced transfer function is evaluated with respect to the robustness against bid manipulation,

10

CHAPTER 1. INTRODUCTION

i.e. to what degree it is beneficial for service providers to deviate from their true
valuation. Results show that even in settings with a low level of competition
strategic behavior of service providers is tremendously limited as a deviation from a
truth-telling strategy is not significantly beneficial even in small service value
networks. The incentive for service providers to increase their services’ degree
of interoperability is numerically evaluated by means of an agent-based simulation. Compared to an equal transfer function which distributes available surplus equally among allocated service providers, it is shown that the ITF extension
implements incentives to foster a higher overall degree of interoperability in settings
with a low level of competition. Thus, the ITF extension supports service value
networks in an early stage of development as a high degree of interoperability increases the multitude of feasible complex service instances that can be offered to
customers. An increase of variety and interoperability leverages network externalities [SV99, FK07, LM94, KS85] and attracts customers which in turn attracts
more service providers to participate in the complex service auction.
Broadening the strategic scope of service providers that participate in the complex service auction, it might be beneficial from a provider perspective – dependent on how they are situated within the service value network– to offer their
services as a bundle together with matching service providers. This question is
addressed by means of an agent-based simulation. It is evaluated if it is beneficial to offer bundled services which decreases flexibility but leverages synergy
effects and reduces costs or if it is beneficial to offer single highly specialized services that are more flexibly composable into various complex service instances. In
summary, there two main strategies analyzed: (i) Competing in quality through
differentiation and flexibility and (ii) competing in price through bundling synergies and cost reduction. Results show that in general service providers that own
services within the service value network which are highly competitive, i.e. they
are likely to be allocated, act best by following an unbundling strategy. In contrary, for service providers with less competitive service offers it is beneficial to
form bundled service offers while leveraging synergy effects. Nevertheless, this
strategic recommendation only holds in settings with a low level of competition.

1.3

Structure

The outline of this work is structured accordingly as depicted in Figure 1.1.
Chapter 2 introduces technologies, concepts and methods, which are fundamental for the work at hand. First, the concepts and key characteristics of dif-

1.3. STRUCTURE

11

Chapter 1
Introduction

Part I
Foundations

Part II
Design &
Implementation

Part III
Evaluation

Chapter 2
Preliminaries & Related Work

Chapter 3
Complex Service Auction (CSA)

Chapter 4
Applicability Extensions

Chapter 5

Chapter 6

Analytical Results

Numerical Results

Part IV

Chapter 7

Finale

Conclusion & Outlook

Figure 1.1
Structure of this work.

ferent kind of services are discussed and corresponding definitions are outlined.
Then service enabler technologies and paradigms such as service-oriented architectures, service value networks, and situational applications are introduced in
detail. Bridging the gap between a more technical to an economic perspective,
the idea of service markets is introduced and motivated in the context of complex services and service value networks. The discussion is followed by the description of the discipline of market engineering, which provides a structured
approach for designing, implementing, and evaluating market mechanisms in
different domains such as the service sector. The approach of mechanism design
underlying the work at hand is introduced as well as important impossibility and
possibility results. Summarizing the preliminaries, economic and applicability
requirements upon a suitable mechanism for trading complex services in service
value networks are discussed The requirement analysis is followed by a detailed
description of related approaches in that particular research area with respect
to stated requirements and identified shortcomings. Chapter 2 concludes with

12

CHAPTER 1. INTRODUCTION

a brief description of research methods, which are used to analyze the research
questions throughout this work.
Introducing the core model and mechanism implementation of the complex
service auction as well as corresponding applicability extensions, Chapters 3 and
4 embody the central part of this work. Based on the design part, Chapters 5 and
6 analyze properties of the complex service auction mechanism following analytical and numerical research methods. For the convenience of the reader, each
chapter entails detailed related work regarding the specific research question addressed additionally to the previously outlined approaches, which are closely
related to the work at hand.
Finally, Chapter 7 summarizes the key contributions of this work, outlines
complementary research and points out further challenges to be addressed in the
future.

1.4

Publications & Research Development

Excerpts of this thesis have been published in European and international academic conferences and as journal articles. This section provides a brief overview
regarding what parts have been presented, discussed and refined in the context
of which research community. This section furthermore illustrates how the work
at hand has been developed focusing on its steps of refinement and extension.
Laying the groundwork for this work at hand in Chapter 2, an analysis about
characteristics of traditional and e-services as well as corresponding service definitions have been published in the Proceedings of the 18th International World
Wide Web Conference (WWW 2009) [MB09]. The service decomposition model
and the conceptual framework for categorizing different service artifacts have
been presented at the Multikonferenz Wirtschaftsinformatik [BS08] and a revised
version at the Joint Conference of the INFORMS Section on Group Decision and
Negotiation, the EURO Working Group on Decision and Negotiation Support,
and the EURO Working Group on Decision Support Systems [BBS08].
Basic ideas and concepts about situational Web applications introduced in the
preliminaries have been published in the Proceedings of the 2nd Workshop on
Mashups, Enterprise Mashups and Lightweight Composition on the Web (MEM
2009, WWW 2009 pre-conference workshop) [BLH09]. A first position paper
about service value networks, their differentiation from related concepts, charac-

1.4. PUBLICATIONS & RESEARCH DEVELOPMENT

13

teristics, components, and an abstract model has been presented at the 11th IEEE
Conference on Commerce and Enterprise Computing (CEC 2009) [BKCvD09].
With respect to Chapter 3, first versions of the auction mechanism and the
idea of applying path auctions to composition problems have been published
in the 10th IEEE Joint Conference on E-Commerce Technology (CEC 2008) and
Enterprise Computing, E-Commerce and E-Services (EEE 2008) [BLNW08]. A
further refined version of the model including first simulation-based evaluations
have been presented at the 16th European Conference on Information Systems
(ECIS 2008) [BNWM08]. The next step of revision and extension of the complex
service auction has been published in the Proceedings of the 9th International
Conference on Business Informatics [CvD09].
The comprehensive model of the complex service auction as introduced in the
work at hand including a complete analytical analysis of the mechanism’s properties with respect to allocation efficiency and incentive compatibility as outlined in
Chapter 5 has been presented at the the 17th European Conference on Information
Systems (ECIS 2009) [BCM09] and published in the Journal of Business and Information Systems Engineering, Special Issue Internet of Services (forthcoming)
[BvDC+ 09].
A simulation-based evaluation of service providers’ bundling and unbundling strategies participating in the complex service auction as introduced
in Chapter 6 has been submitted to the Journal Electronic Commerce Research
and Applications, Special Issue on Emerging Economic, Strategic and Technical
Issues in Online Auctions and Electronic Market Mechanisms [BvDCW09].
As outlined in Chapter 7, complementary and future research with respect
to implementing mechanisms that – in contrary to traditional mechanism design
goals – provide innovative incentives to support service value networks in their
early stage of growth have been presented at the 15th Americas Conference on
Information Systems (AMCIS 2009) [CBSvD09].

Chapter 2
Preliminaries & Related Work
In contrast to a good, a service is not an entity that can exist independently of its
producer or consumer and therefore should not be treated as if it were some special kind
of good, namely an ’immaterial’ one.
[Hil99]
he goal of this chapter is to give a thorough introduction into technical and
economic foundations, which are essential for the remainder of this thesis.
The work at hand focuses on the design and evaluation of an auction mechanism
to coordinate value generation among distributed parties. The mechanism design
provides means for the feasible and efficient allocation and pricing of composite
services in service value networks.

T

This chapter firstly discusses the differentiation between tangible and intangible goods and the central concept of a service. Based on these results, a service
decomposition model is presented that provides a conceptualization scheme for different classes of services and highlights the concept of a complex service. Following
these definitions and classifications, the paradigm of a service-oriented architecture
is introduced, which embodies the key principles leading to enabler technologies for service-centric electronic networks. Technical foundations cover the concept of Web services, emerging technologies with a focus on lightweight protocols,
puristic architectural styles and slim message formats as well as quality of service
aspects and their legal manifestation in service level agreements. As coordination
plays a central role in distributed environments with self-interested parties such
as the Web, frameworks and specifications in the Web service context are introduced that provide means for realizing coordination mechanisms from a technical
perspective.

16

CHAPTER 2. PRELIMINARIES & RELATED WORK

As the work at hand focuses on not only distributed but also networked service environments, the emergence of service value networks as a novel form of
inter-organizational interaction and value generation is described and a model
for capturing essential characteristics is provided. Service value networks allow
for the realization of short-living complex services that fulfil customers’ needs
on a individual basis. Hence, such situational applications and service mashups are
briefly introduced.
Following this introduction of service concepts, definitions and technologies,
the need for auction mechanisms in these environments is discussed. Since this
work targets on providing a comprehensive design and evaluation of a suitable
service coordination mechanism from a technical and an economic perspective,
this chapter introduces the idea of algorithmic mechanism design and the interdisciplinary approach inherent in this emerging discipline. In the context of coordinating distributed and self-interested participants, central economic and computational desiderata, prominent mechanisms, and important impossibility results
are outlined.
Finally, the research methods underlying this work are briefly introduced.
This chapter introduces related work and state of the art that is broadly related
to the research questions at hand. Adjacent literature, a clear differentiation and
a detailed discussion is provided in the remainder of this thesis.

2.1

Service Concepts, Definitions, and Technologies

The whole concept of distributed (service-oriented) computing can be viewed as simply a
global network of cooperating business objects.
(Papazoglou 2000)
The goal of this section is to provide a thorough introduction to the service concept itself, conceptual classification models, related paradigms and technology,
and emerging service-centric environments.
Section 2.1.1 describes the differences between tangible and intangible goods
and the concept of a service by elaborating specific properties that allow for a
more or less strict differentiation. Based on this analysis, the service concept is
defined and its main characteristics are presented in detail. Concretizing the service concept by restricting its production and consumption channels to primarily

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

17

electronic networks, the concept of an e-service is described and its implications
on the general characteristics of a service are argued.
These foundations lay the groundwork for a service decomposition model as
illustrated in Section 2.1.2, which serves as a conceptual classification scheme for
different types of services with respect to their granularity and level of abstraction. Besides utility and elementary services, complex services – as a special type
of service – are introduced in detail as they embody a central concept for the work
at hand.
Section 2.1.3 is concerned with the paradigm of a service-oriented architecture
and its key principles which can be seen as the foundation for enabler-technology
such as Web services. Service-oriented architectures allow for the agile production and consumption of distributed services in electronic networks such as the
Web, that is, they enable value generation from a technical perspective. Value,
created by a service is mainly dominated by intangible elements that are experienced during its performance, which therefore highly depends on the service’s
quality. Hence, the main quality aspects that together constitute quality of service (QoS) are argued and how a legal foundation is constituted by service level
agreements. Distributed service activities that foster value generation and produce an overall quality that is provisioned to the consumer must be coordinated
by suitable mechanisms. By introducing a standardized framework that specifies
how coordination can be realized in the context of Web services, this challenge is
initially addressed from a technical perspective.
Designing suitable mechanisms to coordinate value generation through complex services requires a deep understanding of emerging forms of organization
of distributed service activities. Therefore, Section 2.1.4 presents the concept of a
service value network, its characteristics, the various roles involved and how they
are organized in order to jointly create value for potential service requesters. The
overall objectives underlying this value generation process are individually specified by the services requester and consequently change frequently. This leads
directly to the concept of situational applications and service mashups which is
elaborated from a technical and an economic perspective in the remainder of Section 2.1.4.

2.1.1 Tangibles, Intangibles, and Services
The differentiation between the terms good, intangible good, tangible good and
service is ambiguous and not exhaustive in the literature. Nevertheless a funda-

18

CHAPTER 2. PRELIMINARIES & RELATED WORK

mental understanding of the concepts at hand is inevitable to derive requirements
and implications in the context of service value networks, value generation and
their coordination.

2.1.1.1

Tangible and Intangible Goods

A good is an economic entity with a defined ownership. The ownership is defined by means of a legal right that allows the owner to use the good exclusively
and to prevent others from doing so. According to [Hil99] there are two main
characteristics of a good observable: (i) The existence of a good is independent of
the existence of its owner, meaning that a good’s identity is retained over time. (ii)
Ownership rights can be transferred from one economic entity to another, which
implies that goods are tradable. The owner of a good derives some economic
benefit from it (in contrary to a bad that decreases the utility of its owner). A
more rigorous differentiation between goods and services appears in the context
of production. The production process of goods involves inputs and outputs that
are entirely owned by the producer of the good. A good may be inventoried, sold
or traded, consumed or disposed after production as separated activities. The
fact that production and use are distinct activities is important from an economic
perspective as it allows for the transfer and exchange of goods even multiple
times.
Although most of the goods are material, economic entities exist that expose
all key characteristics of a good but are immaterial. According to [Hil99], “these
consist of intangible entities originally produced as outputs of persons, enterprises, engaged in creative or innovative activities of a literary, scientific, engineering, artistic or entertainment nature.” Although these information goods are
immaterial they are goods because ownership can be defined and transferred
from one economic unit to another. The main value for the consumer is derived
from the information itself. They are also intangible because they expose no physical dimensions (except from the medium the information is stored on, which is
not the economic entity at hand). The production process itself is mostly very
costly and time consuming, whereas the reproduction or copying of information
goods is cheap. The value of information goods generally increases through sharing and use [SV99, BBL99]1 .
1 Note

that this fact is not universally true. E.g. the value of private information about shares
of a company decreases through sharing.

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES
2.1.1.2

19

Services

Analogues to the fact that attributes, properties and characteristics of a service
are rather fuzzy, the concept of a service itself is hardly definable especially in
a consistent way across different application areas. Complementary to a short
definition, this section defines the service concept and differentiates it from adjacent concepts such as goods and products through the identification of its main
characteristics and their implications.
In general a service is some kind of activity or performance. The result of such
an activity is the change of condition of some person or good. This change of state
is based on an agreement of the economic unit owning the good and the one
providing the service [Hil77, Gad92].
Definition 2.1 [S ERVICE ]. A service is an activity which an economic unit A (service provider) performs for another economic unit B (service consumer) that results in a
change of state or condition of an economic unit C whereas The output of that activity
cannot circulate in the economy independently of economic unit C.2
Services expose a set of unique characteristics that have strong implications
from an economic perspective and allow a more or less consistent differentiation
from traditional goods or products. In order to analyze key characteristics of
services, it is important to differentiate the relevant phases of a service’s lifecycle
as depicted in Figure 2.1.
The overall lifecycle is determined and evaluated based on a global strategy,
i.e. the service strategy, that defines requirements and goals of the service portfolio. Based on initial requirements, the service design phase lays the groundwork
while dealing with conceptual decisions regarding a service’s design (e.g. is the
room service available all the time? Which architectural design to choose for
implementing a Web service?). Based on the initial design, the service itself is developed in the service production phase and all necessary resources for the service
provisioning are prepared (e.g. a Web service is implemented using the Ruby programming language, a hotel room is cleaned and the mini bar is refilled). According to the central service characteristic, the uno-actu principle, which is explained
in detail in the remainder of this section, service provision and service consumption
occur simultaneously, i.e. they coincide in time under the presence of a producer
and consumer. It is important to strictly differentiate between service produc2 This

definition is based on [Hil77, Gad00]

3 http://www.itil-officialsite.com/

20

CHAPTER 2. PRELIMINARIES & RELATED WORK
Service Strategy

Service
Design

E.g. architectural
decision:
RESTful ROA vs.
Big Web services
SOA)

Service
Production

E.g. Web service
development and
deployment

Service
Provision

Service
Consumption

E.g. flexible
binding and
execution

E.g. output
processing

Uno-Actu

Figure 2.1
Service lifecycle. Elements are partly derived from ITIL V33
tion and provision, as the latter is the central phase for the following analysis of
service key characteristics.
In literature it has been argued that intangibility is the main characteristic to
differentiate goods from services [Rat66, ZVB96]. Especially in the marketing
area, intangibility has been identified as the most difficult aspect of services to
deal with when it comes to the evaluation of service value creation as well as
quality control and assurance [Lev81, LW01]. Focusing on economic properties
and their implications for the coordination of value creation, intangibility is not
the only fundamental characteristic to differentiate goods from services. The following list of the key service characteristics serves as a basis to derive requirements for adequate market mechanisms to coordinate value generation through
services.
C 2.1 [U NO - ACTU ]. Service provision and consumption are not separable and coincide
in time.
In contrary to goods where the production, use and ownership can be separated from the economic entity itself, a service cannot be treated independently
from its producer or consumer. “Services involve relationships between producers
and consumers” [Hil99]. This implies that the process of production and consumption cannot be separated, meaning that there is no producer without a consumer and the other way around (e.g. a barber can only cut hair if the customer is
present at the same time, which implies that there is no hair cutting activity possi-

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

21

ble without the barber or the customer being present). This principle is also called
uno-actu and states that production coincides with consumption. Uno-actu is the
central and most important key characteristic of services. Hence, it is fundamental to
distinguish services from goods and it causally implicates most of the following
service characteristics.
C 2.2 [N OT STORABLE ]. Services cannot be inventoried or produced on stock.
The main value generated by the consumption of services comes from an action or performance. Service are ephemeral – transitory and perishable – which implies that they cannot be stored or produced on stock. It is not possible to produce
services in advance in order to meet fluctuating demand. It is of great importance to distinguish between the actual performance that leads to an immediate
change in state and its effect on reality. The activity itself on the one hand cannot be produced on stock as it is intangible and perishable. The person or good
that is affected by this activity on the other hand can mostly be preserved over
time [Gad00] (e.g. the actual deed of cutting hair cannot be produced on stock,
whereas the change of condition – the physical cut hair – can be inventoried and
exists over time). It has been argued by [Sta79] that the possibility to store and
transport an economic entity is the main distinguishing element of services. Considering energy as an economic entity, this argumentation does not hold or must
at least be relaxed, which questions its suitability for a strict differentiation.
C 2.3 [C O - CREATION ]. Services are generally co-created by their consumers.
According to Definition 2.1, services are deeds or actions that change the condition of another economic unit. This economic unit – often referred to as external
factor – is mostly brought in by the consumer. The consumer proactively influences the service activity and might therefore influence its result and quality. The
degree of customer participation and co-production in the context of different
service categories is analyzed in [BFHZ97]. Depending on the type of service (i)
customer presence might be required during service delivery, (ii) customer inputs might be required for the actual service creation or (iii) customer inputs are
completely mandatory. Co-production is argued to be the main characteristic to
differentiate services from goods [Fuc68]. However, recent production strategies
of traditional goods heavily integrate customers in the production process – often referred to as mass customization [PMS04] – which shows that co-production

22

CHAPTER 2. PRELIMINARIES & RELATED WORK

does not appear to be a suitable service characteristic in order to strictly distinguish services from goods.
C 2.4 [I NTANGIBLE VALUE CREATION ]. Value creation through services is characterized by intangible elements.
Some services include physical elements in the process of value creation
(i.e. spare parts during a repair process). However, the most value is created
in the form of intangible, immaterial elements. The consumer of a service experiences the performance or activity, which embodies the main portion of created
value [LW01]. Services create value when service consumers benefit from experiencing a service without a transfer of ownership (e.g. booking a hotel room).
Due to this fact, the assessment of quality and its assurance is a critical issue in
the context of services as an experience or an intangible result is hard to measure
and strongly depends on the economic unit to which it is provided. A continuous spectrum from tangible-dominant to intangible-dominant to differentiate
between goods and services is suggested in [Sho85].
C 2.5 [F UZZY INPUTS AND OUTPUTS ]. Service inputs and outputs are fuzzy and tend
to vary more widely.
Implied by the previous characteristic, it is hardly possible to control quality
aspects of a service in a way that outcomes are predictable and constant over time
[GW97]. Services are produced and consumed coincidentally and the value that
is created during this process varies widely due to the lack of control instruments
and various facets of service experience. This issue is even more intensified by
another phenomenon that is specific to services. The quality of a service might
depend on the ”quality” or effort of the service consumer (e.g. in teaching or
consulting) [Gri92]. Due to the fact that the quality or effort of a service consumer
is not under the control of the provider and tends to vary from individual to
individual, the final outcome of a service activity is fuzzy and varies more widely.

2.1.1.3

E-Services

With the rise of information and communication technology and the rapid
growth of the Web, the environment for service development, production, provision and consumption has changed completely. In this context the concept of
e-services emerged. The term e-service stands for a special form of “service that

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

23

is provided over electronic networks” [RK02]. The e-service paradigm [RK03] is
based on a broader view than the concepts of software services or IT services4 .
Definition 2.2 [E-S ERVICE ]. An e-service or electronic service is a service provided
over electronic networks.5
Based on the implications of these novel environments that foster the e-service
paradigm it is necessary to recall the service characteristics introduced in Section
2.1.1.2. As an e-service is a specific type of service, its characteristics are quite similar the characteristics of a general service. Nevertheless they have to be revised
and adapted according to the conditions of the changed surroundings.
C 2.1 (U NO - ACTU) In the context of e-services, the roles “service producer” and
“service consumer” are not strictly definable according to a traditional perspective. In most cases, the consumer of such a service is also an e-service or
another automated electronic entity (e.g. search agents, spiders and robots).
The role of the service producer is analogously hard to specify as e-services
are developed and ready for execution via electronic networks, meaning
that – under the assumption that there are no capacity constraints imposed
by e.g. the network’s bandwidth – these services can be performed anytime in a distributed manner to multiple consumers. Hence, dependent
of how the provision and the actual consumption is defined in the context
of e-services, this fact blurs the definition of the uno-actu principle which
states that service producer and service consumer are contemporaneously
involved in the performance of a service. Although the principle still holds
in the e-service context, its relevance and implications on service provision
and consumption have to be relaxed dependent of how provision and consumption are definable and separable.
C 2.2 (N OT STORABLE) E-services can be developed and stored to be ready for
execution. Although the physical storage of the program code that determines the behavior of the service is possible, the actual execution, which is
the value generating element of the service, can obviously not be performed
on stock. This also implies a fluctuating supply as capacity constraints in the
form of bandwidth or computing power limit the ability to satisfy peaks in
4 “A

Service provided to one or more Customers by an IT Service Provider. An IT Service is
based on the use of Information Technology and supports the Customer’s Business Processes. An
IT Service is made up from a combination of people, Processes and technology and should be
defined in a Service Level Agreement.” [RH07]
5 Based on the definition in [RK02]

24

CHAPTER 2. PRELIMINARIES & RELATED WORK
demand. Resource-focused capacity constraints can partly be overcome by
the use of computer grids or cloud computing environments that allow for
the flexible scaling of computing power and storage.

C 2.3 (C O - CREATION) In order to perform a service, the consumer mostly has to
provide additional information that is either transformed by the service or
used to scope and customize the service execution according to the needs of
the consumer. Although the service consumer does not bring in a physical
economic entity that is a central part of the service activity, the consumer
still influences and co-produces the final outcome of an e-service by providing necessary additional information or data. Thus, co-production is still
a central element of service provision and consumption in the context of
e-services.
C 2.4 (I NTANGIBLE VALUE CREATION) Value that is created through the execution of an e-service is idiosyncratic and highly depends on the preferences of
the service consumer. Although, the experience of a service performance in
an electronic environment also depends on expectations, needs and preferences of the service consumer, e-services partly allow for an objective measurement of service quality, which highly correlates with the value generated. The proportion of value-determining aspects of a service outcome that
can objectively be measured increases in the context of e-services, which
leads to an increase of uncertainty about the value generated through a service activity.
C 2.5 (F UZZY INPUTS AND OUTPUTS) A great advantage of e-services is the possibility to describe their main functionality and capabilities in a standardized manner, which simplifies their usage and management. Inputs and
outputs of e-services can be specified using standardized description languages that are common knowledge to service producers and service consumers. Thus, standardization and common sense about specifications reduce uncertainty about inputs and outputs in the context of e-services. Nevertheless, also in the context of electronic networks service, inputs and outputs highly depend on the state of the environment they ’live’ in. E.g. capacity constraints, network failures and unreliable transportation influence
the service outcome and its quality which increases uncertainty and unpredictability. Another factor that has an impact on the output generated by
the service is the consumer’s information that is either transformed or used
to scope the service execution. Fuzzyness of service inputs and outputs can

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

25

be reduced by means of standardized service description but is still an issue
in the context of e-services.
Summarizing described key characteristics, Table 2.1 shows an overview over
differentiation criteria of tangibles, intangibles, services, and e-services that have
been discussed in this section.

Services

E-Services

Intangibles

Criterion

Tangibles

Table 2.1: Differentiation criteria of tangibles, intangibles, services, and e-services. ( = fully satisfied, G
# = partly satisfied,
# = not satisfied, NA = not applicable)

#

#

#

#

#

NA

NA

Ownership rights definable and transferable
Immaterial

#

Costly initial production
Costly reproduction
Sharing increases value

#

#
G

Uno-actu

#

#

Not storable

#

#

#
G

Co-creation

G
#

#
G

#
G

Intangible value creation

#

Fuzzy inputs and outputs

NA

NA

#
G

#

2.1.2 Service Decomposition Model
This section gives a thorough classification of groups of services that share common characteristics from a technical and economic perspective as depicted in Figure 2.2. The Service Decomposition Model is based on the classification in [BS08] and
the extension in [BBS08]. The model distinguishes three different service layers
grouping Utility Services, Elementary Services and Complex Services.

2.1.2.1

Utility Services

Utility services reflect a vision where services can be accessed dynamically in
analogy to electricity and water: “Utility computing is the on-demand delivery

26

CHAPTER 2. PRELIMINARIES & RELATED WORK

Complex
Services

Enterprise Service
(Procurement Scenario)

IT Service
(Content Management
Sytem)

Economic Service
(Market Service)

Encapsulation

Elementary
Services

Intermediation Service
(Data Transformation)

Database Service
(Data Storage)

Information Service
(Information Retreval)

Virtualization

Utility
Services

Energy
(Electricity, Cooling)

Computation
(CPU)

Memory
(HDD, RAM)

Figure 2.2
Service decomposition model [BBS08].
of infrastructure, applications, and business processes in a security-rich, shared,
scalable, and standards-based computer environment over the Internet for a fee.
Customers will tap into IT resources – and pay for them – as easily as they now
get their electricity or water.” [Rap04]. Utilities are characterized by necessity,
reliability, ease of use, fluctuating utilization patterns, and economies of scale. In
[Rap04], base pricing in utility computing on metering usage (also coined “paywhat-you-use” or “pay-as-you-go”) is suggested, as is the case with classic utilities such as water, telephone and Internet access. With the fast rise of energy
prices, the meaning of utility services is even extended back to the roots where the
name originally came from: Basic computing services in hosting centers need to
be managed explicitly taking into account energy consumption as a relevant optimization criterion [CAT+ 01]. “Heterogeneous server clusters can be made more
efficient by conserving power and energy while exploiting information from the
service level, such as request priorities established by service level agreements”
[BR04]. Even temperature aware computing solutions for data centers are proposed [MSS+ 08].

2.1.2.2

Elementary Services

Elementary services virtualize the utility services layer and encapsulate underlying functionality. They provide rather basic functionality such as data format
converting services, storage services, or pure information services that retrieve in-

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

27

formation from designated sources. Although the type and behavior of these services are mostly standardized, they have multiple attributes with varying characteristics. For instance, storage services may differ according to their capacity,
access time and data throughput. These varying characteristics of the same type
of service, as well as the service itself can be described by means of standardized
description languages. The input and output semantics of these so-called elementary services are well-accepted and interpretable. Examples might be database
services and data format transformation services. Services in this layer are required for several different higher-level applications and, as a consequence, are
utilized by a multitude of different users. Similar to utility services, the provided
quality of service for the same type of service may vary. For instance, a set of
data format transformation services may vary from their offered response time;
however, it is assumed that these characteristics can also be described in a standardized form.

2.1.2.3

Complex Services

While elementary services provide simple functions such as credit checking and
authorization, inventory status checking, or weather reporting, complex services
may appropriately unify disparate business functionality to provide a whole
range of automated processes such as insurance brokering, travel planning, insurance liability services or package tracking [PD04]. A complex service is composed of multiple service components (which are either elementary or complex
themselves), often requiring an interaction or conversation between the user and
services, so that the user can make decisions [MSZ01]. According to [Pap08], a
complex service can be defined as follows:
Definition 2.3 [C OMPLEX S ERVICE ]. Complex (or composite) services typically involve the assembly and invocation of many pre-existing services possibly found in diverse
enterprises to complete a multi-step business interaction.
Complex services combine the functionality and capabilities of modularized
service components (which themselves can be utility, elementary or complex services) by sequential composition in order to generate added value. To illustrate
the idea of complex services this section provides exemplary business cases from
the enterprise sector which are based on current market information.

28

CHAPTER 2. PRELIMINARIES & RELATED WORK

Example 2.1 [C OMPLEX S ERVICE : PAYMENT P ROCESSING ]. Consider a manager
of a mid-size company that distributes flowers over the Internet. As payment processing is
not a core competency of the company, the board decides on the integration of third-party
services into existing business processes in order to decrease the costs of operation and
maintenance. Figure 2.3 shows the overall business scenario and in detail the payment
processing complex service that is intended to be replaced by a third-party service from
external providers.

Order
Processing

Payment
Processing

Logistics

Data
Verification
Service

Transaction
Processing
Service

Database
Service

Storage
Service

Figure 2.3
Business scenario integrating a payment processing service.
Focusing on the payment processing complex service and necessary components, the
diagram in Figure 5.1 sketches an excerpt of the service components of an exemplary
complex service that provides payment processing functionality.
The PaymentProcessingService facilitates service components from Strike Iron6 ,
Duo Share7 and CDYNE8 to verify the customer’s address and credit card information.
Customer data is stored and managed using a StorageService and a DataBaseService
from third-parties. Sample services from decentralized storage providers are Amazon
S39 , Digital Bucket10 and Box.net11 . Services for organizing and managing customer
data are Amazon Simple DB12 and Long Jump DaaS13 . The actual execution of the fi6 http://strikeiron.com/
7 http://duoshare.com/
8 http://cdyne.com/
9 http://aws.amazon.com/s3/
10 http://digitalbucket.net/
11 http://box.net/
12 http://aws.amazon.com/simpledb/
13 http://longjump.com/daas/

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

29

PaymentProcessingService

DataVerificationService

AddressVer

CreditCardVer

DatabaseService

StorageService

TransactionProcessingService

LongJumpDaaS

AmazonS3

JETTISTransactionProcessing

AmazonSimpleDB

DigitalBucket

NetBillingCreditCardProcessing

StrikeIronGlobalAddressLocator

Box.net

DuoShareAddressQualityIntegrator

CDYNEPostalAddressVerification

Figure 2.4
Payment processing service (static view).

nancial transaction through the TransactionProcessingService is provided by JETTIS
Transaction Processing14 and Net Billing Credit Card Processing15 .
The process behavior of the payment processing complex service is depicted in Figure
2.5. Customer data is validated in the first step. After validation the actual transaction
takes place and the customer’s credit card account is charged by a transaction processing
service. The change in state must be updated in the internal database of the company. A
database service updates corresponding customer data that is stored using a decentralized
storage service.
For each step of the complex service there is a potential pool of suitable candidates
to fulfill required business transaction. The result of each transaction is passed to the
successor service. In order to successfully instantiate the complex service the overall
transaction requires a service candidate from each pool.
14 http://jettis.com/
15 http://netbilling.com/

30

CHAPTER 2. PRELIMINARIES & RELATED WORK

Data
Verification
Service

Transaction
Processing
Service

Database
Service

Strike
Iron

Storage
Service

Amazon

JETTIS

Long
Jump

Duo
Share

Digital
Bucket

Net
Billing

Amazon

CDYNE

Box.net

Figure 2.5
Payment processing service (dynamic view).
Example 2.1 shows that core service competencies can be leveraged by procuring complex services from third party providers to close competency gaps in business processes. The granularity of complex services ranges from services that are
parts of a business process to services that cover whole business scenarios as illustrated in the following example.
Example 2.2. To further illustrate the idea of a complex service a business scenario which
is actually delivered to customers as part of SAP’s BusinessByDesign16 is introduced exemplarily. The scenario consists of modular service components that can be provided
by decentralized service providers. The integration scenario “Service Request and Order Management” (cp. Figure 2.6) describes operational processes in a customer service
based on service requests, service orders and service confirmations. From an end-to-end
perspective the scenario includes the integration into related applications such as logistics
planning and execution, invoicing and payment, as well as financial accounting.
The complex service is formed by decentralized service providers that contribute to
the achievement of an overall goal. In the presented scenario this goal is the flawless ex16 http://www.sap.com/solutions/sme/businessbydesign/

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

SCM

CRM

Service
Request
Processing

Service
Order
Processing

Service
Confirmation
Processing

Customer
Requirement
Processing

Logistics
Execution
Control

31

FIN

Supply and
Demand
Matching

Customer
Invoice
Processing

Due Item
Processing

Payment
Processing

Figure 2.6
Business scenario “Service Request and Order Management”
(SROM).
ecution of a business scenario in order to provide defined functionality to the customer.
Many service providers offer differentiated and specialized services covering various types
of functionality within the complex service. They provide service components regarding
customer relationship management (CRM), supply chain management (SCM) and finance (FIN). In this scenario the functionality of each component can be modularized
and therefore performed by different software-as-a-service (SaaS) providers as depicted in
Table 2.2.
Table 2.2: SaaS providers for CRM, SCM and FIN components of
the business scenario SROM.
CRM

SCM

FIN

Salesforce

GXS

Cashview

http://salesforce.com/

http://gxs.com/

http://cashview.com/

Rightnow

7Hills

Opsource

http://rightnow.com/

http://7hillsbiz.com/

http://opsource.net/

Oracle

Intacct

http://oracle.com/crmondemand/

http://intacct.com/

SAP
http://www.sap.com/solutions/sme/businessbydesign/

The rapid growth of the number of on-demand service providers shows the high degree of innovation and market penetration as a result of service modularization. Service
providers offer specialized services and concentrate on their core competencies. Each service provider is responsible for a certain part of the overall functionality, which consequently spreads the risk of an erroneous business process over all contributing service
providers. Furthermore, they partly grant access to their own resources thus supporting
the realization of the overall business scenario.

32

CHAPTER 2. PRELIMINARIES & RELATED WORK

2.1.3 Service-Oriented Architectures
This section introduces fundamentals and basic concepts of service-oriented architectures with a focus on technologies and definitions that serve as a basis for
the remainder of this thesis. In Section 2.1.3.1, service-oriented architecture as
a paradigm for organizing distributed services that are under the control of different domains is introduced. The section provides a definition of the serviceoriented architecture concept and introduces its key principles. The concept of
Web services as the most prominent example of a technology that leverages the
strength of service-oriented architectures is presented in Section 2.1.3.2. The section guides through the Web service technology stack and state-of-the-art specifications and standards. It is well-known that the main value generated by a service activity is determined by its quality characteristics and their manifestation
at run-time. Hence, Section 2.1.3.3 introduces the concept of quality of service
(QoS), relevant factors in the context of Web services and how QoS guarantees
can be formulated in contracts, i.e. service level agreements. Contracts defining
QoS aspects provide the legal basis for the market-based trade of services as a special form of coordination. Thus, technologies and concepts for the coordination
of Web services are introduced in Section 2.1.3.4 that provide means for organizing dependencies among distributed service activities that have to be governed
to achieve an overall outcome.

2.1.3.1

Basic Concepts

Service-oriented architectures (SOAs) have gained a lot of momentum over the
last years. SOA is a paradigm to organize distributed capabilities possibly under
the control of different domains. The paradigm itself and its concrete implementations are fundamental for the development, production, innovation and provision of services via electronic channels. Technology that is based on the SOA
principle can be seen as the enabler technology for service-oriented computing.
Definitions of service-oriented architectures and related concepts are based on
the OASIS Reference Model for Service Oriented Architectures [MLM+ 06].
The main goal of service-oriented architectures is the composition of complex applications out of loosely-coupled service components that provide specific well-defined functionality. Service components are designed to live independently of the application they are part of and are therefore reusable and recomposable in different application contexts [Ley03]. In order to illustrate the idea
of the flexible composition of loosely-coupled service components, the concept of

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

33

a service and its interaction with central roles in the context of service-oriented
architectures have to be elaborated in detail.
Relevant services in the context of service-oriented architectures are a subset of e-services as defined in Section 2.1.1.3. These types of electronic services
are called software services. Software services are self-describing software components that provide certain capabilities through a programmatic interface via
electronic networks such as the Internet. A service interface publishes the service’s
signature describing input and output parameters as well as message types. The
objectives of a service are defined through its capabilities, which are acts or performances that solve problems of an economic unit. They state the conceptual purpose and expected result of the service by using terms or concepts defined in an
application-specific taxonomy [PG03]. Narrowing down Definition 2.1, capabilities are provided through a software service by a service provider and consumed
by a service requester in order to fulfill certain needs. Software services expose
three major properties that are essential for the SOA paradigm:
• The programmatic interface of the service is platform-independent.
• The service can be dynamically located and invoked.
• The service maintains its own state (self-contained).
By means of a well-defined platform independent interface, the service can
be consumed from anywhere, on any operating system and in any programming
language. The service can be discovered by means of a look-up mechanism facilitating a service registry. In any state of its lifestyle the service manages its own
state independently. Compromising this information the definition of software
services is the following:
Definition 2.4 [S OFTWARE S ERVICE ]. A software service is a self-describing, selfcontained mechanism that enables the access to certain capabilities of an encapsulated
software component via an electronic network by means of a well-defined platformindependent programmatic interface. A software service is an open component that can
be dynamically located, bound and invoked.
The definition at hand is more restrictive then Definition 2.2 because it requires the existence of a well-defined platform-independent programmatic interface17 . An example of a software service would be a credit card verification
17 For

the reader’s convenience, the terms software service and service are from now on used
interchangeably.

34

CHAPTER 2. PRELIMINARIES & RELATED WORK

service accessible over the Internet that verifies credit cards at a central authority
based on the card number provided through the service’s interface. In contrary
a Web blog might not be considered to be a software service according to Definition 2.4 as it does not expose a well-defined programmatic interface in the narrow
sense.
In the context of service-oriented architectures there are three primary operations to manage the interaction between the provider and requester roles as
depicted in Figure 2.7. These are the publication of the service descriptions at a
service registry by the service provider, finding of the service descriptions, binding
and execution of the services based on their description by the service requester
[Pap08].

Registry
find

publish

bind

Requester

Provider

execute

Figure 2.7
Roles and primary operations in service-oriented architectures.
Publishing a service at a service registry mainly consists of two steps. The
first step is to describe the service at hand, that is, a description of its interface
and usage conditions. The second step is the actual registration of the service in
order to facilitate discovery and reusability by service requesters. The finding of
a service involves two steps as well: The first step is to create a description in the
form of a query that defines criteria and search terms concretizing the service that
is needed by the service requester. The second step, is the selection of the set of
services retrieved from the discovery agency. Criteria defined in the query consist of the type of service that is needed, quality aspects and other technical as
well as non-functional service characteristics. The query is executed against the
data set stored in the service registry and a subset of services that meet the criterions in the search query are retrieved. In the second step the service requester
has to chose from the set of discovered services either statically at design-time

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

35

or automatically bound at run-time. Binding and invocation are the most important operations in service-oriented architectures. Once a service is chosen either
statically or dynamically, the service requester and the service provider agree to
a well-defined and unambiguous contract that describes the service at hand and
corresponding service level agreements. The invocation can either be performed
directly by the service requester using the technical service description from the
registry or via a mediation through the registry.
Having defined services, related concepts, roles and primary operations in the
context of service-oriented architectures, the paradigm itself, its main goals and
its key principles can be defined
Definition 2.5 [S ERVICE - ORIENTED A RCHITECTURE ]. A service-oriented architecture is an architectural design paradigm to structure, utilize and compose distributed
interoperable software services that are under the control of decentralized ownership domains in order to realize distributed applications.
In order to achieve defined purposes the SOA paradigm relies on the following key principles.
Loose-coupling The term coupling refers to the degree of dependency between
two systems. Therefore, loosely-coupled services can interact more freely
as they do not need to know the location, behavior, implementation or
any other details of communication partners. Systems that are designed
in a loosely-coupled manner are mostly based on asynchronous or eventdriven interaction schemes instead of synchronous communication [Pap08].
A loosely-coupled design allows for the flexible restructuring of processes
and application logic without having to touch the internal structure of the
services involved as they live independently within a service-oriented architecture [Bur04].
Interoperability A main benefit of service-oriented architectures is the heterogeneity of services that can be integrated in a distributed system. This diversity and continuous evolution of services during their lifecycle implies
a high complexity to enable a seamless communication between services
without manual adaption, i.e interoperability. The high degree of standardized formalisms and protocols in service-oriented architectures are key concepts to achieve the desired interoperability of distributed services.
Reusability As services in a service-oriented architecture are self-contained,
loosely-coupled and not bound to a concrete system, they can be reused

36

CHAPTER 2. PRELIMINARIES & RELATED WORK
in different application contexts. Due to reusability, the number of redundant components in a service-oriented architecture is generally much lower
compared to traditional systems. This results in a lower effort for change
management and maintenance in service-oriented architectures.

Discoverability In order to reuse services in a service-oriented architecture, a potential consumer or developer must be able to find the service that matches
the specified requirements. Discoverability is mostly realized by a service
repository that entails services including their description to enable their
search and usage. The process of service discovery can either be performed
manually by consumers or automatically by the system.
The key principles of service-oriented architectures are pursued and enabled
by the architectural design through the encapsulation of infrastructure, application
logic, services and business processes in a transparent manner. Figure 2.8 schematically illustrates the architectural layers of a SOA as well as their interactions.

Business
Processes

Service Bus

Services

Application
Logic

Virtualization

Infrastructure

Figure 2.8
SOA layers.

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

37

The infrastructure layer comprises physical resources providing computing
power, storage, memory and bandwidth. Encapsulation and flexible resource
provisioning is achieved by the adoption of virtualization technologies that allow for the dynamic instantiation and migration of virtual resource environments
independent from their physical hosting location [BDF+ 03]. Virtualization is an
important step towards a service enablement of physical resources, which fosters
a service-oriented management of hardware units.
Above the virtualized infrastructure is the application logic layer, which entails
applications and application systems that provide the actual functionality in the
form of software components. These systems are a mixture of up-to-date application systems and old legacy systems. Applications in the application logic layer
are enhanced by service definitions to enable encapsulation and abstraction in
order to be manageable in a service-oriented context.
The application logic layer is abstracted by services in the service layer. They
encapsulate functionality in a self-describing, self-contained and loosely-coupled
manner and provide access through well-defined interfaces. The service bus is the
main component of a service-oriented architecture. It functions as the connecting
element between the set of services providing loosely-coupled functionality and
business processes reflecting organizational criterions and real-world business
procedures. The service bus enables the retrieval, provision and binding of services [Ley03] while supporting standards to facilitate distributed communication
and message exchange between services.

2.1.3.2

Web Services

Over the last decade the Web has evolved from a content- or document-oriented
environment to a service-centric environment. This is due to the rise of the concept of Web services. The term Web service in general does not per se imply a
concrete form of realization. Web services are a way to expose functionality in a
standardized manner that is accessible over the Web in order to realize complex
distributed applications. The use of standard Web technology reduces heterogeneity and enables the reuse and integration of distributed functionality independent of platforms and programming models. In contrary to traditional intercompany middleware that is centrally organized and controlled by a single company, the Web service paradigm allows for the integration of globally distributed
services across organizational boundaries.

38

CHAPTER 2. PRELIMINARIES & RELATED WORK

A huge body of work has been done defining Web services. The most prominent definitions range from a very generic perspective to a strict and languageoriented view. Nevertheless, only focusing on the aspect that Web services are
applications that are accessible over the Web to other applications [ABC+ 02] is
certainly not practical. In contrary, the notion of the World Wide Web consortium (W3C) [AGB+ 04] is much stricter as it limits Web services to those services
that expose interfaces that are described using the eXtensible Markup Language
(XML) [BPSM+ 06]. The W3C defines a Web service as “[...] a software system
identified by a URI [BLFM98], whose public interfaces and bindings are defined
and described using XML. Its definition can be discovered by other software systems. These systems may then interact with the Web service in a manner prescribed by its definition, using XML based messages conveyed by Internet protocols.” This definition excludes Web services that exchange messages in a more
lightweight manner facilitating formatting standards that in contrary to XML reduce payload. In order to include these types of Web services the definition by
W3C has to be relaxed regarding language limitations.
Definition 2.6 [W EB S ERVICE ]. A Web service is a software service identified by a
URI [BLFM98] that exposes a public interfaces, based on Internet standards. A Web
service can be discovered by other software systems. These systems may then interact
with the Web service in a manner prescribed by its definition, using Internet standard
based messages conveyed by Internet protocols.
Conceptually Web services can be divided in two main categories depending
on the architectural style used for their realization, i.e. RESTful Web services18 and
Big Web services. [PZL08].
Recently, RESTful Web services have increased attention not only because of
their usage in the context of Web 2.019 , service mashups and situational applications, but also because of the presumed simplicity and their lightweight character.
RESTful Web services are based on an architectural style that is used for realizing distributed hypermedia information systems (e.g. the Web). Messages are
transported via the HTTP protocol without the need for an envelope on-top such
as SOAP that generates extra XML payload. RESTful Web services expose unique
document processing interfaces. The signature consists of the scoping information
specified by a URI (e.g. “/reports/open-bugs/”) and method information specified in the HTTP header (e.g. GET, HEAD, PUT, DELETE). Due to the strict and
18 The
19 cp.

term Representational State Transfer (REST) was firstly introduced in [Fie00]
http://programmableweb.org/apis/

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

39

exclusive use of standardized HTTP methods valuable properties are retained,
i.e. safety and idempotence. Safety refers to the property that – assuming a correct
implementation of a RESTful Web service– the execution of HTTP methods GET
and HEAD does not change the state of the corresponding service. Idempotence
is a property of an operation that states that the result of an operation is independent of the number of executions20 . It is important that HTTP methods such
as PUT and DELETE are idempotent operations due to the unreliable nature of
the Web and the uncertainty of a successful method execution. Therefore, it is
possible to invoke the same method multiple times without having to care about
the implications of the repeated calls. Furthermore, RESTful Web services are
addressable, connected and stateless meaning that they can be uniquely identified,
they mostly point to other services that make sense in a certain context, and any
information that is necessary to understand a message is enclosed in the HTTP
message.
Up to now the lightweight nature of RESTful Web services and the lack of
expensive service descriptions have been regarded as feature of the approach especially in the context of service mashups and situational applications. However,
as applications become more complex and the number of services grows, the lack
of a service description becomes increasingly problematic (see also discussion in
[PZL08, Pau08]). Therefore, first approaches for annotating RESTful Web services
have been proposed. Similar to the approach used in SAWSDL [FL07] for WSDLbased services, SA-REST [SGL07, LGS07] can be used to attach model reference
annotations to HTML using RDFa [AB08]. It can thus be used to annotate RESTful
Web services.
Recently, many service providers claim to offer RESTful Web services but
mostly violate important properties that are outlined in this section [RR07].
Prominent examples of service providers that offer correctly implemented RESTful Web services are Amazon and Yahoo!. Amazon offers storage capacity
through its Simple Storage Service (S3)21 that is fully accessible and manageable in the manner of REST. Most of Yahoo!’s Web services22 are also available
as RESTful Web services.23
To pursue SOA principles such as interoperability and platform independence, Web service technology is based on standardized Internet protocols and
20 e.g.

the function f ( x ) = 1 · x is idempotent as f ( f ( x )) = f ( x ) and in general f ◦ · · · ◦ f = f

21 http://aws.amazon.com/s3/

22 http://developer.yahoo.com/
23 Note

[RR07].

that also most static Web sites are accessible and manageable as RESTful Web services

40

CHAPTER 2. PRELIMINARIES & RELATED WORK

description languages to allow for the interoperable automation of distributed
applications without the need for human intervention. Thus, Web services are
not built in a monolithic manner but rather founded on a stack of complementary
standards encapsulating several functional layers as illustrated in Figure 2.9.
Orchestration &
Choreography

WS-BPEL, WS-CDL

Big WS Stack

WS-Coordination

WS-Context

Discovery

UDDI

WSDL

WS-Policy

RESTful
WS Stack

Description

Messaging

SOAP

XML, XML Schema

Coordination &
Context

JSON

HTTP

Forma!ing

Transport

Figure 2.9
Web service technology stack.
Due to this design principle, new standards in the context of Web services
emerge quickly as they are developed on-top of existing functionality24 .
Transport
Web services facilitate basic Internet infrastructure technology such as the
Hypertext Transfer Protocol (HTTP) [FGM+ 99], the Simple Mail Transfer Protocol (SMTP) or the File Transfer Protocol (FTP). The HTTP protocol enables
transportation, ensures almost universal reach and support and is the most
prominent transport protocol used by Web servers and browsers. It allows for
the stateless interoperability of distributed, collaborative information systems. In
order to enable the unique addressing for transportation, resources on the Web
are identified using a Unique Resource Identifier (URI) [BLFM98].
Formatting

24 The interested reader is referred to http://www.innoq.com/soa/ws-standards/poster/

for a comprehensive overview of state-of-the-art Web service standards.

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

41

Messages that are exchanged via the transport layer are structured based on
formatting standards. The most prominent example that is widely used is the
eXtensible Markup Language (XML) [BPSM+ 06] but there are also lightweight
formats mainly pushed through Web 2.0 technology such as the JavaScript Object
Notation (JSON) [Cro06].
Messaging
Message exchange in distributed environments such as the Web have to be
organized using standardized specifications. Specifications for the exchange of
messages are developed on top of the transport layer and protocols such as HTTP,
SMTP or FTP and function as an envelope that defines how messages should
be exchanged between communication partners. A well-established framework
for Web service information exchange is the Simple Object Access Protocol
(SOAP) [BEK+ 00]. SOAP is a further development of XML-RPC [Win99]. It
is a network protocol that enables the XML-based message exchange between
distributed software systems in the manner of a Remote Procedure Call (RPC)
architectural style. It specifies how messages should be structured, formatted
and interpreted independent of semantics and application-specific information.
SOAP messaging can be enhanced by complementary Web service standards
such as WS-Security [NKMHB06] to allow for integrity and confidentiality of
information exchange procedures.
Description
The publish-find-bind-execute paradigm as illustrated in Figure 2.7 allows service providers to publish services at a central registry, that can then be discovered, bound and executed by service requesters. In order to enable such roles,
operations and interactions in a service-oriented architecture, Web services need
to be described in a consistent manner. Thus, only if a service requester is able to
gather all necessary information on a service’s interface and the type and structure of the messages being exchanged, services can be assembled and composed
into value-added complex services that expose business functionality. Service
description reduces the need for a common understanding and custom programming and is a key driver of loosely-coupling in service-oriented architectures.
It is a machine-understandable description of a service’s structure, operational
characteristics and non-functional properties [Pap08].

42

CHAPTER 2. PRELIMINARIES & RELATED WORK

The Web service Description Language (WSDL) [CCMW01] is widely used
especially for the description of SOAP-enabled Web services. Generally, WSDL
describes what a service does, that is, the operations the service provides, where
it is located, and how to invoke it. WSDL is based on XML consisting of an abstract part and a concrete part. A service’s interface consisting of operations and
corresponding data types of input and output messages are specified in the abstract part by means of a port type. The concrete part binds the abstract port type
to a message encoding protocol and adds a concrete end point address to each port
type.
Although the Web is mainly based on HTTP as the transport protocol, WSDL
and SOAP hardly use the features of HTTP at all (e.g. SOAP only uses HTTP
response codes “200” and “500”). Nevertheless, it is also possible to leverage
the power of HTTP by facilitating all features originally provided by HTTP 1.1
in order to describe Web services. Exemplary, the Web Application Description
Language (WADL) [Had06] describes resources or services that respond to
HTTP’s uniform interface by grouping their operations into a single end point.
Discovery
The full potential of reusable loosely-coupled Web services can only be utilized
if there exist mechanisms that enable service providers to publish information on
the capabilities of their service offers and how to access and use them. Service
requesters should be able to discover adequate services that match their requirements and the necessary information to bind and invoke them. Service discovery
is the process of querying a service registry and retrieving published Web service
descriptions that specify the Web service’s properties, its capabilities and how to
properly interact with it. The discovery process can be differentiated in two basic
types, static and dynamic discovery [GSB+ 02]. Static discovery queries a registry
and receives necessary information at design-time while dynamic discovery proceeds these steps during run-time. After having retrieved a set of Web services
that match the query criteria, the service requester has to select a service to be
invoked.
The Universal Description, Discovery, and Integration (UDDI) [CHvRR04] is
a framework representing a central registry to publish and discover Web services
in a global and open manner. Information provided by a UDDI registry is threefold. White pages provide contact information on companies that publish their
services in a UDDI registry. Yellow pages provide the classification of information

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

43

based on standardized industry taxonomies. Green pages accommodate service
requesters with necessary technical information regarding exposed Web services.
Coordination & Context
In distributed environments with decentralized service providers, the coordination of transactions is a fundamental concept in order to govern interactions
of participants to achieve a desired outcome. A detailed introduction to the
WS-Coordination specification [NRFJ07] is provided in Section 2.1.3.4.
Orchestration & Choreography
Generating value from a business perspective is achieved by loosely-coupled Web
services that are composed into complex applications as the main objective of
the SOA paradigm. There are essentially two types of service composition as
depicted in Figure 2.10 that have to be differentiated.

Orchestration X
Service
X1

Service
X2

Service
X3

Choreography XY

Service
Y1

Service
Y2

Service
Y3

Orchestration Y

Figure 2.10
Service orchestration versus service choreography.

44

CHAPTER 2. PRELIMINARIES & RELATED WORK

Service orchestration completely describes the composition procedure of internal or external services controlled by a central element. Each service that
is part of an orchestration has a limited scope that restricts its decision radius. Activities that run internally within a service component are transparent and hidden to other services. A specification of a service orchestration
describes service components, conditional dependencies and alternatives
within a composition.
Service choreography is the description of a protocol that defines rules for the
interaction between service components and their function within the composition. There is no central element to control and assure a correct behavior
of each service component and the composition itself. A service choreography focuses on the exchange of messages between services components and
the definition of necessary protocols.
In short the difference between service orchestration and choreography can
be narrowed down as follows:
Orchestration defines procedure, choreography defines protocol.
From a business perspective the goal of a service-oriented architecture is to
provide the architectural design that enables a flexible customization of business
processes in order to align IT and business. As business processes are volatile
and change frequently, service-oriented architectures allow for an ad-hoc adaption of business processes according to situational needs and changing market
requirements. The final process flow is instantiated at run-time, which enables
just-in-time reflection of real-world business processes in a way that IT aligns
with business and not vice versa.
Web service standards such as SOAP, WSDL and UDDI provide means for the
realization of relatively simple Web services that fulfill limited tasks by providing adequate functionality. Extending the vision of a loosely-coupled serviceoriented architecture that overcomes physical boundaries and enables an interand intra-organizational integration of business functionality requires standardized formalisms to describe Web service orchestration into business processes and
their choreography in a seamless manner. A Web service business process describes
how operations are composed out of a set of potential Web services, how they
interact, share information and what partners are involved in order to create the
required business value.

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

45

The Web Service Business Process Execution Language (WS-BPEL) [AAA+ 07]
provides a standardized description language for specifying business processes
composed of operations that are exposed by WSDL-based Web services.
Hence, WS-BPEL supports service composition models, recursive composition,
separation of composability of concerns, stateful conversation and lifecycle
management, and recoverability properties [WCL+ 05]. WS-BPEL mainly contains five sections, i.e. the message flow, the control flow, the data flow, the process
orchestration, and the fault and exception handling section as illustrated in Listing
2.1.

1

<process name="paymentProcessing" ...>

3

<partnerLinks> ... </partnerLinks>

5

<variables> ... </variables>

7

<correlationSets> ... </correlationSets>

9

<!- Activities -->

11

<faultHandlers> ... </faultHandlers>

13

<compensationHandlers> ... </compensationHandlers>

15

<eventHandlers> ... </eventHandlers>

17

</process>

Listing 2.1: WS-BPEL Structure

The selection of services for composition and for the definition of relationships
among services revolves around the notion of partner links. WS-BPEL maintains
the state of the process and control data which is stored in variables analogous
to variables in programming languages which are specified by names and types.
Partner links describe a pair of roles which exchange messages and port types
that the service playing these roles has to implement. Enabling the mapping of
messages to composition instances, correlation sets can be defined that describe
how to correlate messages with concrete instances.
The component model of WS-BPEL consists of basic and structured activities.
Structured activities define the actual orchestration whereas basic activities specify the components itself and correspond to the invocation of a WSDL operation.
As basic activities, WS-BPEL provides invoke activities, that invoke operations,
as well as receive and reply activities which correspond to the receipt of a client’s
message and to the reply in response to an operation invoked. Structured activities however are capable of defining more sophisticated process logic by combin-

46

CHAPTER 2. PRELIMINARIES & RELATED WORK

ing other activities (basic and structured). Constructs of structured activities are
sequences, switches, picks, whiles and flows.
Providing means for exception handling, fault handlers define how certain
exceptions should be managed. fault handlers specify a catch element which
defines the fault it manages and the corresponding activity that is triggered in
case an exception occurs. Combining exception handling and transactional techniques, compensation handlers define the logic required to undo the execution of
activities as a compensation. In contrary to the try-catch-approach, event handlers continuously monitor certain events and define activities to be triggered in
case that particular event occurs.

2.1.3.3

Quality of Service (QoS)

The value generated by a service is mainly embodied through intangible elements
exposed at execution (cp. service characteristic C 2.4). Therefore, a service consumer expects a service to function reliably and to deliver a consistent outcome
at a variety of levels, i.e. quality of service (QoS). To provision, control and assure QoS it requires not only for focus on functional properties of a service but
also on non-functional aspects. The context of a service also influences its quality, which is experienced by the consumer, e.g. the partner network that comes
with a service, its reputation in certain communities or advertisement campaigns
promoting the service. From an economic perspective, QoS is the most important
characteristic that differentiates service offerings and leverages market advantage, as price competition is tough due to low variable costs of service provisioning. Thus, QoS is the key criterion to keep the business side competitive as it has
serious implications on the provider and consumer side [Pap08]. The provision
of services with a defined QoS over electronic networks such as the Web is challenging due to issues like infrastructure problems, unpredictable reliability, low
performance of Web protocols and many more. In addition, the distributed nature of Web service environments and their high degree of complexity requires a
comprehensive description of Web service quality characteristics, both functional
and non-functional. The main aspects of QoS in a Web service context, which are
partly derived from [MN02, ZBD+ 03, LNZ04, CSM+ 04, Pap08] are as follows:
Availability Service availability is the likelihood of absence of downtime, i.e. the
probability that a service is available for invocation. Small values indicate
an unpredictability of the service to be accessible at a certain point of time.
This probability can be estimated by incorporating historical data on a ser-

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

47

vice’s downtime. The ration of observed average downtime and total time
of potential availability results in an estimated probability of unavailability
for the future, whereas the probability of the complementary event reflects
an estimated probability of availability of a service.
Reliability Service reliability refers to the characteristic to function correctly and
consistently, i.e to produce the desired outcome or result. This is usually
expressed in transaction failures over a defined period of time. It can be be
measured using historical data of previous invocations and a corresponding
successful delivery.
Scalability The ability to service requests independently of volume is referred
to as the scalability of a service. Scalability is important in periods with
high peaks of demand with uncertain occurrence and hardly predictable
patterns.
Performance The service quality aspect performance consists of two parts,
throughput and latency. A service’s throughput refers to the number of requests that can be served at a defined time period. Latency of a service is the
time between sending a request and receiving the outcome or result. This
means that high throughput and low latency characterize a service with a
high degree of performance.
Security As Web services are usually provided over the Internet, security is an
important issue for service providers and consumers. Especially in order
to represent long-lived mission-critical business transactions that involve
private business information, Web services must fulfill serious security requirements such as access control (authentication, authorization), confidentiality, and integrity of information.
Reputation The reputation of a service is a measurement of its trustworthiness.
The value creation of a service is mostly dominated by intangible elements
and is therefore subjective to the individual that experiences a service’s outcome. As the sum of individual experiences is a suitable indicator for service quality, reputation is an important aspect that takes consumers’ experiences and opinions into account25 .
An agreement between service provider and service consumer about the QoS
to be delivered must be founded on a legal basis, i.e by specifying a service level
25 A

star ranking mechanism is a possible solution to capture consumers’ valuations for a service. An example can be found at http://aws.amazon.com/.

48

CHAPTER 2. PRELIMINARIES & RELATED WORK

agreement. A service level agreement is a contract that defines mutual understandings and expectations of a service between service provider and service consumer [JMS02]. It defines service characteristics and the quality to be delivered
by the provider and monetary penalties in case of non-performance. Such a contract represents a guarantee for the service consumer, which assures the delivery
of the defined quality or an adequate charge-back mechanism.
Depending on the frequency by which a service level agreement can be redefined and adapted according to changed requirements or conditions, two types
of service level agreements can be differentiated, static and dynamic service level
agreements. Static service level agreements generally remain unchanged for a
long period of time or multiple service time intervals. The quality of situational and short-termed Web services is covered by dynamic service level agreements that change from period to period. This type of service level agreement
is inevitable in highly dynamic environments where Web services are composed
and provisioned on-demand and roles of service provider and consumer change
quickly.

2.1.3.4

Web Service Coordination

Environments in which distributed units provide functionality in a looselycoupled manner (according to the SOA paradigm) require some sort of process
or set of rules to align activities in order to generate a desired outcome, i.e. they
require coordination. The objective of coordination is to make a set of entities –
either by providing incentives or establishing constraints upon them – pursue a
common goal, e.g. producing a defined outcome.
Definition 2.7 [C OORDINATION ]. Coordination is managing the dependencies of activities.26
Coordination can be formalized by designing adequate mechanisms, i.e sets of
rules that govern the interaction between the various entities. Coordination is
the key instrument to organize multiple activities especially in distributed environments. In the context of Web services two specifications provide frameworks to implement coordination scenarios, WS-Coordination [NRFJ07] and WSCF [CNLP05]. This section focuses on WS-Coordination as it is a finalized standard in contrary to WS-CF, which is still a public review draft. A detailed com26 The

definition of coordination is based on [MC94] and is consistent with literature from organization theory [Gal73]

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

49

parison of WS-Coordination and WS-CF can be found in [LW03] and [Kra05].
WS-Coordination is based on concepts and roles that are represented by Web services. Initiator, coordinator and participants communicate using a common context
that glues their interaction to the coordinated activity. The framework allows for
different coordination protocols to be plugged in to coordinate domain-specific
work between clients, services and participants. Work is defined as activities
performed by one or more distributed parties. Examples for specific transaction protocols are WS-AtomicTransaction [NRLW07] and WS-Business Activity
[NRFL07]. WS-AtomicTransaction specifies a rudimentary ACID27 transaction
protocol focusing on ad-hoc short-term transactions in a general manner. In
contrast, WS-BusinessActivity defines transactions with relaxed ACID properties
with the purpose to coordinate long-term business transactions.
The process of coordination and the roles involved according to the WSCoordination specification are depicted in Figure 2.11. The sequence diagram
illustrates the main phases activation, registration, invitation and communication.
—’’Š˜›

˜˜›’—Š˜›

Š›’Œ’™Š—

›˜˜Œ˜•ȱŽ›Ÿ’ŒŽ ŽšžŽœ’—ȱŽ›Ÿ’ŒŽ Œ’ŸŠ’˜—ȱŽ›Ÿ’ŒŽ Ž’œ›Š’˜—ȱŽ›Ÿ’ŒŽ ›˜˜Œ˜•ȱŽ›Ÿ’ŒŽ ŽšžŽœ’—ȱŽ›Ÿ’ŒŽ ›˜˜Œ˜•ȱŽ›Ÿ’ŒŽ

›ŽŠŽ˜˜›’—Š’˜—˜—Ž¡

›ŽŠŽ˜˜›’—Š’˜—˜—Ž¡Žœ™˜—œŽ
Ž’œŽ›


Ž’œŽ›Žœ™˜—œŽ
›˜˜Œ˜•ȱŽœœŠŽ





›˜˜Œ˜•ȱŽœœŠŽ
˜› Š›ȱ›ŽŠŽ˜˜›’—Š’˜—˜—Ž¡Žœ™˜—œŽ
Ž’œŽ›



Ž’œŽ›Žœ™˜—œŽ
›˜˜Œ˜•ȱŽœœŠŽ



›˜˜Œ˜•ȱŽœœŠŽ
˜–™•Ž’˜—ȱŽšžŽœ
˜–™•Ž’˜—ȱŒ”—˜ •ŽŽ–Ž—



˜–™•Ž’˜—ȱŽšžŽœ
˜–™•Ž’˜—ȱŒ”—˜ •ŽŽ–Ž—

Figure 2.11
WS-Coordination sequence diagram.

27 ACID

stands for atomicity, consistency, isolation and durability, which are properties that guarantee a reliable transaction.

50

CHAPTER 2. PRELIMINARIES & RELATED WORK

Activation The WS-Coordination framework exposes an activation service that is
responsible for the creation of specific coordinator instances with concrete
protocols and associated context. To start a coordination process, the initiator sends a CreateCoordinationContext message to the endpoint of
the activation service in an asynchronous manner. The coordinator either
replies with a CreateCoordinationContextResponse message or an
error message. A CreateCoordinationContext message has the following structure:
The CoordinationType points to a uniform resource identifier that speci1
2
3
4
5
6

<CreateCoordinationContext ...>
<CoordinationType> ... </CoordinationType>
<wsu:Expires> ... </wsu:Expires>
<CurrentContext> ... </CurrentContext>
...
</CreateCoordinationContext>

Listing
2.2:
Structure
CreateCoordinationContext Message

of

a

fies the type of coordination to be used in the coordination process (e.g. WSAT, WS-BA). wsu:Expires is an optional argument that defines a time-out
value for the corresponding coordination context. The semantic of this argument depends on the coordination type used. The CurrentContext
argument is also optional and can be used to hand over an existing context
(activity import). In this case, the coordinator participates at the running
activity instead of creating a new context.
In case the activation is successful, the coordinator replies asynchronously
with a CreateCoordinationContextResponse message that is structured as follows:
The CoordinationContext consists of a unique Identifier that guar1
2
3
4
5
6
7

<CreateCoordinationContextResponse ...>
<CoordinationContext>
<Identifier> ... </Identifier>
<CoordinationType> ... </CoordinationType>
<RegistrationService> ... </RegistrationService>
</CoordinationContext>
</CreateCoordinationContextResponse>

Listing
2.3:
Structure
of
CreateCoordinationContextResponse Message

a

antees a well-defined mapping from message to activity. The argument
CoordinationType defines the type of coordination. The actual endpoint
reference to the registration service exposed by the coordinator is specified

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

51

using WS-Addressing [BCC+ 04] in the RegistrationService section.
The registration service is responsible for handling registration requests
from participants that intent to participate in the activity.
Registration Once a coordinator has been activated by the activation service, a registration service is exposed that allows for participants to
register for being part of the activity and to send – if this is supported
by the coordination protocol – and receive protocol messages. Via the
CoordinationContextRespond message, the initiator receives and
endpoint reference to the registration service. By sending a Register
message to this uniform resource identifier, the initiator’s registration
is confirmed by the coordinator with a RegisterRespond message.
The RegisterRespond message contains and endpoint reference to the
protocol service of the coordinator that is responsible for managing the
communication between participating roles. A Register message is
structured as follows:
The ProtocolIdentifier argument specifies the coordination protocol
1
2
3
4
5

<Register ...>
<ProtocolIdentifier> ... </ProtocolIdentifier>
<ParticipantProtocolService> ... </ParticipantProtocolService>
...
</Register>

Listing 2.4: Structure of a Register Message

that is supported by the chosen coordination type of the coordination context. An endpoint reference to the protocol service of the initiator is defined
in the ParticipantProtocolService section as the destination for
further communication. In case of a successful registration, the coordinator
sends a RegisterRespond message to the initiator that is structured as
follows:
The registration response message contains the endpoint ref1
2
3
4

<RegisterResponse ...>
<CoordinationProtocolService> ... </CoordinationProtocolService>
...
</RegisterRepsonse>

Listing 2.5: Structure of a RegisterResond Message

erence to the protocol service of the
CoordinationProtocolService section.

coordinator

in

the

52

CHAPTER 2. PRELIMINARIES & RELATED WORK

Invitation Recall, the CreateCoordinationContextResponse message
contains the endpoint reference to the registration service of the coordinator and can therefore be used as an invitation or call for participation. By
forwarding the message to potential participants they obtain the possibility
to register for the activity at hand. Although the initiator normally invites
further participants, one can think of multiple scenarios with different roles
to be the inviting party in the process. The coordinator can step into the
role of pushing the invitation process using a UDDI registry to find suitable participants. It is also possible to reverse the roles in such a lookup
scenario, meaning that potential participants are proactively searching for
suitable coordination services. Potential participants could also subscribe
to a notification service – analogue to the observer design pattern – using
the WS-Notification [GNC+ 04] specification in order to automatically be
informed if an adequate coordination service is available.

Communication Initiator and participants share common knowledge about the
endpoint reference of the coordinator’s protocol service. Depending on the
coordination type and the activity that is realized by the coordination process, initiator and participants use the protocol service of the coordinator to
exchange messages in an asynchronous manner. The registration phase also
provides the coordinator with the necessary address information about the
active parties to be able to respond to incoming messages.

Completion Termination of the coordination process is usually initiated by the
initiator. The initiator sends a completion request message to the coordinator that acknowledges the request by a completion acknowledge message.
The coordinator informs all registered participants by sending a completion request message. A confirmation of each registered participant is then
responded as a completion acknowledge message back to coordinator.
Example 2.3 [WS-C OORDINATION COMPLIANT REVERSE AUCTION ]. To illustrate the specification of a coordination model according to the WS-Coordination framework, an auction mechanism is introduced as a special type of coordination, i.e a single
item sealed bid reverse auction. There is one buyer that intents to procure a single good or
service from multiple sellers. The auction conduction including the type of messages to
be exchanged between the participants is specified by auction rules which are controlled
and enforced by an auctioneer. The mapping between roles and entities in a reverse
auction and a coordination model is depicted in Figure 2.12.

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

Reverse
Auction

53

Auctioneer
Seller

Buyer

Coordination
Model

Auction
Rules

Seller

Coordinator
Participant

Initiator

Coordination
Protocol

Participant

Figure 2.12
Mapping of a reverse auction to a coordination model.
The buyer starts the auction by announcing a request for the desired good or service.
The auctioneer receives sealed offer bids from the sellers by a public deadline. After the
deadline the winner determination is performed by the auctioneer, the good or service is
transferred and the winning seller receives its payment.
Based on the WS-Coordination framework, the buyer is represented by the initiator
and the sellers are instances of the participant role. The auctioneer as the coordinator is responsible for the coordination protocol, that is, the set of auction rules. The initiator starts
the activation phase and receives a coordination context from the coordinator. The invitation phase is generally done by the initiator according to [NRFJ07]. Nevertheless this
not practicable for the reverse auction scenario as the buyer is not necessarily responsible
for the discovery and selection of potential sellers. As the WS-Coordination framework
provides a generic coordination model independent of a domain-specific application logic,
a tailored invitation process can be implemented on-top in order to shift responsibilities.

2.1.4 Service Value Networks and Situational Applications
Complete industries are moving from integration to specialization. Hierarchically
organized firms that started to cooperate in firmly-coupled strategic networks

54

CHAPTER 2. PRELIMINARIES & RELATED WORK

with stable inter-organizational ties recently explore the benefits of exploiting
more loosely-coupled configurations of legally independent firms. In theory,
complex products or services can be produced by a single vertically integrated
company. However, doing so, the company cannot focus on its core competencies since it has to cover the whole spectrum of the value chain. Also, it has to
burden all the risks in a complex, changing and uncertain environment by itself.
2.1.4.1

Networks as a Type of Governance Form

As a consequence, business networks (BNs) have been proposed as the superior governance form for today’s highly dynamic and complex business world
[MS86]. Business networks evolve from a pool of potential horizontal as well as
vertical business partnerships. In this respect they differ both from strategic alliances, comprising only horizontal business partners, and supply chains, denoting
purely vertical relationships. The advantages of business networks compared to
more traditional governance forms are manifold:
• Insurance against uncertainty in demand and supply.
• Balancing adaptability to highly complex tasks while maintaining control.
• Protection of business knowledge through modularization.
• Market-based forces as coordination mechanism to ensure efficiency.
A bulk of managerial and academic literature deals with variations of such
business networks, whose complete characterization would be far beyond the
scope of this section. In this section, Service Value Networks (SVNs) as a special
type of business networks are identified and the differences to related organizational forms, which are to described in the following are described.
Virtual Organizations (VOs) are temporary networks of independent enterprises that bring in complementary competencies and resources for mutual benefit [DM93]. Virtual organizations stress the complementarity of firms’ core competencies in the value creation process and the temporary nature of the interaction. However, virtual organizations often suffer from trust related problems and
are therefore usually constituted among firms in a closed pool of known network
partners.
Smart Business Networks (SBNs) are one way beyond the virtual organization framework and particularly stress the smart use of information and communication technology (ICT) as a facilitator to network interaction. Smartness is

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

55

thereby a relative term, which refers to effectiveness and a comparative advantage through the use of ICT. Moreover, ICT is also seen as an enabler of network
agility, i.e. the network’s ability to “rapidly pick, plug, and play” business processes [vHV07]. Furthermore, nodes in a smart business network need to meet
specific requirements in order to be ready to contribute to ad-hoc joint value creation. This modularity of potential network members allows not only for spontaneous network orchestration, but also for better protection of a firm’s core competencies as compared to virtual organizations. Trust problems are thus not as
severe and the smart business networks may therefore recruit members from a
more open pool of potential partners. The instantiations of smart business networks are also more short-lived than those of virtual organizations. However,
like in virtual organizations, the network pool itself is sustainable over time.
Business Webs are defined as “customer-centric, hetrarchical organizational
forms that consisting of legally independent but economically interdependent
specialized firms that co-opetitively contribute modules to a product system
based on a value-enabling platform under the presence of network externalities which are supported by extensive usage of information and communication
technologies.” [Ste04]. Business Webs stress the internet as the primary channel
for business communications [TLT00]. Moreover, the so-called “shaper-adapter
configuration” is an important assumption: A shaper (i.e. a focal company or
nucleus) controls the central element in a business web, while adapters (i.e. context providers) add complementary elements. A closely related field of research
considers Business Ecosystems whose quintessence is each participant’s ultimate
connection to the fate of the network as a whole [IL04].
In this context, service value networks are a special type of smart business networks with features of business webs. They exhibit the crucial features of smart
business networks, such as the smart use of ICT, agility, ad-hoc value creation
and sustainability of the network pool. With respect to business webs, service
value networks share the feature of being enabled through ubiquitously available ICT, foremost the Internet. However, service value networks are distinct to
business webs because they do not follow the shaper-adapter paradigm and are
rather constituted by market-based composition from an open pool of network
partners.

2.1.4.2

Service Value Networks

Companies tend to engage in networked value creation, which allows participants to focus on their strengths. Partners in such ecosystem-like environ-

56

CHAPTER 2. PRELIMINARIES & RELATED WORK

ments can leverage the know-how and capital assets of partners, at the same
time spreading risk and sharing investment cost. Focusing on core competencies
does not put constraints on the company or limit its reach. In contrary, by reaggregating with partners, a network of companies can broaden its range of customer attraction. Especially in complex and highly dynamic industries, forming
such open networks is more than an attractive strategic alternative. Service value
networks bring together mutually networked, permanently changing, legally independent actors in customer centric, mostly heterarchical organizational forms
in order to create joint value for customers. Specialized firms co-opetitively contribute modules to an overall value proposition under the presence of network
externalities.
There is still only few research in the context of service value networks, especially regarding attempts to provide a definition. Service value networks are
constituted by loosely-coupled formations of companies that provide modularized services while concentrating on their core competencies. These Web-enabled
services expose standardized interfaces and foster an ad-hoc composition in order to jointly generate added value for customers in an on-demand fashion. This
argumentation leads to the following definition:
Definition 2.8 [S ERVICE VALUE N ETWORK ]. Service value networks are goaloriented business networks, which provide business value through the agile and marketbased composition of complex services from a steady, but open pool of complementary as
well as substitutive standardized service modules by the use of ubiquitously accessible
information technology.
To foster a fundamental understanding of the service value network concept,
Figure 2.13 depicts the main components and their interdependencies in a simplified manner.
A service value network consists of a set of service providers (s ∈ S) that supply a portfolio of service offers (v ∈ V) that provide specified functionality. Each
service provider can own one or multiple service offers, indicated by an ownership relation. The example in Figure 2.13 shows a service value network with four
service offers (v1 , v2 , v3 , v4 ) that are owned by three service providers (s1 , s2 , s3 ).
Service offers that are substitutes – which provide roughly similar functionality –
are clustered in candidate pools (Y ∈ Y ). A candidate pool is a set of potential service offers that are substitutes and can therefore be replaced on-demand. Service
offers that are compatible, this is, they are interoperable regarding their interfaces
and input and output capabilities, expose a directed composition relation. Service

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

s1

s2

57

Caption

s3
s

Service Provider
Ownership
Relation

v1

v2

v

Service Offer
Composition
Relation
Source Node

Sink Node

v3

v4
Candidate Pool
Y

Yb

Ya

Complex Service

Figure 2.13
Service value network model.
offers – clustered into candidate pools – and their connections form a graph-like
structure that is directed and a-cyclic starting from a source node and ending at
a sink node. Each feasible connected set of service offers within this graph is
called a path and represents a possible instantiation of a complex service consisting
of functionality from each candidate pool. According to the example in Figure
2.13, a complex service can be instantiated either by a composition of v1 and v2 or
v1 and v4 or v3 and v4 .
Service Providers and Service Offers The number of service providers offering
various types of utility, elementary and complex services in ecosystem-like
environments is constantly increasing.
Exemplarily, Amazon offers utility services based on their infrastructure as
a computing and a storage service called Elastic Compute Cloud (EC2)28
and Simple Storage Service (S3)29 that are accessible and manageable
through simple highly standardized interfaces based on REST and WSDL.
In most cases, such cloud computing infrastructures are organized in a
cluster-like structure facilitating virtualization technologies. Nevertheless,
there are service providers that focus on offering computing on-demand
28 http://aws.amazon.com/ec2/
29 http://aws.amazon.com/s3/

58

CHAPTER 2. PRELIMINARIES & RELATED WORK
through a server Grid such as the Sun Grid Computing Utility30 . Among
providing pure utility services, providers such as RightScale31 often enrich
their offerings through value-added elementary services for managing the
underlying hardware (i.e. scaling, migration) that are accessible via Web
front-ends.
Service providers such as StrikeIron32 offer a comprehensive portfolio of
elementary and complex Web services that provide functionality in the context of communications, customer relationship management (CRM), data
enhancement, e-commerce, finance, and marketing. Especially in the financial sector, companies (e.g. Xignite33 ) sell Web services providing financial
information such as real-time stock quotes, options, historical data, commodity prices, mutual funds, currency rates, and financial market indices.
Nevertheless, not only rather simple, but also complex services supporting
multi-step business processes are offered modularized in an on-demand
fashion. For instance, providers like salesforce.com34 or Netsuite35 successfully entered the business software ecosystem with their entirely Webbased on-demand customer relationship management (CRM) suites. Components offered within these suites can be dynamically composed to customized complex services. AppExchange36 , the service marketplace offered
by salesforce.com, offers a range of pre-integrated complementary services
provided by third-party vendors grouped around the core service Salesforce
CRM.

Service Requester The open and dynamic character of service value networks
enables customers to request customized complex services from whatever
service value network they prefer that satisfy their needs and match market
requirements. Service requesters creatively create their complex services by
composing adequate service components from multiple candidate pools in
a plug-and-play fashion in order to receive added value. By concentrating
on their core competencies, companies are not forced to provide solutions
covering the whole range of a business process but they are able to complement their service portfolio by requesting complex services from service
value networks (cp. Example 2.1).
30 http://www.network.com/
31 http://www.rightscale.com/
32 http://www.strikeiron.com/
33 http://www.xignite.com/
34 http://www.salesforce.com/
35 http://www.netsuite.com/
36 http://www.salesforce.com/appexchange/

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

59

Candidate Pool The structure of service value networks, characterized by their
participants and their interrelations, is not static and predefined but formed
on-demand in a short term, goal-oriented fashion. The formation process
requires a steady pool of distributed and loosely-coupled service providers
that offer predefined functionality through modularized services to be
ready on call. In order to participate in service value networks, i.e. participate in candidate pools to be ready for service provision, service providers
must register at a central registry and satisfy a set of minimum requirements
such as interoperability through well-defined interfaces based on Internet
standards. The process of registration can be activated by switching initiators, meaning that also an operator of a central registry might query and
proactively invite suitable service providers to join a candidate pool. The
open character of service value networks allows any service provider to potentially participate in value creation as long as minimum requirements are
met.
Candidate pools group service offers of multiple service providers by functionality and capabilities exposed. Service offers covering the same spectrum of functionality (e.g. login/ID services such as OpenID37 and Google
Accounts38,39 ) are categorized in identical candidate pools. These services
are replaceable and represent service substitutes form an economic perspective. The actual formation process occurs when a concrete service request
is addressed to the loosely formation of service providers. Based on the required functionality and capabilities described by the request, feasible candidate pools are iteratively arranged in a way that they together contain the
potential to jointly generate desired value. A coordination mechanism is
required to chose a single service offer from each candidate pool based on a
set of rules in order to efficiently instantiate the requested complex service
to be provided to the service requester.
Complex Service The final outcome that is produced by a service value network
is realized through a sequence of modularized service offers from a set of
iteratively arranged candidate pools (cp. Figure 2.13), that is, a complex
service. This final outcome is the added value generated for the service
requester. The concept of a complex service, its characteristics and the way
it is composed is explained in detail in Section 2.1.2.3.
37 http://openid.net/
38 https://google.com/accounts/
39 Note

that the Google Accounts service is not an adequate candidate to participate in an service value network in a strict sense, as it is proprietarily bound to Google services and does not
expose a well-defined interface to be accessed in an open manner.

60

CHAPTER 2. PRELIMINARIES & RELATED WORK

Coordination Mechanism In environments with distributed, self-interested entities that jointly contribute to an overall goal, mechanisms are needed that
coordinate procedures from multiple parties with possibly colliding objectives. Service value networks are a prominent example of such complex
environments and their success therefore highly depends on adequate and
efficient coordination mechanisms. As already mentioned in Section 2.1.3.4,
coordination is managing the dependencies of activities. It is obvious that there
exist various facettes of coordination forms that have to be chosen according
to the characteristics and requirements imposed by the type of environment.
The continuum of coordination ranges from market-based approaches to
hierarchical control and dictatorships [Tho91, MC94]. Market-based approaches manage the activities of distributed, self-interested entities only
indirectly by institutionalizing a rule set that incentivizes market participants to act in a desired manner in order to achieve an overall goal. Actors
and dependencies of their activities are managed ’invisible’ and ’unseen’
driven by rational behavior of utility-maximizing economic entities and incentivized by rules to perform a social choice and compensate the entities’
efforts. Nevertheless, there are situations in which this ’liberal’ form of coordination results in inefficient outcomes. In this case, the economic entities
need to be consciously organized in hierarchical forms to streamline activities in an efficient manner.

The problem of efficiently choosing adequate service offers from candidate
pools to instantiate a complex service that meets the requirements imposed
by the service requester is a traditional problem of coordination. Service
providers are self-interested, act rational and therefore try to maximize their
utility without accounting for a system-wide solution (e.g. a solution that
maximizes welfare). Thus, the design of adequate coordination mechanisms is crucial to the efficiency and success of a service value network.
Example 2.4 [SVN R EALIZING A CRM C OMPLEX S ERVICE ]. This example shows
the formation of a service value network that is ready to instantiate a complex service
based on the requirements imposed by a service request. A service requester requires a
complex service that scans calendar entries within the upcoming week with regard to
future meetings within a company. Based on the the meetings’ descriptions, the complex
service queries soft skills of all meeting participants by browsing their profiles in social
communities. Gathered information is then updated in a CRM data base that is stored by
on-demand storage infrastructure (Figure 2.14).

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

61
Caption

Salesforce
Service Provider
Google

Amazon
Facebook

LinkedIn

Strategic
Alliance
Ownership
Relation
Service Offer

Google
Calendar

Facebook
Browser

S3

Composition
Relation
Source Node

Sink Node

LinkedIn
Browser

App
Engine

Figure 2.14
Example of a service value network realizing a CRM complex
service.
A set of service providers participates in the service value network by offering services
grouped in candidate pools. Google offers its Google Calendar service40 and Google App
Engine41 which provides a scalable infrastructure for service development and storage.
The social community platforms Facebook42 and LingedIn43 provide services to browser
profiles of registered users. Amazon offers flexible storage capabilities through its Simple
Storage Service (S3)44 . As depicted in Figure 2.14 the requested complex service can be
realized in four different versions by selecting feasible service combinations (e.g. Google
Calendar, LinkedIn Browser and Amazon S3).
This example shows that service value networks foster the ad-hoc creation
of short-living complex services that fulfill individual needs of a variety of consumers. This type of complex service is also called service mashup or situational
application. The following section introduces fundamentals of situational applications and service mashups, explains their role within service-ecosystems, and
introduces key principles they are based on.
40 http://google.com/calendar
41 http://code.google.com/appengine/
42 http://facebook.com/
43 http://linkedin.com/
44 http://aws.amazon.com/s3/

62

CHAPTER 2. PRELIMINARIES & RELATED WORK

2.1.4.3

Situational Applications and Service Mashups

Competitive forces in today’s markets result in the fact that dealing with change
is a necessity for companies. This needs to be exploited and enabled by achieving
flexibility in the organization and IT infrastructure [Eva91, GS06, AB91]. Flexibility is mainly concerned with the quick development of new applications to
support changing business processes. In the past, IT departments have fallen
short to satisfy the demand for new applications. Typically, situational applications that are needed only for a limited time span never made it into realization
in favor of strategically important applications as part of the development backlog. Nowadays, most of the efforts of the IT departments are devoted to maintenance leaving many application wishes unfulfilled. With the advent of Web
2.0 technologies and the renaissance of HTTP appreciation, the possibilities to
build “good enough” applications have greatly increased and traditional roles of
service provider and service consumer blur.
A so-called service mashup is an application or Web site that aggregates content such as data feeds, applications, widgets, or gadgets from different sources
[Mer06]. The number of publicly available mashups is dramatically increasing and can be checked at programmableweb.org45 . While the first mashups
were dedicated to small consumer mashups, where simple data (e.g. RSS feeds
[BDBD+ 00]) is integrated in the Web browser, mashup technology promises to
integrate enterprise applications. In fact, mashups can be considered to provide
solutions for the long tail of applications [And06].
As depicted in Figure 2.15, standard applications (such as ERP modules) are
standardized, but need customization. This mass market exhibits only small degrees of customization but enjoys demand by many customers, i.e. volume business. Software companies have been exploiting these market segments. However,
there is also a long tail of applications, which require highly specialized features
– accordingly, this highly specialized software cannot be offered to many customers in scalable manner. It is thus not astonishing that these segments around
the long tail have so far not been exploited. Summarizing, the long tail of applications is very fat in a sense that the demand for customized and quality differentiated software is immense, i.e. value business. Due to the diversified demand
there are numerous, hitherto unexploited niche markets, where the project set-up
costs exceed the benefit.

45 http://programmableweb.org/

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

Mass Market

Niche Market

Situational/Tailored
(Service Mashups)

Demand

Satisfaction

Off -the-Shelf
(SaaS)

63

Service Customization
Figure 2.15
Situational applications address the long tail of business.
With the technology of mashups, it is now possible to exploit the long tail as
customization becomes cheaper through the aggregation of small services. Big
and RESTful Web services encapsulate functionality and put it behind clearly
defined interfaces based on SOAP, WSDL and HTTP respectively. Typically, it
is distinguished between consumer, data and enterprise mashups. In fact, consumer mashups combine data elements from different sources and hides them behind a simple GUI (e.g. TuneGlue being an interactive visualization of the music
artists available at Last.fm46 which is linked with Amazon customer data). Data
mashups combine data streams from different sources into one single data feed
with one dedicated user interface attached to it. Enterprise mashups integrate
data and other services (e.g. infrastructure services) from internal and external
sources creating composite Web applications. Because of the simplicity in setting
up composite applications, mashup technologies are expected to evolve significantly. Fierce competition and the corresponding needs for applications coerce
companies into imperatives of the modern service-oriented economy that opens
up the long tail of strong differentiation of their service offerings, and customercentricity in the creation of services.
Service mashups also allow end-users to create customized applications by
combining content, presentation functionality and business logic from heterogeneous sources using lightweight Web technologies. Through the extensive reuse
46 http://lastfm.com/

64

CHAPTER 2. PRELIMINARIES & RELATED WORK

of existing resources and simple programming models mashups facilitate the adhoc development of highly situation-specific applications which are often used
for a short time only. Mashups therefore support the long tail of business, which
cannot be served by traditional off-the-shelf software. Situational applications
embody the next step in service-oriented computing and their ease of use heralds
the next generation of flexibly recombined services. The following principles encompass the key innovation of situational applications:
Principle 2.1 [S IMPLIFICATION AND S TANDARDIZATION ]. Service mashups and
the way they are developed is a prominent result of a clear trend towards the simplification and standardization. Even complex services are increasingly exposed in the manner
of puristic service descriptions and interfaces. As explained in Section 2.1.3.2, RESTful architectural styles leverage the power of the highly standardized and interoperable
HTTP protocol. HTTP methods (e.g. GET, DELETE and CREATE) are used to build the
most elementary signatures encapsulating scalable functionality in a distributed fashion. Unlike heavy-weight RPC-style architectures with high XML payload and complex
programming-language-like interfaces, RESTful Web services are founded on unified interfaces based on HTTP methods and scoping information encoded in the service’s URI.
Principle 2.2 [L IGHTWEIGHT C OMPOSITION AND F LEXIBLE B INDING ]. Puristic
Web APIs such as REST and other lightweight approaches to Web service protocols and
messages formats (e.g. JSON) enable ad-hoc composition and flexible binding of replaceable services [Jhi06]. Situational applications mostly focus on simple data manipulation
and can therefore be piped sequentially. Well-defined building blocks as components of
these sequences can be composed, decomposed and rearranged dynamically and enable
demand-driven customization and satisfaction of individual consumer needs. A high degree of standardization regarding service interfaces allows for the specification of reusable
service blueprints that define a skeleton of service mashups. Service components within
these blueprints can be bound and instantiated at run-time as they are replaceable and
puristic in nature.
Principle 2.3 [M ASS C OLLABORATION AND C USTOMIZATION ]. The central principle of a continuous development of situational applications is collaboration and customization [Mul06]. Participants are part of a mass co-production process that blurs
the border between creation and consumption. Users contribute their individual knowledge about the existence, capabilities and compatibilities of feasible service components to
service mashup models. A high degree of customization and self-selection continuously
generates new demand and satisfies niche markets in the long tail [And06].

2.1. SERVICE CONCEPTS, DEFINITIONS, AND TECHNOLOGIES

65

Principle 2.4 [P ERPETUAL B ETA ]. The development of service mashups is comparable
to agile software development and extreme programming [Mul06]. Multiple users continuously create and re-engineer service compositions using components that are mostly
under the control of distributed owners. Service mashups are living applications that
never reach a final state. They are created and improved through a trial-and-error-process
that involves many participants manipulating models according to their needs and mostly
self-interest.
The following example illustrates the idea behind service mashups and how
key principles are realized in the context of consumer mashups.
Example 2.5. As an example consider a user Anna who wants to blog links about horseback riding on Iceland. The link list should be updated automatically as new articles
about this topic are published on the Web. Since manual creation of the link list is therefore not possible, Anna decides to quickly create a tiny mashup for gathering, tagging and
displaying the links.

Newsfeed

Tagging

Translation

Google
Search

Tagthe.net

Google
Language

Yahoo!
Search

Yahoo!
Term
Extraction

Yahoo!
Babel Fish

Microsoft
Live Search

Zingo
Tag FInder

Figure 2.16
Blueprint of a translation and tagging service mashup.
As depicted in Figure 2.16, the mashup requires a newsfeed, tagging and translation
service. Newsfeed services take the desired topic as input and return relevant news ar-

66

CHAPTER 2. PRELIMINARIES & RELATED WORK

ticles. In the following, relevant tags have to be determined for these articles. As Anna
would like to keep her blog consistent in German, a service is required to translate the
foreign language tags.

2.2

Markets in a Service World

The community is a fictitious body, composed of the individual persons who are
considered as constituting as it were its members. The interest of the community then is,
what? – the sum of the interests of the several members who compose it?
[Ben38]
This section elaborates the idea, necessity and applicability of markets in servicedominated environments which are constantly evolving in almost any field of
society. Providing a first insight and a general motivation to the topic, Section
2.2.1 provides a thorough line of argument answering the question why auctions
should be applied in the context of complex services and how they can serve
to coordinate distributed activities to enable a flawless composition. The argumentation builds upon the general service characteristics as introduced in Section
2.1.1.2 and proclaims the need for auction-based dynamic allocation and pricing
of service components generating added value through the composition of complex services.
Laying the groundwork for the design of mechanisms, Section 2.2.3 introduces the approach of mechanism design, elaborates economic objectives that are
desirable when implementing a social choice, and briefly introduces prominent
mechanisms along with a set of impossibility theorems. Bringing mechanism design in the context of service value networks and information systems design, the
idea behind algorithmic mechanism design is motivated.
As the process of designing market-mechanisms for a specific domain is complex and involves many steps and multiple factors, Section 2.2.2 introduces the
concept of an electronic market and provides a market engineering process as a
structured approach for the discipline of market engineering. Each phase within
the market engineering process is iteratively mapped on the structure of the work
at hand.
The Section 2.2.4 concludes with a detailed analysis of economic and applicability requirements, an auction mechanism has to meet to support dynamic
allocation and pricing of complex services in networked environments such as

2.2. MARKETS IN A SERVICE WORLD

67

service value networks. Based on the requirements analysis, related work is presented and evaluated illustrating the research gap which is filled by this thesis.

2.2.1 Why Auctions for Complex Services?
In general, an adequate approach for allocation and pricing of complex services
has to account for service characteristics as introduced in Section 2.1.1.2. As stated
by [Smi89] “auctions flourish in situations in which the convential ways of establishing price and ownership are inadequate”. Smith concretizes the argumentation by briefly pointing out the main characteristics of such situations which are
predestinated for the application of auctions by focusing on the roles and items
involved: “costs cannot be established, [...], there is something special or unusual
about the item, ownership is in question, different persons assert special claims, [...].”
Although this statement is rather fuzzy, the characterization of the type of ’item’
which price is best established by the application of an auction mechanism opens
up an analogy to the service concept. Recall, in Section 2.1.1.2 services are characterized by the coincide of production and consumption (uno-actu), they cannot
be inventoried, value creation is dominated by intangible elements, consumer
co-production and fuzzy inputs and outputs.
Smith points out that auctions are preferable in situation where costs cannot
be established. From an microeconomic perspective such costs refer to internal
costs that are private information to the one producing the item, i.e. the producer’s
individual valuation for the item. In the context of services, this argument also
holds for the consumer side. According to the service characteristic C 2.4, value
that is generated for the service consumer is mostly dominated by intangible elements and therefore hard to determine. An objective measurement of quality
which might be an indicator for the consumer’s valuation is also hardly applicable due to a service’s fuzzy inputs and outputs according to characteristic C 2.5.
The complexity of value elicitation and the problem of establishing adequate prices
even increases in scenarios with joint value creation through service compositions
(e.g. in service value networks where complex services are produced). Analogue
to Smith’s argumentation, such problems can be addressed by the design of a
suitable auction mechanism that induces incentives for service providers to report their private valuations truthfully. Auctions haven proven to be the ideal
instrument to aggregate information from distributed parties which results in an
aggregated valuation [PS00, Jac03]. Without prior knowledge about the valuations of each participant, auctions can provide suitable incentives to make truth
revelation an equilibrium strategy and therefore automatically aggregate neces-

68

CHAPTER 2. PRELIMINARIES & RELATED WORK

sary information from self-interested participants to determine adequate prices for
complex services.
Another criterion that is crucial to establishing a suitable approach for allocation and pricing according to [Smi89] is if the item subject to trade exposes
special or unusual characteristics. The uno-actu principle (C 2.1) implies that
in the context of services there cannot be a producer without at a consumer as
production and consumption coincides in time. This service characteristic has fundamental implications on coordination aspects as service cannot be inventoried
in order to balance demand and supply. Following the same direction, LuckingReiley enriches this argumentation by adding an economic perspective which
explicitly focuses on the trade of services by stating that “[...] in the future we
may see much more auctioning of services [...]. Services are particularly attractive for auctions because they are in relatively fixed supply – unlike durable goods,
one cannot store surpluses or draw down inventory in order to meet fluctuating demand.” [LR00]. Market mechanisms such as auctions are preferable in situations
with a fast changing demand and supply ratio as dynamic pricing smoothes high
amplitudes. This property is crucial to success of efficient allocation and pricing
especially when perishable services are traded [Eso01].
The rapid growth of information and communication technology has tremendously decreased transaction costs for service provision and consumption.
Computing power and storage raises exponentially while prices drop antiproportionally for hardware as illustrated by Moore’s Law. This development
directly leads to a tough price competition for service providers. In order to stay
competitive, service providers have to differentiate their service offers with respect
to quality (not price) [Dev98, MV98, DLP03, LSW01, BP91]. Quality is the main
value-determining factor in the context of services as service consumers experience
a service activity mainly based on the quality provided. Quality is idiosyncratic
to the individual and often determined by various factors and the interplay of
multiple service components that are part of a service composition. Hence, it
is unbearable for service consumers to reason about all feasible combinations of
single services and the resulting quality provided by the service composition in
order to meet their requirements. Therefore an auction mechanism is needed
which accounts for different preferences of service requesters defined for a variety of
quality characteristics that are determined by each component that is part of feasible complex service instances (cp. Section 2.1.2.3). Especially in the context of
situational complex services provided by distributed parties in service value networks, a QoS-sensitive auction mechanism allows for the provision and pricing of
highly customized short-term solutions to various types of customers leveraging

2.2. MARKETS IN A SERVICE WORLD

69

the nature and benefits of situational applications and service mashups (cp. Section 2.1.4.3). As a consequence, service providers in service value networks are
able to address the long tail of business by satisfying a great amount of individual
service requests [And06]. In these environments, it is assumed that service offers are under the control of distributed self-interested owners. In the absence of
central control, non-performance or complete drop-outs of service components
maybe rare but inevitable. Auction mechanisms that are computational feasible allow for reallocation and price adaption during run-time enabling dynamic failovers
in unreliable environments [FKNT02].

2.2.2 Electronic Markets and Market Engineering
Coordination of transactions requires an adequate form of organization and coordination mechanism. From an economic theory perspective, two extreme forms
have to be distinguished: markets and hierarchies. Markets coordinate transactions by means of a rule set which constraints the way transactions may take
place. The coordination itself results from a balance between demand and supply and consequently determines dynamic prices, quantities, quality and so forth.
In the past, markets have been used in environments with relatively simple products with respect to attributes and quality and low specificity (e.g. commodity
goods) due to high coordination costs for message exchange and matching of
demand and supply (cp. Figure 2.17). In the absence of modern information and
communication technology, complex products or services are costly to coordinate
(e.g. complex descriptions require complex bidding languages and messages as
well as highly sophisticated matching algorithms) [MS84]. Traditionally, in scenarios with complex products, hierarchies have proven to perform quite well due
to a higher degree of planning and control, which results in lower coordination
costs (less messages have to be exchanged and no complex matching is required).
A detailed analysis of trade-offs between markets and hierarchies with respect to
transaction and coordination costs can be found in [Wil79, Mal85, MS84, Mal87].
However, this argumentation does not hold under the presence of modern
information and communication technology and powerful dynamic infrastructures built upon the principles of the SOA paradigm. Due to more efficient and
sophisticated information and communication infrastructures, market-based coordination in electronic environments can be realized [MYB87]. Therefore the
following definition of an electronic market can be concluded:

Low

High

CHAPTER 2. PRELIMINARIES & RELATED WORK

Complexity of Product Description

70

Hierarchy

Market

Low

High
Asset Specificity

Figure 2.17
Characteristics of products and services affect forms of
organization [MYB87].

Definition 2.9 [E LECTRONIC M ARKET ]. An electronic market is an institutions built
upon information and communication technology that establishes a market-based coordination of transactions by enabling the ubiquitous trade of products and services between
multiple distributed participants.
Designing market mechanisms in electronic environments is a complex process that requires knowledge and expertise in the area of economics and computer science. Interdependencies between economic desiderata such as allocation efficiency (cp. Section 2.2.3) and technical applicability requirements such as
computational tractability have to be identified and feasible trade-offs have to
be analyzed in order to achieve desired goals [WNH06]. Different aspects from
technical and economic viewpoints often lead to colliding objectives that have
to be resolved through relaxation of requirements and objectives or designing
suitable trade-offs between conflicting goals. Relying on existing market mecha-

2.2. MARKETS IN A SERVICE WORLD

71

nisms originally designed for other environments may often lead to poor market
performance and inefficient outcomes [Lai05].
Hence, the process of designing markets for a specific domain must be wellstructured and based on a solid engineering methodology. The market engineering process according to [Smi82, Neu04, WNH06] is structured as depicted in
Figure 2.18. It mainly consists of four stages: Environmental analysis, design and
implementation, testing, and introduction. Each stage is briefly introduced in the
remainder of this section.
Operating Electronic Market

Introduction

Tested Electronic Market

Testing & Evaluation

Preliminary Electronic Market

Design & Implementation

Specification of Requirements

Environmental Analysis

Formalization of Objectives and Strategies

Figure 2.18
Stages of the market engineering process [Neu04].

2.2.2.1

Environmental Analysis

The environmental analysis is the first phase of the market engineering process and
comprehends the phases environmental definition and requirement analysis.
The environmental definition targets the gathering of necessary information
that allows for an efficient market design. This information covers the characteristics and types of objects that are subject to trade, possible market participants,

72

CHAPTER 2. PRELIMINARIES & RELATED WORK

their objectives and possible strategies as well as information about intermediaries in the market as analyzed in Chapter 2. Based on this information, potential
market segments are identified and evaluated comparatively.
Hence, this analysis serves as a basis for deriving requirements and desiderata
for the design phase, i.e. the requirement analysis. A thorough environmental
analysis is fundamental to the success of an efficient market design. The results
of the environmental analysis of this work are outlined in Section 2.2.4.

2.2.2.2

Design and Implementation

Having derived desiderata and requirements for a domain-specific market design, the next stage covers the conceptual design phase as the central element of
the market engineering process. Analogously to the design of systems and architectures in the computer science domain, markets are meaningfully composed
out of modularized elements in order to achieve a desired market performance
and outcome. The conceptual design constitutes a set of institutional rules in
an abstract manner independent of a concrete implementation (analogue to a
platform- and programming-model-independent design of a software artifact
e.g. in UML [OMG07]). The conceptual design of this work that comprehends
the design of a bidding language to express service offers and requests as well as
a mechanism design with additional extensions is introduced in Section 3 using
an implementation-independent mathematical formalization.
The conceptual design lays the groundwork for the actual implementation of
the market into an information system. This phase is distinguished in the embodiment phase and the implementation phase. In the embodiment phase, the conceptual
design is refined, concretized and extended where required into a more specific
market scheme but still remains implementation-independent. This phase of the
market engineering process is realized in the work at hand in Chapter 4.
The condensed market scheme is subsequently modeled into a formal process
model describing the domain-specific market to be prototypically realized. Section 3.5 introduces the process model for the auction conduction which serves as
procedural blueprint for the subsequent implementation phase.
Finally, in the implementation phase, the prototypical implementation of the
market design is realized based on the results of the previous phases. A prototypical implementation of the work at hand is introduced and briefly described
in Section 3.6.

2.2. MARKETS IN A SERVICE WORLD
2.2.2.3

73

Testing and Evaluation

Having completed the conceptual design phase, the embodiment phase and the
implementation phase, the created artifacts are tested and evaluated with respect
to the specified desiderata and requirements in the environmental analysis. In
the evaluation phase, both, technical and applicability requirements (e.g. support
for service compositions) as well as economic requirements (e.g. incentive compatibility) are evaluated and verified in this phase.
Depending on the aspect subject to evaluation, adequate methods and approaches have to be chosen and selected based on their applicability. Exemplary,
the economic desideratum, which states that the mechanism shall implement a
social choice function that is weakly budget-balanced can be theoretically evaluated using mathematical proofs. Strategic behavior of market participants with
respect to bundling strategies might be too complex to be theoretically investigated but requires an agent-based simulation approach to evaluate such aspects.
The evaluation phase of the work at hand is therefore divided into an analytical
evaluation part in Chapter 5 and an numerical evaluation part in Chapter 6.
Based on the obtained information out of the testing and evaluation phase
about the satisfaction of requirements by the market design and the achievement
of desired outcomes, a final refinement takes place to complete the market for
operative introduction.

2.2.2.4

Introduction

The introduction phase constitutes the final phase of the market engineering process. In this phase, the evaluated and refined electronic market is introduced and
initiates its operation cycle.

2.2.3 Mechanism Design
Mechanism design is a subfield of game theory that pursues the idea of designing institutions that determine decisions as a function of the information that is
known by the individuals in the economy in order to achieve a desired outcome
[Mye88]. Mechanisms serve as a unifying conceptual structure, which allows for
analyzing and comparing economic institutions with respect to their properties
and suitability in order to foster certain outcomes. Analogue to traditional game
theory, mechanism design assumes individuals in an economy to be rational-

74

CHAPTER 2. PRELIMINARIES & RELATED WORK

acting and self-interested, meaning they pursue individual utility maximization.
According to [Par01] the mechanism design problem can be defined as follows:
Definition 2.10 [M ECHANISM D ESIGN ]. The mechanism design problem is to implement an optimal system-wide solution (social choice) to a decentralized optimization
problem with self-interested agents with private information about their preferences for
different outcomes.

2.2.3.1

Social Choice

The main goal of mechanism design is to provide mechanisms that implement a
social choice. A social choice function is an aggregation of the preferences of multiple participants into a single joint decision [NRTV07]. In environments with
decentralized, rationally-acting agents that have private information about their
preferences for different outcomes, the implementation of a social choice function
is necessary to achieve an overall goal due to the absence of complete information.
Given the agent’s type θi ∈ Θi with i ∈ I , the preferences for different outcomes
ρ ∈ R result in the agent’s utility ui (ρ, θi ). A social choice function selects – given
the agents’ types – the optimal outcome ρ∗ .
Definition 2.11 [S OCIAL C HOICE F UNCTION ]. A social choice function ω : Θ1 ×
· · · × Θ I → R selects an optimal outcome ω (θ ) = ρ∗ with ρ∗ ∈ R given the agent’s
types θ = (θ1 , . . . , θ I ). The outcome ρ is decomposable into a choice ωo (θ ) ∈ Ωo and
payments made by each agent ωti (θ ) ∈ Ωt . 47
The outcome of a social choice function is a system-wide solution that can
not be solved directly as the agent’s types are private information to the agents.
Thus, an adequate mechanism is needed that defines a set of game theoretic rules
to implement the solution to the social choice function accounting for rational
and selfish behavior of the agents. The behavior of agents is game theoretically
defined by means of strategies. A strategy describes a complete and contingent
plan that defines the actions an agent will select in every possible state of a game
[Gib92, Par01]. A strategy ψi (θi ) of an agent i is defined as ψi (θi ) ∈ Ψi where θi
denotes the type of agent i and Si all possible strategies depending on its type.
47 Decomposition

into a choice and a payment component is only feasible under the assumption of quasi-linear preferences which is common in game theory.

2.2. MARKETS IN A SERVICE WORLD

75

Based on the concept of a social choice function and agents’ behavior by means
of their strategies, a mechanism is defined as follows:
Definition 2.12 [M ECHANISM ]. A mechanism M = (Ψ1 , . . . , Ψ I , m(·)) defines an
outcome rule m(·) that maps strategies Ψ1 , . . . , Ψ I of agents 1, . . . , I to an outcome ρ ∈ R
such that m : Ψ1 ×, . . . , ×Ψ I → R. The outcome rule m(o (·), t(·)) consists of a choice
or allocation rule o (·) and a payment or transfer rule t(·) that determines the monetary
transfer to the agents. 47
Hence, a mechanism determines the agents’ strategy space and defines a
certain outcome given the chosen strategies. The outcome defines an allocation
(e.g. agent sr gets service v from agent s p ) and the monetary exchange – the transfer – between agents (e.g. agent sr has to transfer an amount x to agent s p ).
Recall that the goal of mechanism design is to implement an optimal systemwide solution (social choice) to a decentralized optimization problem even
though the participants are self-interested and have private information about
their preferences for different outcomes. As agents are assumed to act rational
and therefore to maximize their individually utility, a solution in such a scenario
must be a state where no agent gains by changing its own chosen strategy unilaterally, i.e. an equilibrium in game theoretic terms. The goal of a mechanism is
to implement a social choice function, that is, a mechanism constitutes an equilibrium that yields the same outcome as the optimal solution to the social choice
function for all possible agent preferences.
Definition 2.13 [M ECHANISM I MPLEMENTATION ]. A social choice function ω (θ )
with outcome ρ∗ ∈ R is implemented by a mechanism M = (Ψ1 , . . . , Ψ I , m(·)) if
m(ψ1∗ (θ1 ), . . . , ψ∗I (θ I )) = ρ∗ with (ψ1∗ , . . . , ψ∗I ) ∈ Ψ1 ×, . . . , ×Ψ I and (θ1 , . . . , θ I ) ∈
Θ1 ×, . . . , ×Θ I where strategy profile (ψ1∗ , . . . , ψ∗I ) is an equilibrium strategy given mechanism M.
One can distinguish between direct and indirect mechanisms. In a direct
mechanism, agents submit their messages once to the mechanism and the outcome is computed subsequently. In an indirect mechanism, agents may submit
several messages to the mechanism an receive feedback which is incorporated by
the agents. The focus of the work at hand is restricted to direct mechanisms. A
direct-revelation mechanism is defined as follows:

76

CHAPTER 2. PRELIMINARIES & RELATED WORK

Definition 2.14 [D IRECT-R EVELATION M ECHANISM ]. A direct-revelation mechanism restricts the strategy set for all agents i ∈ I to strategies where agent i reports the
type θ´i = ψi (θi ) based on its actual preferences θi .
The relation between a mechanism, its implementation and the achievement
of the same outcome as a social choice function depicted in Figure 2.19, which is
based on the illustration in [Rei77].

ω (θ )

Type

Outcome

θ

ρ

Mechanism

ψ( θ )

M

m(ψ(θ ))

Figure 2.19
Triangle relation of mechanism implementation and social
choice [Rei77].

In distributed environments with self-interested agents, a system-wide solution to a social choice problem is not solvable directly as rational-acting agents
cannot be assumed to reveal their private information e.g. for the sake of welfare. The agents’ primary objective is to maximize their individual utility, which
mostly collides with a truth-telling strategy. In the absence of complete information regarding agents’ preferences for different outcomes, a mechanism M
must be designed that implements a desired social choice function by means of a
rule set that specifies how to allocate and how to transfer payments. The mechanism implementation induces incentives that constitute an equilibrium strategy
profile which yields the same outcome as the social choice function such that
m(ψ(θ )) = ω (θ ).

2.2. MARKETS IN A SERVICE WORLD
2.2.3.2

77

Properties of Social Choice and Mechanism Implementations

The objective of mechanism design is to implement a social choice function in
equilibrium strategies that yields desired properties. Such properties are often
referred to as mechanism properties. Nevertheless mechanisms do not directly
expose these properties but they implement social choice functions that do. For
the reader’s convenience properties of social choice are also referred to as mechanism properties in the remainder of this thesis. For an extended introduction
to mechanism and social choice properties, the interested reader is referred to
[Par01].
Desideratum 2.1 [A LLOCATIVE E FFICIENCY ]. A social choice function ω (θ ) =
(ωo (θ ), ωt (θ )) is allocative efficient if it maximizes the total utility over all agents. Let
ωo∗ (θ ) ∈ Ωo be an allocative efficient choice, then no alternative choice ώo (θ ) ∈ Ωo yields
a higher utility for all agents such that:
(2.1)

∑ ui (ωo∗ (θ ), θi ) ≥ ∑ ui (ώo (θ ), θi ),
i ∈I

∀ώo (θ ) ∈ Ωo

(AE)

i ∈I

Desideratum 2.2 [(D OMINANT S TRATEGY ) I NCENTIVE C OMPATIBILITY ]. A
mechanism M is incentive compatible if agents report truthful information about their
preferences in equilibrium. A mechanism M is strategy-proof or dominant-strategy
incentive-compatible if each agent i’s best response to any strategy of the other agents
is revealing its true type, i.e. reporting true information about the preferences is a dominant strategy in equilibrium. In other words there is no incentive for agents to announce
untruthful information about their preferences in order to increase their individual utility. Let ψi∗ (θi ) = θi be the truth-revelation strategy for agent i. For a strategy-proof
mechanism M it is required that
(2.2)

ui (m(ψi∗ (θi ), ψ−i (θ−i )), θi ) ≥ ui (m(ψ́i (θi ), ψ−i (θ−i )), θi ),

∀ψ́i ∈ Ψi \ {ψi∗ },

∀ψ−i ∈ Ψ−i ,

∀i ∈ I

which means that the truth-revelation strategy is a dominant strategy for all agents. Furthermore it is required that the strategy profile
(2.3)

ψ∗ = (ψ1∗ (θ1 ), . . . , ψ∗I (θ I ))

is an equilibrium given mechanism M.

(DSIC)

78

CHAPTER 2. PRELIMINARIES & RELATED WORK

Desideratum 2.3 [I NDIVIDUAL R ATIONALITY ]. A mechanism M is individual rational if it implements a social choice function ω (θ ) = (ωo (θ ), ωt (θ )) = ρ that guarantees that agents are not worse-off by participating. Let ui (ρ, θi ) be the utility of agent i
in case of participation and ūi (θi ) the utility of its outside option, i.e. its utility if agent i
does not participate.
(2.4)

ui (ρ, θi ) ≥ ūi (θi ),

∀i ∈ I

(IR)

Assuming a mechanism where an agent can withdraw once it knows the outcome ex-post
is individual rational if participation makes the agent not worse-off compared to the outside option of not participating for all possible agent types in the system. In mechanisms
where agents are not able to observe the outcome, meaning the decision to participate has
to be done ex-ante, the concept of interim individual rationality is introduced, which
is a weaker property from an ex-ante perspective.
(2.5)

E(ui (ρ, θi )) ≥ E(ūi (θi )),

∀i ∈ I

(IIR)

The expected utility E(ui (ρ, θi )) for agent i from participation is not worse then its expected utility E(ūi (θi )) from not participating.
Desideratum 2.4 [B UDGET B ALANCE ]. A social choice function ω (θ ) =
(ωo (θ ), ωt (θ )) is strong budget-balanced if all payments made by the agents are distributed among all agents. This means that there are no outside payments necessary to
realize transfers according to the outcome of the social choice function.
(2.6)

∑ ωti ( θ ) = 0

(BB)

i ∈I

There are no net transfers neither into the system nor out of the system. A weaker version of budget balance is if there are transfers out of the system but not into the system,
i.e. weak budget balance.
(2.7)

∑ ωti ( θ ) ≥ 0

(WBB)

i ∈I

Although all of these valuable properties of social choice and mechanism
implementations are desired from an economical perspective, they cannot be
achieved at the same time due to impossibilities, which are presented in detail
in Section 2.2.3.4.

2.2. MARKETS IN A SERVICE WORLD
2.2.3.3

79

Possibility Results

Maybe the most important possibility result in mechanism design is the revelation principle as it implies that it is sufficient to restrict to direct incentive compatible mechanisms. The principle is defined as follows:
Definition 2.15 [R EVELATION P RINCIPLE ]. Any mechanism M that implements a
social choice function ω (·) in dominant strategies48 can also be implemented by an incentive compatible direct-revelation mechanism that implements the same social choice
function ω (·) in dominant strategies.
The intuition behind the revelation principle can be illustrated as follows: Assuming the agents’ strategy profile ψ∗ = (ψ1∗ , . . . , ψ∗I ) in equilibrium in a mechanism M leads to an outcome ρ(ψ∗ ). Now, the behavior of the agents is simulated
by a mechanism Ḿ called a simulator which computes the optimal strategies of
the agents based on their reported preferences. Hence, for each agent i ∈ I it is
a dominant strategy to report its type truthfully to the mechanism Ḿ. Consequently the simulator Ḿ implements the same social choice function as M.
To illustrate the idea of the revelation principle the following example
presents a general mechanism and an equivalent incentive compatible directrevelation mechanism that leads to the same outcome. The example is a slightly
changed variant of an example in [Mye88] with an extensive analysis.
Example 2.6 [I NCENTIVE C OMPATIBLE D IRECT-R EVELATION M ECHANISM ].
Consider a game where two agents i and −i have private valuations vi and v−i for a
good g. Both agents separately put amounts bi and b−i in two different envelops. The
agent that reports the higher amount gets the good and the other one gets both envelopes.
Presented game is symmetric and therefore both agents try to maximize the same expected
utility. Without loss of generality, agent i’s expected utility is analyzed.
(2.8)

Ei (·) = P(bi > b−i ) [vi − bi ] + P(bi < b−i ) [bi + b−i ]

Two cases must be considered:

48 Note

that the first version of the revelation principle in [Gib73] is restricted to mechanisms
that implement a social choice function in dominant strategies. In [Mye82] the principle is extended
to the general case for all equilibrium concepts e.g. Bayesian-Nash equilibria.

80

CHAPTER 2. PRELIMINARIES & RELATED WORK
1. Getting the good g yields a higher utility for agent i then getting both envelopes
such that
(2.9)

( v i − bi ) > ( bi + b − i )

(2.10)

vi − 2bi > b−i

Consequently agent i wants to maximize the probability of winning the good.
P(bi > b−i ) is maximized by reporting an amount bi = vi − 2bi which leads to
the strategy of reporting an amount bi = 31 vi .
2. Getting the good g yields a lower utility for agent i then getting both envelopes
such that
(2.11)

( v i − bi ) < ( bi + b − i )

(2.12)

vi − 2bi < b−i

Consequently agent i wants to maximize the probability of getting both envelopes
and loosing the good. P(bi < b−i ) is maximized by reporting an amount bi =
vi − 2bi which leads to strategy of reporting an amount bi = 31 vi .
The strategy of announcing an amount bi∗ = 13 vi is the best response of agent i not knowing agent −i’s strategy. As the game is symmetric this argumentation also holds for agent
−i. Consequently, the strategy b∗ = 31 v constitutes an equilibrium.
Without loss of generality let agent i be the agent that wins the good g such that
bi > b−i . Thus, the outcome of the game based on the agents’ equilibrium strategies
evolves as follows:
(2.13)
(2.14)

2
v
3 i
1
1
u−i (·) =
v −i + vi
3
3
ui (·) =

According to the revelation principle (Definition 2.15) an equivalent incentive compatible
direct-revelation mechanism can be designed that yields the same outcome:
The mechanism allocates the good g to the agent that reports the higher amount and
charges one-third of that amount. The other agent that does not receive the good gets onethird of both reported amounts. Analogously to the previous game, the expected utility of
agent i is analyzed.
(2.15)




1
1
1
Ei (·) = P(bi > b−i ) vi − bi + P(bi < b−i ) bi + b−i
3
3
3


2.2. MARKETS IN A SERVICE WORLD

81

Two cases must be considered:
1. Getting the good g yields a higher utility for agent i then getting one-third of both
reported amounts such that
(2.16)
(2.17)

1
1
1
( v i − bi ) > ( bi + b − i )
3
3
3
3vi − 2bi > b−i

Consequently agent i wants to maximize the probability of winning the good.
P(bi > b−i ) is maximized by reporting an amount bi = 3vi − 2bi which leads to
the truth-telling strategy bi = vi .
2. Getting the good g yields a lower utility for agent i then getting one-third of both
reported amounts such that
(2.18)
(2.19)

1
1
1
( v i − bi ) < ( bi + b − i )
3
3
3
3vi − 2bi < b−i

Consequently agent i wants to maximize the probability of getting both envelopes
and loosing the good. P(bi < b−i ) is maximized by reporting an amount bi =
3vi − 2bi which also leads to the truth-telling strategy bi = vi .
Without loss of generality let agent i be the agent that wins the good g such that bi > b−i .
Thus, the outcome of the game based on the agents’ equilibrium truth-telling strategies
evolves as follows:
(2.20)
(2.21)

2
v
3 i
1
1
u−i (·) =
v −i + vi
3
3
ui (·) =

The example at hand illustrates the idea of the revelation principle by showing
that there exists a direct-revelation mechanism that yields the same outcome as
the general mechanism in a truth-telling equilibrium, i.e its incentive compatible.
Note that the example demonstrates the application of the more general revelation principle according to [Mye82] that extends results in [Gib73] – that restrict
the revelation principle to dominant strategy equilibria – to the general case for
multiple equilibrium concepts e.g. Bayesian-Nash equilibria.
Summarizing, with the results of the revelation principle, impossibility results
can be proven over the space of direct-revelation mechanisms, and possibility
results can be constructed over the space of direct-revelation mechanisms.

82

CHAPTER 2. PRELIMINARIES & RELATED WORK

Maybe the most prominent family of direct-revelation mechanisms are the
Vickrey-Clarke-Groves (VCG) mechanisms [Vic61], [Cla71] and [Gro73]. VGC
mechanisms belong to the class of Groves mechanisms and are individual rational, allocatively-efficient and strategy-proof direct-revelation mechanisms. For a detailed analysis of the family of VCG mechanisms and their properties, the interested reader should refer to [Par01].

2.2.3.4

Impossibility Results

Despite of possibility results such as the revelation principle, there are important
impossibility results that have strong limitations to design goals that can be simultaneously pursued. In fact, it is impossible to achieve certain combinations of
design desiderata as outlined in the previous section. Among the most prominent
are the following theorems:
Theorem 2.1 [H URWICZ (G REEN -L AFFONT ) I MPOSSIBILITY T HEOREM ]. There
is no double-sided mechanism that is at the same time allocative efficient, budget-balanced,
and truthful in settings with quasi-linear preferences [GL78, Wal80, HW90].
The Theorem 2.1 restricts its proposition and applicability to dominantstrategy equilibria, whereas the following theorem by Myerson and Satterthwaite
makes a more generic statement:
Theorem 2.2 [M YERSON -S ATTERTHWAITE I MPOSSIBILITY T HEOREM ]. There is
no double-sided mechanism that is at the same time allocative efficient, budget-balanced,
Bayesian-Nash incentive compatible, and (interim) individually rationality, even in settings with quasi-linear preferences [MS83].
Theorem 2.2 extends the former theorem also to situations where reporting
ones true type is a Bayesian-Nash equilibrium where participants intent to maximize their expected utility instead of their ex-post utility. By extending their
proposition, Myerson and Satterthwaite add the condition that the mechanism
must be individual rational.
In summary, the Myerson-Satterthwaite Impossibility Theorem implies that
at most two desiderata out of allocation efficiency, individual rationality, and
budget balance can be achieved when designing truthful mechanisms in settings
where agents are assumed to have quasi-linear preferences.

2.2. MARKETS IN A SERVICE WORLD
2.2.3.5

83

Algorithmic Mechanism Design

Algorithmic mechanism design – firstly introduced by [NR01] – broadens the economic focus by considering problems that are inherent in the mechanism design
problem from a computer science and algorithmic perspective such as complexity and computational feasibility of computing an optimal system-wide solution.
Internet protocols for example are designed under the implicit assumption that
each participant within the system behaves according to a deterministic procedure or program. Nevertheless, this assumption does not hold in environments
such as the Web as participants and owner of computer systems and applications
are self-interested and act according to their individual objectives.
Many challenges in computer science involve selfish behavior of decentralized participants and thus, require adequate mechanisms to implement an efficient solution such us internet routing, scheduling and task allocation, resource
allocation, and service composition [NRTV07]. In such scenarios, agents cannot
be assumed to follow a deterministic algorithm but try to maximize their own
utility which might collude with other objectives and a system-wide solution.
Especially the coordination of service composition requires a mechanism design that accounts for selfish behavior of distributed service providers by implementing the right incentives to jointly achieve a common goal that serves the
objectives and well-being of the overall system. Despite of such economic challenges, this scenario puts further technical requirements upon a potential mechanism design in order to be applicable for the coordination of composite service
creation. Hence, this broadens the view of mechanism design regarding the field
of algorithms and information systems design [DJP03].

2.2.4 Environmental Analysis and Related Work
This section outlines requirements upon a mechanism in order to be applicable
in the context of coordination in service value networks from an economic and
technical perspective (Section 2.2.4.1). Based on the requirement analysis, Section 2.2.4.2 introduces and describes related work and critically examines their
shortcomings in the context of stated requirements and the approach at hand.

2.2.4.1

Requirements

There is a number of requirements a mechanism must and partly should satisfy
in order to be applicable in the context of service composition in service value

84

CHAPTER 2. PRELIMINARIES & RELATED WORK

networks from an economic and technical perspective. Requirements upon a
mechanism are basically dividable into economic requirements and applicability requirements. Economic requirements are explained in detail in Section 2.2.3.5 and
are therefore only outlined briefly at this point:
Requirement 1 [A LLOCATIVE E FFICIENCY ]. A mechanism is said to be allocative
efficient if it always determines the outcome that maximizes the overall utility across
all participants (consumer and provider surplus), i.e. it always maximizes the system’s
welfare (cp. Desideratum 2.1).
Requirement 2 [I NCENTIVE C OMPATIBILITY ]. A mechanism is said to be (dominant
strategy) incentive compatible or truthful if the truth-telling strategy is an equilibrium
in weakly dominant strategies (cp. Desideratum 2.2).
Incentive compatibility is an important requirement as it functions a precondition for the allocative efficiency requirement. In distributed environments incentive compatibility enables the transition from incomplete (private) information
to the situation in which participants reveal their true types voluntarily. This reported information is necessary for a welfare-maximizing solution to be always
computable as stated in Requirement 1. Furthermore, truthfulness tremendously
reduces the complexity of the strategy space of participants. Under the presence
of a weakly dominant strategy there is no need to reason about the other participants’ preferences.
Requirement 3 [I NDIVIDUAL R ATIONALITY ]. A mechanism implements a social
choice that is said to provide the property of individual rationality if agents cannot suffer
a loss in utility from participating in the mechanism, i.e. the option to participate in the
mechanism is not worth than the outside option.
Requirement 4 [B UDGET B ALANCE ]. A mechanism is said to be (weakly) budgetbalanced if its transfers do not require external subsidization by outside payments, i.e. the
requester’s willingness to pay covers payments transferred to providers (cp. Desideratum
2.4).
Budget balance and individual rationality are crucial for a sustainable implementation of a mechanism with respect to the underlying business model. If
budget balance is not met, the mechanism must continuously be subsidized by
outside payments which is not feasible from the strategic perspective of e.g. a
service platform provider. Additionally if individual rationality is not me by the

2.2. MARKETS IN A SERVICE WORLD

85

mechanism, agents will not voluntarily participate in the mechanism as they face
the risk of being worse off compared to their outside option.
For a mechanism in order to be applicable in the context of complex services
in service value networks from a technical and domain-specific perspective, the
following requirements have to be met:
Requirement 5 [C OMPUTATIONAL T RACTABILITY ]. A mechanism is said to be
computational tractable if it computes an allocation and corresponding prices in polynomial runtime in the size of its inputs, i.e. e.g. the number of service offers and their
feasible compositions into a complex service.
Computational tractability is important for mechanisms that need to perform
in online systems, i.e. they need to compute an allocation and prices at runtime
within a feasible time frame. Especially in the context of service value networks,
the number of feasible paths through the network – that is, the number of feasible
complex service instances – increases rapidly (exponentially) as the number of
service providers and candidate pools increases49 .
Requirement 6 [S ERVICE C OMPOSITION S UPPORT ]. Service compositions, in contrary to service bundles, only generate value for the requester in the right order of their
components. Thus, a mechanism in a broader sense is said to support service composition
if its bidding language and allocation function accounts for the well-defined sequence of
service components in order to form a feasible complex or composite service.
Support for service composition is a rare requirement in the context of combinatorial mechanisms. Although most approaches in this area provide rich bidding languages, they only support bundles in an economic sense which ignores
the order of the entities the bundle consists of50 .
Requirement 7 [Q O S-S ENSITIVITY ]. A mechanism in a broader sense is said to be
QoS-sensitive if it accounts for complex QoS characteristics by providing an adequate
bidding language and allocation function that is implemented by a corresponding allocation algorithm.
49 Based

on the service value network model in Section 2.1.4, the number of feasible paths
depends on the number of candidate pools and service offers per candidate pool. Assuming an
 |V \{v ,v }| K
s f
equal number of service offers per pool, the number of paths is
, with K denotes the
K
number of candidate pools.
50 E.g. its not possible to express a preference like ( A, B ) ≻ ( B, A )

86

CHAPTER 2. PRELIMINARIES & RELATED WORK

Requirement 8 [S ERVICE L EVEL E NFORCEMENT ]. A mechanism in a broader sense
is said to provide service level enforcement if it incorporates information about the fulfillment of QoS aspects. Based on this information, the mechanism’s transfer function
provides means for rewarding or penalizing agents.
Requirements 6 and 8 together are important to provide a sustainable support
for the coordination and trade of complex services as it enables differentiation in
quality and a trustworthy environment for service contracts.
2.2.4.2

Related Work

This section outlines research approaches that are closely related to the work
at hand and highlights research gaps and shortcomings that are addressed and
partly solved by this approach.
A double-sided market mechanism for trading Grid resources is presented in
[Sto09]. The computation of the allocation is based on a greedy heuristic which is
scalable and performs well also in large-scale settings while minimizing efficiency
loses compared to an optimal solution that is computational intractable. In the
work, two pricing schemes are presented. The first, a proportional critical value
pricing scheme that successfully limits strategic behavior of market participants
on the expense of computational costs. The second pricing scheme, k-pricing
is highly scalable while sacrificing incentive compatibility to a certain degree.
Nevertheless, only low-level resource-oriented services (cp. the bottom layer in
the service decomposition model in Section 2.1.2) are tradable as the mechanism
and the bidding language do not support compositions of services, complex QoS
characteristics and their enforcement.
Allowing the trade of service bundles, MACE (Multi-Attribute Combinatorial Exchange [Sch07]) and the Bellagio System [ACSV04] provide mechanism for
the trade of infrastructure resources. Resource service are specified by rudimentary quality attributes and can be requested and provisioned as bundled services.
Although the trade of service bundles is supported, their is no support for service compositions as the bidding language is only capable of capturing bundle
specifications independent of the sequence of entailed service components. Furthermore, preferences for service attributes can only be specified by means of
rudimentary operations such as AND, OR, and XOR whereas only simple quality attributes such as response time are supported. From an economic perspective, neither mechanism implements truthfulness with respect to resource prices
which allows for strategic behavior of participants that is only partly limited by

2.2. MARKETS IN A SERVICE WORLD

87

the pricing scheme. From a technical perspective, the winner determination problem in both mechanisms is computational intractable which does not allow for
their application in large-scale online settings.
In [LS06], the MACE exchange is extended by means of semantic concepts and
technologies. A combinatorial double auction is presented that is continuously
cleared. Corresponding bidding language supports the trade of service bundles
but is not capable of capturing information about sequential compositions. Services are specified by means of semantically describable quality attributes which
allows for highly differentiated service offers with respect to their QoS characteristics. Nevertheless, from an economic perspective, the auction mechanism
does not provide incentives for truth-revelation of private valuations and QoS
attributes of traded services. Furthermore, in settings which require the timely
allocation of services, the auction mechanism is not applicable as it exposes exponential run-time behavior.
Focusing on mechanisms for allocation and pricing of service compositions
that expose a well-defined control sequence, a combinatorial auction for QoSaware dynamic web services composition is proposed in [MNM+ 07]. Their composition model heavily relies on the work in [ZBD+ 03] where feasible service
compositions are predefined based on a statechart graph. Based on this model,
a QoS-sensitive combinatorial auction mechanism is proposed which allocates
the composition of services which yields the highest quality level based on the
requesters preferences subject to budget constraints which results in a computational intractable problem. From an economic perspective, the mechanism’s
design does not implement incentives for truth-revelation of QoS attributes and
private valuations. The mechanism neither verifies the services’ performance expost nor incorporates penalties at the current state of the work.
In summary, as comprised in Table 2.3, a lot of work has been done with respect to designing suitable mechanisms for allocation and pricing of services in
different levels of granularity (utility, elementary and complex services). Nevertheless, there still exist various research gaps especially in the context of incorporating feasibility of service compositions in the allocation problem as well as
QoS-sensitivity and adequate ex-post verification mechanisms to impose penalties for non-performance.

88

CHAPTER 2. PRELIMINARIES & RELATED WORK

Approach

(R 8) Service Level Enforcement

(R 7) QoS-Sensitivity

(R 6) Service Composition Support

(R 5) Computational Tractability

(R 4) Budget Balance

Economic Requirements

Applicability Requirements

Stößer 2009

#

G
#

#

#

#

Schnizler 2007

#

#

#

#

G
#

#

Lamparter et al. 2006

#

#

#

#

Mohabey et al. 2007

#

#

#

This Work
This Work (extended)

2.3

(R 3) Individual Rationality

(R 2) Incentive Compatibility

(R 1) Allocative Efficiency

Table 2.3: Requirements satisfaction degree of related approaches ( = fully satisfied, G
# = partly satisfied, # = not satisfied).

#
#

#
G
#

G
#

Research Methods

The primary goal of the work at hand is not to analyze existing mechanisms but
to design novel mechanisms that expose desired properties and induce desired
behavior of participants in a particular domain. As pointed out in [Rot02], an “engineering approach” is required for designing suitable market mechanisms. This
work is founded on the approach of mechanism design [Mye88, NR01] which is
introduced in detail in Section 2.2.3.5. In order to evaluate the properties and the
behavior of participants in the developed auction mechanism, the complex service auction, this work heavily relies on two methodologies: theoretical analysis
and simulations which are briefly introduced in the remainder of this section.

2.3. RESEARCH METHODS

89

2.3.1 Theoretical Analysis
To study the main properties of the auction mechanism, concepts and methods
from game theory are employed. This implies to make strong assumption about
the market participants with respect to the information about other participants
and the utility functions [MCWG95]. There exist multiple solution concepts in
game theory such as Nash equilibria and dominant strategy equilibria. Theoretical analysis provides strong results. Nevertheless, in order to apply analytical
game theoretic evaluations, models usually rely on strong assumptions that do
not necessarily reflect real world settings.

2.3.2 Simulations
Evaluating certain mechanism properties or behavior of participants in settings
with a multitude of variable factors, a theoretical analysis is not applicable in
most of the cases due to the high complexity of the system. As a remedy, numerical simulations provide a useful tool to analyze particular properties of a mechanism by means of randomly generated problem sets, i.e. the variable factors are
randomly generated for multiple simulation runs. Numerical simulations can
provide insights into the general problem structure, performance aspects of the
algorithm that solves the winner determination problem, mechanism properties
and strategic behavior of participants.
Focusing on more complex settings with participants that face large strategy
spaces which precludes theoretical solutions, the methodology of agent-based
simulations has proven to be promising [Bon02]. Strategic behavior is simulated by means of collections of computerized agents that implement the ability
to learn their surroundings and the space of feasible solutions. In contrary to a
traditional game theoretic analysis, agent-based simulations provide means for
the evaluation of rare strategies which are more complex and occur in special
domains. Nevertheless, it is crucial to design reasonable strategies and learning
behavior and incorporate them into software agents. However, a lot of work has
been done in the area of agent-based simulations and a whole set of different
strategies has been shown to work well in many settings [Phe08].

Part II
Design & Implementation

Chapter 3
Complex Service Auction (CSA)
I believe that in the future we may see much more auctioning of services [...]. Services
are particularly attractive for auctions because they are in relatively fixed supply –
unlike durable goods, one cannot store surpluses or draw down inventory in order to
meet fluctuating demand.
[LR00]
he fundamental paradigm shift from vertical integration to horizontal specialization and the coherent transformation of traditional value chains to
highly dynamic value networks is predominantly observable in the service sector. At the same time, customers’ demand for sophisticated, customized services has considerably been rising in recent years. Open standards and serviceoriented architectures have emerged as important building blocks for innovative
service value networks tying together the competencies of specialized contributors. Thus, by modularization, complex services are increasingly able to be
composed in a “plug-and-play”-manner [VvHPP05]. This novel form of value
creation in loosely-coupled service ecosystems is unique from a coordination and
incentive engineering perspective as it exposes cooperative and non-cooperative
aspects. Participants in such service value networks are both, self-interested –
i.e. they try to maximize their individual utility – but also fully bound to the
success of the whole system.

T

It is a well-known result from Market Engineering (cp. Section 2.2.2) that there
is no general mechanism that fits any possible setting [WHN03]. An adequate
mechanism depends amongst others on the properties of the trading objects –
which are service components and complex services in the work at hand – and the
goals of the designer (e.g. welfare vs. revenue maximization). Having analyzed

94

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

the characteristics of services in general in Section 2.1.1.2, and special aspects of
software services in Section 2.1.3 as well as their composition into complex services in service value networks in Section 2.1.2 and 2.1.4, the set of requirements
and desiderata from a technical and an economic perspective upon a suitable
mechanism were outlined in Section 2.2.4.
Section 3 focuses on the design of an auction mechanism – the Complex Service Auction (CSA) – that enables based on service offers and requests the allocation of multidimensional service components which are sequentially composed into feasible complex service instances. An abstract model is introduced
that comprehends a bidding language to describe information objects that are exchanged during the auction process. Additionally the model provides means
to formalize service value networks in a graph-based structure. The mechanism itself is capable of allocating service components and determining dynamic
prices and corresponding QoS characteristics of complex services. Furthermore,
in Chapter 4 extensions to the complex service auction are developed in order
to meet the applicability requirements such as QoS-sensitivity and service level
enforcement and to achieve budget balance.
For the remainder of this section it is useful to refer to the design framework
for market mechanisms depicted in Figure 3.1. Analogue to the structure of this
section, there are three fundamental components in the design of a market mechanism [DVVfMSiES03]: the bidding language (cp. Section 3.2), that provides means
for formalizing information objects and all their necessary parts for the requester
and the provider side that are exchanged during the conduction of e.g. the complex service auction; the allocation function (cp. Section 3.3.1) which determines
which trading object(s) are allocated to which participant(s); and the transfer function (cp. Section 3.3.2) that determines based on the allocation the monetary transfers that have to be realized among the participants. Focusing on the realization
of a mechanism implementation, the concrete allocation algorithm that computes
the allocation function is a central design issue. In this context, design desiderata such as computational tractability and allocative efficiency strongly depend
on the design of the allocation algorithm. Counteracting complexity, heuristic algorithms might restore computational tractability by sacrificing optimality to a
certain extent [Sto09]. In contrary, exact algorithms enable the computation of an
allocative efficient outcome (assuming incentive compatibility) but might result
in exponential run-time [Sch07].
Based on the impossibility results as described in Section 2.2.3.4, there is an
inherent trade-off between design desiderata (cp. Section 2.2.4.1) that has to be
considered when constructing the mechanism’s components.

3.1. SERVICE VALUE NETWORK MODEL

95

Mechanism

Bidding Language

Allocation Function

Transfer Function

Allocation Algorithm

Heuristic

Exact

Figure 3.1
Framework for the design of mechanisms.

For the reader’s convenience, the formal notation that is used throughout this
section, is outlined in Section A.1 in tabular form.

3.1 Service Value Network Model
Recall that Section 2.1.4 is concerned with an initial description of service value
networks, their main characteristics and the various roles involved in value creation. In addition to this first outline, this section focuses on providing a mathematical model of a service value network that captures the presented aspects in a
comprehensive technical manner.
A service value network is described by means of a simplified statechart
model [HN96] and is aligned with the representation in [ZBD+ 03] as depicted
in Figure 3.2. Statecharts have proven to be the preferred choice for specifying
process models as they expose well-defined semantics and they provide flow
constructs offered by prominent process modeling languages (e.g. WS-BPEL) and
therefore allow for simple serialization in standardized formalisms.
Hence, a service value network is represented by a k-partite, directed and
acyclic graph G = (V, E). Each partition Y1 , . . . , YK of the graph represents a candidate pool that entails service offers that provide the same (business) functional-

96

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)
t5

t1
t4
t2

t3

t6

Caption
State

AND-State

Transition

Initial State

Final State

Figure 3.2
Statechart formalization [HN96, ZBD+ 03].
ity. The set of N nodes V = {v1 , . . . , v N } represents the set of service offers1 with
u, v, i, j being arbitrary service offers. There are two designated nodes vs and v f
that stand for source and sink in the network and are not part of any partition
Y = (Y1 , . . . , YK ), hence V = Y1 ∪ · · · ∪ YK ∪ {vs , v f }. Services are offered by a set of
Q service providers S = {s1 , . . . , sQ } with s being an arbitrary service provider. The
ownership information σ : S → P (V \ {vs , v f }) that reveals which service provider
owns which services within the network is public knowledge2 . The set of edges
E = {eij |i, j ∈ V } denotes technically feasible service composition such that eij
represents an interoperable connection of service i ∈ V with service j ∈ V 3 . If two
services are not interoperable at all, they are not connected within the network.
A service configuration A j of service offer j ∈ V is fully characterized by a vector
of attributes A j = ( a1j , . . . , a Lj ) where alj is an attribute value of attribute type l ∈ L
of service offer j’s configuration. Attribute types can be either functional attribute
types or non-functional attribute types (e.g. availability or privacy). A service’s
configuration represents the quality level provided and differentiates its offering
from other services. According to [Lam07], a service configuration can be defined
as follows:
Definition 3.1 [S ERVICE C ONFIGURATION ]. A service configuration A j of a service
j ∈ V selects a value alj for each attribute type l ∈ L of a service and thereby unambiguously defines all relevant service characteristics. The choice of configuration might affect
the functional and non-functional aspects of a service and is a major determinant of the
price.
1 For

the reader’s convenience the terms service offer, service and node are used interchangeably
: V \ {vs , v f } → S maps service offers to single service
providers that own that particular service
3 For the reader’s convenience the notion e is equivalent to e
vi v j representing an interoperable
ij
connection of service i ∈ V with service j ∈ V.
2 The reverse ownership information σ −1

3.1. SERVICE VALUE NETWORK MODEL

97

Furthermore let cij denote the internal variable costs that the service provider
that owns service j has to bear for that service being interoperable with service
i and for the execution of service j as a successor of service i. The representation of a detailed cost structure of service providers is intentionally omitted
which serves a better understanding and does not restrict the generalization of
the model. It is assumed that the representation of internal variable costs reflects the service providers’ valuations for their service offers being executed in
different composition-related contexts.
Example 3.1 [C ONTEXT-D EPENDENT C OST S TRUCTURES ]. In order to illustrate
the idea of context-dependent cost structures of service providers refer to Figure 2.1. For
simplification, the complex service is reduced to the first two business transactions, data
verification and the transaction processing. Figure 3.3 shows the service value network with service offers and corresponding costs dependent on the preceeding service.
Data verification can be performed by either Strike Iron (s A ) and its service offer A or
CYDNE (s B ) offering service B. The execution of the actual monetary transaction is done
by Net Billing (sC ) offering service C.
Caption
Data
Verification
Service

Transaction
Processing
Service

v

Service Offer
Composition
Relation

Strike
Iron (A)

accA = false

Source Node

c AC = 0.8
cij

Internal Costs

accj

Credit Check
A"ribute

Net
Billing (C)

CDYNE (B)

aBcc = true

c BC = 0.5

Figure 3.3
Context-dependent cost structures of service providers.

98

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

A mandatory step of the overall payment processing service is the credit assessment.
As a precondition, a transaction processing service has to check if the customer is credit
worthy in order to charge the corresponding account. The credit assessment has to be
performed at a central authority (e.g. Equifax, Experian or Trans Union) and generates
variable costs each time it is executed. In the concrete scenario, Net Billing has to bear
higher costs of 0.8 in case it is executed as a successor of the Sales Force data verification
service as it does not provide a credit check in advance. In contrary, the service offered by
CYDNE is capable of performing a credit check, which results in lower internal costs for
Net Billing of 0.5.
As already illustrated in Section 2.1.2.3 and Section 2.1.4, the instantiation of
a complex service is represented by a path from source to sink within the service
value network. Let F denote the set of all feasible paths from source to sink. Every
f ∈ F with f ⊂ E represents a possible instantiation of the complex service4 .
Definition 3.2 [S ERVICE VALUE N ETWORK M ODEL ]. A service value network
model is an acyclic, k-partite and directed graph such that
(3.1)

G = (V, E)

with the set of nodes V representing service offers and the set of edges E that denotes
technically feasible service compositions. G contains two designated nodes vs and v f
representing source and sink such that every feasible path f ∈ F connecting both nodes is
a possible instantiation of the complex service.
For illustration purpose, Figure 3.4 shows the model of a service value network with service offers V = {v1 , . . . , v4 } ∪ {vs , v f } and service providers S =
{s1 , . . . , s3 }. Every feasible path f ∈ F connecting source node vs and sink node v f
represents a possible realization of the overall complex service.

3.2

Bidding Language

As a formalization of information objects which are exchanged during auction
conduction a bidding language is introduced that is based on bidding languages
4 Focusing

on the presence or absence of a particular service i ∈ V, F−i represents the set of
all feasible paths from source to sink in the reduced graph G−i without node i and without all its
incoming and outgoing edges. In contrary, let Fi be the subset of all feasible paths from source to
sink that explicitly entail node i.

3.2. BIDDING LANGUAGE

s1

99

s2

Caption

s3
s

Service Provider
Ownership
Relation

v1

cs1

1
1

a

v2

c12
a

…

L
… a1

1
2

v

Service Offer

…

L
… a2

Composition
Relation

c14

vf

vs

v3

cs 3
a

1
3

a

… a

L
3

Source Node

vf

Sink Node

v4

c34

…

vs

1
4

Candidate Pool

…

… a

L
4

Y

Complex Service

Y2

Y1

Figure 3.4
Service value network model.
for products with multiple attributes as discussed in [EWL06]. The formalization is aligned to multiattribute auction theory as presented in [PK02, RL05] and
assures compliance with the WS-Agreement specification [ACD+ 04] in order to
enable realization in decentralized environments such as the Web.

3.2.1 Scoring Function
A complex service – represented by a path f – is characterized by a configuration A f . The importance of certain attributes and prices of a requested complex
service is idiosyncratic and depends on the preferences of the requester. The requesters’ preferences are represented by a scoring function S of the form:

(3.2)

L

S(A f ) =

∑ λl kAlf k

l =1

!

The scoring function S represents the requesters’ preferences for a configuration A f of the complex service represented by f analog to the definition of scoring
rules in [AC08]. It maps the configuration of a complex service to a value repre-

100

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

senting the requester’s score such that S : A → [0; 1]5 . The scoring function is
determined by a vector of weights Λ = (λ1 , . . . , λ L ) with ∑lL=1 λl = 1 that defines
the requester’s preferences of each attribute type l ∈ L. The configuration A f of
the complex service f is constituted by the aggregation of all attribute values of
contributing services with incoming edges on the path f such that

A f = (A1f , . . . , A Lf ) with Alf =

(3.3)

M

alj

eij ∈ f

The aggregation operation
for attribute values depends on their type
(e.g. the attribute type encryption is aggregated using a Boolean AND operator whereas response time is aggregated by a sum operator). Table 3.1 shows
different types of aggregation functions for sample multiple attribute types.
L

Table 3.1: Aggregation operations for different attribute types.
Attribute Type

Aggregation

l∈L

L

eij ∈ f | j6=v f

alj

Response Time (rt)

∑eij ∈ f | j6=v f art
j

Encryption Type (et)

V

eij ∈ f | j6=v f

aet
j

Error Rate (er)

maxeij ∈ f | j6=v f aer
j

Throughput (tp)

mineij ∈ f | j6=v f a j

Probability of Default (pd)

1 − ∏eij ∈ f | j6=v f (1 − a j )

tp
pd

The list of aggregation operations in Table 3.1 only shows a rather trivial subset of possible and practical aggregation operations for different quality aspects of
services and is not exhaustive. The bidding language also supports rich semantic
approaches towards more complex aggregation operations in order to deal with
various non-functional service attributes. For example, services are capable of
different types of encryption algorithms and a requester prefers asymmetric ciphers, semantic subsumption can be used to evaluate if a complex service fulfils
the requester’s requirements and therefore to determine the score. Bidding, ag5 Note

that the scoring function is only capable of expressing soft policies and no goal policies
(cp. [Lam07]). Nevertheless, in Section 4.3 an extension is introduced which enables the specification of more complex QoS characteristics and corresponding goal policies.

3.2. BIDDING LANGUAGE

101

gregation and management of complex QoS aspects within the CSA is presented
in detail in Section 4.3.
To assure comparability of attribute values from different attribute types
and to express requesters’ preferences more sophisticated, the aggregated attribute values are normalized on an interval [0; 1] using preference functions with
lower (bottom) and upper (top) boundaries. Boundaries are defined by a vector
Γ = ((γ1B , γ1T ), . . . , (γBL , γTL )) for each attribute type l with γlB 6= γTl ∀l ∈ L. γlB represents the attribute value boundary that results in a zero utility for the requester
with respect to attribute type l (bottom boundary). γTl denotes the attribute value
boundary for type l ∈ L that just leads to a maximum utility of 1 for the requester
(top boundary). The mapping of attribute values is specified by the following
piecewise defined function.

(3.4)




gl (Alf )






1




0
l
kA f k =


hl (Alf )





1





0

,if γTl > γlB ∧ γlB < Alf < γTl
,if γTl > γlB ∧ Alf ≥ γTl
,if γTl > γlB ∧ Alf ≤ γlB
,if γTl < γlB ∧ γTl < Alf < γlB
,if γTl < γlB ∧ Alf ≤ γTl
,if γTl < γlB ∧ Alf ≥ γlB

The function g : A → [0; 1] is a monotonically increasing utility function such
that gl represents the requesters’ utility function for attribute type l. An increasing utility function gl indicates that the requesters utility increases with higher
values of attribute type l. Attribute types such as response time result in a loss of
utility the higher the attribute value. The preference for these types of attributes is
expressed by a monotonically decreasing utility function such that h : A → [0; 1].
Example 3.2 [S CORING F UNCTION C OMPUTATION ]. This example illustrates how
different attribute types are aggregated along a path of composed service offers in service
value networks. It furthermore shows how the requester’s weights and boundaries for
different attribute types are used to compute the requesters individual score for feasible
service compositions constituting complex service instances.
As depicted in Figure 3.5 the service value network contains four service offerings
unambiguously specified by attribute values for the types response time (rt) and encryption (enc). Each feasible path f a = {es1 , e12 , e2 f } and f b = {es3 , e34 , e4 f } from source to
sink represents a possible instantiation of the complex service. Attribute values for the

102

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)
v1
rt
1
enc
1

v2

a = 100
a

=1

rt
2
enc
2

Caption

a = 50
a

v

=1

Service Offer
Composition
Relation

vf

vs

v3
rt
3
enc
3

v4

a = 10
a

=0

rt
4
enc
4

vs

Source Node

vf

Sink Node

a = 150
a

=1

Figure 3.5
Service value network with service offers and corresponding
configurations.

complex service are computed using suitable aggregation operations according to Table
3.1. Hence, the upper path has a response time of Artfa = 150 and an encryption level
rt
enc
Aenc
f a = 1. Analogue for the lower path: A f b = 160 and A f b = 0.
In this example, the requester’s reported vector of boundaries is Γ =
((200, 20), (0, 1)). For simplicity it is assumed that its utility functions for each attribute
type are linear such that
hrt (Artf ) =

200 − Artf
200 − 20

enc
and genc (Aenc
f ) = Af

According to the piecewise defined normalization function (cp. Equation (3.4)), the
requester’s utility for different types of attributes and their values is illustrated in Figure
3.6.
Normalization of the attribute values according to Equation (3.4) leads to the following values for each feasible complex service instance:
rt
enc
kArtfa k = 0.28, kAenc
f a k = 1, kA f b k = 0.22, kA f b k = 0

In the example at hand it is assumed that response time is more important to the
service requester then encryption, which leads to the vector of weights Λ = (0.7, 0.3).
According to Equation (3.2) the requesters final score for each complex service instance
computes as follows:

3.2. BIDDING LANGUAGE
‖A rt‖
1

103
‖A enc‖
1

0
rt
200 a

20

(a) Requester Utility for
Different Levels of
Response Time

0
0

1

a enc

(b) Requester Utility for
Different Levels of
Encryption

Figure 3.6
Requester utility for different attribute types.

S(A f a ) = 0.7 · 0.28 + 0.3 · 1 = 0.496
S(A f b ) = 0.7 · 0.22 + 0.3 · 0 = 0.154
Based on the requester’s preferences (specified by the vector of boundaries), the utility
functions and the vector of weights for different attribute types, the complex service f a
yields a higher individual score, i.e. it is preferable for the service requester.

3.2.2 Service Requests
Having defined how the score for certain outcomes is computed based on the
requester’s preferences, a specification of the willingness to pay is introduced
that determines the rate of substitution between score and price. Let T f = ∑s∈S ts
represent the sum of all monetary transfers to service providers, i.e. the overall
price of the complex service denoted by f . Hence, the requester’s utility gained
from purchasing a complex service specified by a path f with a configuration A f
evolves as follows:

(3.5)

U fR (α, Λ, Γ, A f , T f ) = αS(A f ) − T f

104

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

The factor α represents the requester’s willingness to pay for a ”perfect” configuration A f with score S(A f ) = 1 based on reported preferences. In other
words α defines the individual substitution rate between quality and price such
that the requester is indifferent between an increase of 1 score unit and α monetary units. Incorporating that information, a service request for a multidimensional complex service is defined as follows:
Definition 3.3 [M ULTIDIMENSIONAL S ERVICE R EQUEST ]. A multidimensional
service request for a complex service is a vector of the form:
(3.6)

R := (Y , α, Λ, Γ)

such that Y = (Y1 , . . . , YK ) represents all candidate pools with the service value network,
i.e. necessary information for each service provider about preceeding service offers6 . The
maximum willingness to pay for a configuration that yields a score of 1 is denoted by α.
The set of weights Λ represents the requesters’ preferences for different attribute types
l ∈ L. Γ denotes the set of lower and upper boundaries for each attribute type.
Example 3.3 [M ULTIDIMENSIONAL S ERVICE R EQUEST ]. Recalling Example 3.2, a
multidimensional service request of a requester with a willingness to pay of α = 100 is
denoted by
R = ({v1 , v3 }, {v2 , v4 }, 100, (0.7, 0.3), ((200, 20), (0, 1)))
For realization in a distributed environment such as the Web, compliance with interoperable and standardized exchange formats such as the WS-Agreement specification
[ACD+ 04] is preferable. As the representation of α, Λ and Γ is straightforward, the information about the service value network topology requires an intermediate XML-based
serialization such as the Graph eXchange Language (GXL) [Win02].

3.2.3 Service Offers
Having specified the bidding language for requesters we define a notation for the
provider side. A multidimensional service offer consists of an announced service
configuration A j and a corresponding price pij that a service provider wants to
charge for the service j being invoked depending on the predecessor service i. An
offer bid bij = ( A j , pij ) is a service offer for invocation of service j as a successor of
6 Note

that there are no preceeding service offers for services v with v ∈ Y1 .

3.2. BIDDING LANGUAGE

105

service i. A service provider s announces a matrix of bids Bs ∈ B for all incoming
edges to every service it owns:
Definition 3.4 [M ULTIDIMENSIONAL S ERVICE O FFER ]. A multidimensional service offer is a matrix of bids of the form:

 b = ( A , p ),
ij
j ij
s
B :=
( Ā , −∞),

(3.7)

j

i ∈ τ ( j ), j ∈ σ ( s )
otherwise

with τ (v) denotes the set of all predecessor services to service v with τ : V → V and σ (s)
the set of all services owned by service provider s. Ā j is an arbitrary service configuration.
Example 3.4 [M ULTIDIMENSIONAL S ERVICE O FFER ]. Recall, the computation of
a scoring function is illustrated in Example 3.2. This example is extended with respect
to internal costs that occur on the provider side for the invocation of a service offer in a
certain context. Figure 3.7 shows the extended service value network.

c s1 = 10

v1
rt
1
enc
1

rt
2
enc
2

=1

a

Service Offer

v4

a = 10
cs 3 = 8

a

=0

Composition
Relation

vf

v3
rt
3
enc
3

v

=1

c14 = 6

vs

Caption

a = 50

a = 100
a

v2

c12 = 12

rt
4
enc
4

vs

Source Node

vf

Sink Node

a = 150
c34 = 7

a

=1

Figure 3.7
Service value network with service offers and internal costs.

It is assumed that service offers v1 and v4 are owned by a service provider s1 and
service offers v2 and v3 are owned by another service provider s2 . Therefore, the ownership
information σ (s1 ) = {v1 , v4 } and σ (s2 ) = {v2 , v3 } is public knowledge. For simplicity,
it is further assumed that service providers follow a truth-telling strategy, that is, they
report their multidimensional types truthfully. According to Definition 3.4 the service
offer bid matrixes for service providers s1 and s2 evolve as follows:

106

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)



B s1

B s2

3.3

−∞

−∞

−∞
= 
−∞


−∞
−∞

−∞

−∞

−∞
= 
−∞


−∞
−∞

((100, 1), 10)
−∞
−∞
−∞
−∞
−∞

−∞
−∞
−∞
−∞
−∞
−∞


−∞
−∞


−∞

−∞


−∞
−∞

−∞
−∞


−∞

−∞


−∞
−∞

−∞
−∞
−∞ ((150, 1), 6)
−∞
−∞
−∞ ((150, 1), 7)
−∞
−∞
−∞
−∞

−∞
−∞
((10, 0), 8)
−∞ ((50, 1), 12)
−∞
−∞
−∞
−∞
−∞
−∞
−∞
−∞
−∞
−∞
−∞
−∞
−∞

−∞
−∞
−∞
−∞
−∞
−∞

Mechanism Implementation

To design a procurement auction for complex services we follow the approach of
algorithmic mechanism design as introduced in [NR01]. The discipline of mechanism design forms a subset of game theory that focuses on solving social choice
problems from an engineering perspective accounting for technical constraints
and preconditions. The central objective is to maximize the system’s welfare
by allocating adequate service offers from a set of decentralized, self-interested
and rationally acting service providers. All service providers have private information about their internal costs and the quality of their services representing
the providers’ multidimensional types. The challenge is to design a mechanism
m = (o, t) consisting of an allocation function o and a transfer function t that incentivizes service providers to report their types truthfully to the auctioneer with
respect to all dimensions of all their service offerings. Such truthful information is
necessary in order to achieve the system-wide solution as desired. The allocation
outcome of such a mechanism yields the same solution as the overall problem
based on the same social choice in a fictive setting with complete information
about the agents’ types.
The auctioneer has to solve the problem of allocating a path f ∗ from source
to sink connecting selected service offers within the network G that yields the
highest welfare as the sum of all utilities (consumer and provider surpluses). The
main challenge in such a setting is that types are private information to service
providers. Therefore the auctioneer is not capable of solving the welfare maxi-

3.3. MECHANISM IMPLEMENTATION

107

mization problem directly but instead has to implement adequate incentives to
make truth-telling a dominant strategy equilibrium.

3.3.1 Allocation
Let U f denote the overall utility of path f based on the reported types. Let further
P f be the sum of all price bids for allocated service offers on the path f such that
P f = ∑eij ∈ f pij . The allocation function o : B → F maps the service providers’ bids
B ∈ B – their reported types – to a feasible path from source to sink f ∗ ∈ F7 such
that:

(3.8)

o ( B) := argmax U f = argmax αS(A f ) − P f
f ∈F

f ∈F




Having defined an allocation function to perform a desired social choice that
selects a set of edges within G that determine the instance of the complex service, a function that specifies monetary transfers to service providers has to be
designed. Let U ∗ 8 denote the overall utility of the allocated path meaning the
∗
utility of the path f ∗ , which maximizes the overall utility. Furthermore, let U−
s
denote the overall utility of a path f −∗ s that yields the maximum welfare in a
reduced graph G−s without every service owned by service provider s and without incoming and outgoing edges of these service offers, i.e. the complex service instance that maximizes welfare in an service value network without service
provider s’s participation.
Definition 3.5 [C RITICAL VALUE ]. The critical value ∆tcrit,s of a service provider s
represents its contribution to the system as the difference between the overall utility U ∗
∗ without service
in the complete graph and the overall utility in the reduced graph U−
s
offers owned by service provider s and incoming and outgoing edges of these services such
that
(3.9)
7 For

∗
∆tcrit,s = U ∗ − U−
s

the sake of simplicity, the expression “allocated service offer” means that this service
offer has an incoming edge that is entailed in the allocated set of edges f ∗ . Analogously, the
expression “allocated service provider” means that a service provider owns at least one “allocated
service offer”.
8 For the reader’s convenience, the notion U ∗ is short for U
o ( B) which denotes the overall
utility of the path f ∗ allocated by o ( B) based on service providers’ bids.

108

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

The following example shows the computation of service provider s’s contribution to the system.
Example 3.5 [C RITICAL VALUE AND I NDIVIDUAL C ONTRIBUTION ]. The service
value network in Figure 3.8a consists of four service offers a, b, c and d and source and sink
nodes s and f . Service provider s1 owns two services b and c such that σ (s1 ) = {b, c}. For
simplicity there are no quality attributes of service offers, which implies one dimensional
types of service providers.
0.1

a

0.3

b

0.1

0.2

a
0.2

s

f

s

f

0.1
0.1

c

0.9

(a) Complete Graph with
Participation of z

d

d

(b) Reduced Graph without
Participation of z

Figure 3.8
Critical value and individual contribution.
Values on the edges within the graph denote price bids of service providers for all
incoming edges of service offers they own. Focusing on service provider s1 , there are bids
bab = 0.3, bcb = 0.2 and bsc = 0.1. Assuming a service requester’s willingness to pay of
α the path f ∗ = {esc , ecb , ec f } is allocated by o ( B) as it yields the highest overall utility of
U ∗ = α − 0.2, which represents the highest welfare.
In order to determine service provider s1 ’s critical value ∆tcrit,s1 – i.e. s1 ’s utility
∗ in the reduced graph depicted in
contribution to the system – the overall utility U−
s1
Figure 3.8b without s1 ’s participation is computed. In the absence of service provider
s1 ’s service offers b and c only a single path from source to sink remains. Hence, the path
f −∗ s1 = {esa , ead , ed f } is allocated and represents the only feasible complex service instance
∗ = α − 0.3.
which results in an overall utility of U−
s1
Consequently the critical value evolves as ∆tcrit,s1 = 0.1, which represents service
provider s1 ’s contribution the overall system.

3.3.2 Transfer
Every service provider s receives a monetary transfer ts for all services s owns that
are allocated by o ( B). Analogue to the idea of a second-price auction, a monetary

3.3. MECHANISM IMPLEMENTATION

109

compensation ts − ∑eij |eij ∈o,j∈σ(s),i∈τ ( j) pij for service provider s that owns service
offers j ∈ σ (s) corresponds to the monetary equivalent of the utility gap between
the allocated path and the allocated path in the reduced graph without s and all
its incoming and outgoing edges, i.e the critical value of service provider s. In
other words the additional payment ts − ∑eij |eij ∈o,j∈σ(s),i∈τ ( j) pij ≥ 0 is a monetary
equivalent to the utility service provider s contributes to the overall utility of the
system. Thus, the transfer ts represents the price that service provider s could
have charged without loosing its participation in the winning allocation:

U ∗ − U−∗ s = ts −
t

s

∑

pij

eij |eij ∈o,j∈σ(s),i ∈τ ( j)

∑

=

eij |eij ∈o,j∈σ(s),i ∈τ ( j)

ts =

∗
pij + (U ∗ − U−
s)

pij + ∆tcrit,s

∑
eij |eij ∈o,j∈σ(s),i ∈τ ( j)

Consequently, the transfer function ts for service provider s is defined as

(3.10)

s

t :=


∑

i ∈τ ( j) ∑ j∈σ(s) pij

+ (U ∗ − U−∗ s ), if eij ∈ o

0,

otherwise

The transfer function belongs to the class of VCG-based payment schemes
which implements valuable mechanism properties that are extensively analyzed
in Chapter 5.
Costs cs that service provider s has to bear for performing offered and allocated services result accordingly:

(3.11)

cs :=


∑
0,

i ∈τ ( j) ∑ j∈σ(s) cij ,

if eij ∈ o
otherwise

3.3.3 Summary
The goal of the mechanism implementation is to incentivize service providers
to report their types truthfully to the auctioneer. This fosters a system-wide so-

110

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

lution in a decentralized environment that maximizes welfare among all participants although they are assumed to act self-interested. The properties of the
implemented social choice are extensively analyzed in Chapter 5.
Summarizing the presented mechanism implementation for the complex service auction, Figure 3.9 depicts the mechanism implementation triangle underlaying the complex service auction.




ω(θ ) = argmax αS (A f ) − ∑ cij 

f ∈F
eij ∈ f


Type

Outcome

θ = {θ s | ∀s ∈ S}†

ρ

Mechanism
ψ(θ )

† s

θ = {( A j , cij )| ∀j ∈ σ ( s), ∀i ∈ τ ( j )}

M

m( ψ(θ )) = m( o( B)†† , t( o , B)††† )

††

o( B) = argmax (αS (A ff ) − P
f ∈F

††† s

t ( o , B) =

∑ ∑p

ij

)

+ ( U * − U * −s )

j∈σ ( s ) i∈τ ( j )

Figure 3.9
Triangle relation of the CSA mechanism implementation and
social choice.

3.4

Related Work

Recently, an enormous body of work has been done that blurs the border between game theory and computer science [Pap01]. Especially the discipline of
mechanism design that focuses on the problem to coordinate self-interested participants in pursuing an overall goal are introduced by [NR01]. The authors design suitable mechanisms to standard optimization problems in the area of task

3.4. RELATED WORK

111

scheduling and routing. In incentive compatible mechanisms agents are incentivized to choose the strategy of revealing their true type. Incentive compatible
mechanisms such as the celebrated Vickrey-Clarke-Groves (VCG) mechanism are
firstly introduced and extensively investigated by [Vic61, Cla71, Gro73, GL78].
Most of the research has been done with respect to truth-telling of onedimensional types. The field of designing incentive compatible mechanisms,
that induce truth-telling of multidimensional properties of goods or services, still
lacks deeper research. A thorough analysis and investigation in the area of multidimensional optimal auctions and the design of optimal scoring rules has been
done by [CIoWM93, Bra97, AC05]. An investigation of the winner determination problem in configurable multiattribute auctions from an operational research
perspective without accounting for mechanism design aspects such as incentive
compatibility has been done in [BK05]. In [PK02, PK05], iterative multiattribute
procurement auctions are introduced while focusing on mechanism design issues
and on solving the multiattribute allocation problem. Preferences for multidimensional goods and multidimensional types in scoring auctions are extensively
investigated in [AC08] and extended to combinatorial auctions in [MPW08]. Nevertheless their work does not consider compositions and sequences of services as
well as their technical feasible interrelations in order to coordinate value generation. All of these approaches assume bundles of goods in scenarios where the
sequence and order does not matter and therefore cannot be applied to composite
services that only fulfil their objectives in the right sequence of composition.
Nevertheless, combinatorial auctions yield major drawbacks regarding computational feasibility that result from an NP-hard complexity. Computational feasibility implies a trade-off between optimality and valuable mechanism properties such as incentive compatibility. Several authors propose approximate solutions for incentive compatible mechanisms to overcome issues of computational complexity [MN08b, NR07, Ron01, RL05]. Path auctions as a subset of
combinatorial auctions reduce complexity through predefining all feasible service combinations in an underlying graph topology and are investigated by
[FRS06, HS01, AT07]. In their work, path auctions are utilized for pricing and
routing in networks of resources such as computation or electricity. Applicationrelated issues of auctions to optimal routing are examined by [FCSS05, MT07].
All of these approaches deal with the utility services layer according to the service classification by [BS08, BBS08] and hence do not cover the problems related
to elementary services and complex services.

112

3.5

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

Auction Process Model & Architecture

The auction conduction is divided in two main phases: a solicitation phase and the
actual auction phase as depicted in Figure 3.10.

Ž›Ÿ’ŒŽȱŽšžœŽ›

Ž›Ÿ’ŒŽȱ—Ž›–Ž’Š›¢

Ž›Ÿ’ŒŽȱ›˜Ÿ’Ž›

˜•’Œ’Š’˜—ȱ‘ŠœŽ

˜–™•Ž¡ȱŽ›Ÿ’ŒŽȱ˜•’Œ’Š’˜—

Ž•ŽŒȱ—Ÿ’Š’˜—ȱŠ—’ŠŽœ
˜–™•Ž¡ȱŽ›Ÿ’ŒŽȱ˜•’Œ’Š’˜—

—’’Š•ȱŽ›Ÿ’ŒŽȱŽ›

•Š——’—

žŒ’˜—ȱ‘ŠœŽ

Ž›Ÿ’ŒŽȱŽšžŽœ

Ž›Ÿ’ŒŽȱŽ›

Žœ˜•ž’˜—
˜’’ŒŠ’˜—

˜’’ŒŠ’˜—

Figure 3.10
Process model of the CSA.

3.5. AUCTION PROCESS MODEL & ARCHITECTURE

113

The solicitation phase serves as an initial screening phase regarding the service request and potential service provider candidates to be invited to participate
in the auction. The service requester sends a complex service solicitation to the service intermediary which initiates the coordination process. The complex service
solicitation specifies required modularized functionality which determines the
candidate pools that are sequentially involved in the production of the complex
service requested.
Based on this information, the service intermediary reasons about potential
service providers to be invited to participate in the auction phase. There are different forms of finding and defining suitable participants. The service intermediary can step into the role of pushing the invitation process using e.g. a registry to
find adequate service providers. It is also possible to reverse the roles in such a
lookup scenario, meaning that potential participants are proactively searching for
suitable coordination services provided by a service intermediary. Potential participants could also subscribe to a notification service – analogue to the observer
design pattern – in order to automatically be informed if an adequate auction
service is available.
Having defined the set of potential service providers to participate in the auction phase, the service intermediary sends out the complex service solicitation
and additional information as an invitation to the candidates. This information
enables service providers to register their service offerings to be part of the service value network and to be considered in the auction phase by sending initial
service offers.
Combining the information about the complex service solicitation and the initial service offers from service providers, the service intermediary plans the topology of the service value network and proceeds its virtual formation (cp. Section
2.1.4 and Section 3.1). This step concludes the solicitation phase and lays the basis
to the actual auction phase.
The auction phase embodies the central coordination process to allocate and
price complex services. Messages and information objects exchanged during the
auction conduction are fully specified according to the bidding language in Section 3.2. The topology information about the service value network as well as the
requester’s preferences and willingness to pay is sent as a service request (cp. Section 3.2.2) to registered service providers. Having received the requester’s information, the service providers privately submit their service offers – as specified in
Section 3.2.3 – to the service intermediary. Having collected necessary information from requester and provider side, the service intermediary resolves the auc-

114

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

tion by computing the winner determination and resulting monetary transfers.
The auction process concludes with notifications about the final outcome and
corresponding transfers sent to the service requester and the service providers.
Providing an architectural overview, Figure 3.11 shows service providers that
intent to participate in the auction, their service offers which are realized in a
lightweight manner and necessary big Web services that enable the overall coordination of the auction process.
Complex Service Auction Platform

WSDL
Interface

Abstract
Composition

Coordinator
Service

Candidate binding

Candidate binding

Candidate binding
Auction process coordination

Service
Offer

Service
Offer
REST
Interface

Service
Provider

Service
Offer
REST
Interface

Service
Offer
REST
Interface

Service
Provider

Participant
Service
WSDL
Interface

REST
Interface

Participant
Service
WSDL
Interface

Figure 3.11
Architectural overview of the CSA.
The CSA platform as the central coordination unit communicates with potential participants via a coordinator service implemented as a Web service with a
WSDL interface. Analogously, each service provider exposes a participant service
for the message exchange with the coordinator. After the coordination phase
is completed, concrete candidate service instances are bound to each step in
the abstract composition in a lightweight manner leveraging the simplicity of

3.6. REALIZATION & IMPLEMENTATION

115

REST/HTTP interfaces. The final composition embodies the outcome of the coordination process in the form of a concrete complex service instance.

3.6 Realization & Implementation
This section provides an in-depth analysis of the ComputeAllocation algorithm
which performs the winner determination in the complex service auction. Special
challenges that result from aggregation operations such as min and max as well
as Boolean operations which are used in the context of semantic QoS extensions
(cp. Section 4.3) are outlined and adequate remedies are discussed. The procedure of the algorithm is illustrated stepwise by means of an extensive example.
Furthermore, this section introduces a prototypical implementation of a service
value network planner tool and an agent-based simulation tool to analyze the
complex service auction.
From an algorithmic mechanism design perspective computational feasibility
according to Requirement 5 is a central desideratum in order to implement the
mechanism in an online system which requires on-the-fly computation at runtime.
It is well-known that solving the winner determination problem in general
combinatorial auctions is N P -complete. Focusing on finding efficient computational approaches, several algorithms have been proposed to solve the winner
determination problem [PS98, RPH98, SSGL05].
The solution to the allocation problem in (3.8) can be compute in polynomial
time using well-known graph algorithms to determine the “shortest” path within
a network such as the Dijkstra algorithm [Dij59].
According to the payment scheme in (3.11) the allocation must be computed
twice for each allocated service offer – based on the graph with the service offerings of the service provider receiving the payment and without its participation.
In the second case the graph can be preprocessed and reduced by all service offerings owned by the service provider that receives the payment. After the reduction the allocation can be computed accordingly which yields the same time
complexity.
Nevertheless, the extension of the complex service auction with respect to
complex QoS aggregation using also aggregation operations that require complete information about predecessors’ attribute values – memory-dependent attribute types (e.g. cp. Section 4.3) – such as min, max and Boolean operations may

116

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

result in sub-optimal solutions using the traditional Dijkstra algorithm. Analogue
to the problem of negative edge weights which is well-known in literature [Dij59],
memory-dependent operations may result in non-monotone utility characteristics. Such behavior conflicts with the main procedure of the Dijkstra algorithm,
that is, it truncates a sub-path which is directly dominated by another sub-path
that intersects it at the point of intersection. Considering an attribute type encryption which is aggregated by a Boolean AND operation according to Table 3.1.
A sub-path f s1 dominates another sub-path f s2 as it yields a higher utility which
results from an aggregated value for encryption of TRUE. In case both sub-paths
intersect at a certain node, the Dijkstra algorithm only considers f s1 and drops f s2
as f s1 yields a higher overall utility so far. Nevertheless, this might be error prone
if the subsequent service offer does not support encryption which leads to an aggregated encryption value for f s1 of FALSE. Hence, the former decision of dropping f s2 might have been incorrect since now both sub-paths are not encrypted
and f s2 might dominate f s1 in price.
To overcome illustrated shortcomings of the Dijkstra algorithm, Algorithm 3.1
accounts for attribute types which are aggregated by memory-dependent operations always yielding an optimal solution.

Algorithm 3.1 ComputeAllocation
Require: V, E, B
1: Q ← getNodesPoolWise (V )
2: for all u ∈ Q do
states [u] ← getNonMonotoneStates (u)
3:
4:
for all w ∈ states [u] do
5:
utility [u][w] ← −∞
6:
path [u][w] ← ∅
7: while getNextNode ( Q ) 6 = null do
8:
u ← getNextNode ( Q)
9:
removeNode (u, Q)
10:
for all v ∈ getSuccesors (u, E) do
11:
for all w ∈ states [u] do
12:
w̄ ← computeState (w, euv , B)
13:
altUtility ← computeUtility (path [u][w] ∪ {euv }, B)
14:
if altUtility > utility [v][w̄] then
15:
utility [v][w̄] ← altUtility
16:
path [v][w̄] ← path [u][w] ∪ {euv }
∗
17: w ← argmaxw∈states [v ] (utility [ v f ][ w ])
f
18: return path [ v f ][ w∗ ]

3.6. REALIZATION & IMPLEMENTATION

117

In order to describe the procedure of the ComputeAllocation algorithm and
its complexity, Algorithm 3.1 is divided into 3 parts, namely the initialization phase
(lines 1-6), the main phase (lines 7-16) and the consolidation phase (lines 17-18).
Initialization phase In the initialization phase, required variables are initialized
and set to their starting values. In contrary to the traditional Dijkstra algorithm, the ComputeAllocation algorithm visits every node within the
graph which is equal to the worst-case behavior of a Dijkstra search. Therefore the node queue Q entails all nodes u ∈ V ordered by the sequence
of the candidate pools in the network such that getNodesPoolWise(V) =
(u11 , . . . , u1|Y | , . . . , u1K , . . . , u|KY | )9 with {u11 , . . . , u1|Y | } = Y1 and {u1K , . . . , u|KY | } =
K
K
1
1
YK . The function getNonMonotoneStates (u) retrieves all possible combinations of memory-dependent attribute values of service offer u. Exemplary, if service offer u is only characterized by an encryption attribute type
with boolean values, hence getNonMonotoneStates (u) = {TRUE, FALSE}.
Let the set W entail all possible states after aggregation, then the time complexity of the initialization phase is O(|V | · |W |).
Main phase In the main phase, the algorithm iterates over all nodes in Q and
removes each node after processing until there is no entry left in the queue.
Each successor v of the current node u is evaluated for all states of u. The
utility of the sub-path including v is computed based on the overall utility U f introduced in Section 3.3.1. These alternatives are compared to the
current utility entry for node v and updated in case of improvement. The
variables utility and path store for each node u and each state the highest
utility and the corresponding path respectively. Traversing all successors of
every node in Q, the ComputeAllocation algorithm processes every edge in
the main phase and compares every state of each node. This leads to a time
complexity of the main phase of O(| E| · |W |).
Consolidation phase After the main part has terminated once Q is empty, i.e. all
nodes have been processed, the consolidation phase evaluates the results.
The path from source to sink is analyzed and the state s∗ that maximizes
the overall utility is determined. Based on this state the final allocation
path [v f ][s∗ ] is returned. Implemented as a linear search, the consolidation
phase yields a time complexity of O(|W |).
The time complexity of the ComputeAllocation algorithm consisting of the
initialization phase, the main phase and the consolidation phase evolves as
9 The

order within each candidate pool is not important.

118

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

O(|V | · |W | + | E| · |W | + |W |). Assuming a worst case number of edges with
|V |−2
respect to the number of nodes | E| can be substituted by ( 2 )2 + (|V | − 2).
This leads to an overall complexity of O(|W | · |V |2 ). The time complexity regarding the number of service offers and connecting edges, the number of paths
respectively, is polynomial which means that the algorithms run-time is robust
with respect to a changing number of participants and feasible complex service
instances. In contrary to the N P -complete complexity in general combinatorial
auctions this is a valuable achievement that enables the conduction of the complex service auction in online systems.
Nevertheless, with respect to the number of memory-dependent attribute
types and the number of their discrete values, the computational complexity is
exponential (e.g. assuming N Boolean attribute types, |W | = 2 N ). From a domainspecific perspective, the impact of this theoretical result is rather weak, as the
number of states that have to be iterated by the algorithm decreases rapidly in the
average case. Figure 3.12 illustrates the run-time performance of the ComputeAllocation algorithm in a scenario with 100 service offers in 10 candidate pools
(cp. Figure 3.12a) and 1000 service offers in 100 candidate pools (cp. Figure 3.12b).
The service value network is assumed to be fully connected which means that
each service offer has the maximum number of incoming edges which results in
the maximum number of feasible paths from source to sink. The algorithm’s performance is evaluated dependent on the number of memory-dependent attribute
types. Attribute types are assumed to be Boolean and their values are uniformly
distributed for each service offer. Although the theoretical worst case analysis
of the computational complexity is exponential with respect to the number N of
memory-dependent attribute types ( O(2 N )), the average case with boolean attribute types results in a linear increasing computation time. The ComputeAllocation algorithm quickly solves the winner determination problem even for huge
instances and satisfies Requirement 5 (computational tractability).
Example 3.6 [A LLOCATION C OMPUTATION WITH M EM .- DEPENDENT Q O S].
This example illustrates the procedure of the ComputeAllocation algorithm in a stepwise manner based on the service value network as depicted in Figure 3.13.
The service value network consists of 6 service offers V = {1, 2, 3, 4, 5, 6} ∪ {s, f }.
Each service offer u is unambiguously configured through a boolean attribute value aenc
u
for the attribute type encryption whereas 1 ≡ TRUE and 0 ≡ FALSE. Values on incoming
edges pij represent price bids of service providers. It is assumed that the service requester’s
willingness to pay αS(A f ) for a complex service depending on its QoS characteristics A f
evolves as

3.6. REALIZATION & IMPLEMENTATION

(a) Performance analysis with 100 service offers in 10 candidate pools.

(b) Performance analysis with 1000 service offers in 10 candidate pools.
Figure 3.12
Performance analysis of the ComputeAllocation algorithm.

119

120

CHAPTER 3. COMPLEX SERVICE AUCTION (CSA)

ps 1 = 1

1
enc
1

a

=1

p12 = 6

2
a

enc
2

=1

p23 = 2

3
a

enc
3

Caption

=1

v

Service Offer

p15 = 2

p26 = 2

Composition
Relation

f

s

s

Source Node

f

Sink Node

p42 = 1
5

4
ps 4 = 2

a4enc = 0

p45 = 2

a5enc = 1

6

p56 = 1

a6enc = 0

Figure 3.13
Service value network with service offers exposing
memory-dependent attribute types.


15, if A = 1
f
αS(A f ) =
12, if A = 0
f

Table 3.2 illustrates the algorithm’s procedure to find an optimal allocation based on
the allocation function in Section 3.3.1 accounting for the memory-dependent attribute
type encryption representing the QoS of service offers.
In the last step when node f is processed, the optimal path given a not encrypted
∗
complex service results as f FALSE
= {es1 , e15 , e56 , e6 f } and yields an overall utility of
∗
∗
= 8. Given a encrypted complex service, the optimal allocation is f TRUE
=
U fFALSE
∗
∗
{es1 , e12 , e23 , e3 f } with an overall utility of U fTRUE
= 6. Thus, the state s = FALSE
yields an optimal path f ∗ = {es1 , e15 , e56 , e6 f } that maximizes the system’s overall utility
U ∗ = 8.

1

1

2

2

3

3

4

4

5

5

6

6

f

f

utility
path

utility
path

utility
path

utility
path

utility
path

utility
path

utility
path

utility
path

utility
path

utility
path

utility
path

utility
path

utility
path

utility
path

FALSE

TRUE

FALSE

TRUE

FALSE

TRUE

FALSE

TRUE

FALSE

TRUE

FALSE

TRUE

FALSE

TRUE

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

12
∅

s

utility
path

FALSE

s

utility
path

TRUE

{1, 4, 2, 5, 3, 6, f }

{s, 1, 4, 2, 5, 3, 6, f }

15
∅

Q

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

10
{es4 }

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

14
{es1 }

12
∅

15
∅

s

-

Node

−∞
∅

−∞
∅

−∞
∅

−∞
∅

−∞
∅

12
{es1 , e15 }

10
{es4 }

−∞
∅

−∞
∅

−∞
∅

−∞
∅

8
{es1 , e12 }

−∞
∅

14
{es1 }

12
∅

15
∅

{4, 2, 5, 3, 6, f }

1

−∞
∅

−∞
∅

−∞
∅

−∞
∅

8
{es4 , e45 }

12
{es1 , e15 }

10
{es4 }

−∞
∅

−∞
∅

−∞
∅

9
{es4 , e42 }

8
{es1 , e12 }

−∞
∅

14
{es1 }

12
∅

15
∅

{2, 5, 3, 6, f }

4

−∞
∅

−∞
∅

7
{es4 , e42 , e26 }

−∞
∅

8
{es4 , e45 }

12
{es1 , e15 }

10
{es4 }

−∞
∅

7
{es4 , e42 , e23 }

6
{es1 , e12 , e23 }

9
{es4 , e42 }

8
{es1 , e12 }

−∞
∅

14
{es1 }

12
∅

15
∅

{5, 3, 6, f }

2

−∞
∅

−∞
∅

8
{es1 , e15 , e56 }

−∞
∅

8
{es4 , e45 }

12
{es1 , e15 }

10
{es4 }

−∞
∅

7
{es4 , e42 , e23 }

6
{es1 , e12 , e23 }

9
{es4 , e42 }

8
{es1 , e12 }

−∞
∅

14
{es1 }

12
∅

15
∅

{3, 6, f }

5

7
{es4 , e42 , e23 , e3 f }

6
{es1 , e12 , e23 , e3 f }

8
{es1 , e15 , e56 }

−∞
{es1 , e12 , e26 }

8
{es4 , e45 }

12
{es1 , e15 }

10
{es4 }

−∞
∅

7
{es4 , e42 , e23 }

6
{es1 , e12 , e23 }

9
{es4 , e42 }

8
{es1 , e12 }

−∞
∅

14
{es1 }

12
∅

15
∅

{6, f }

3

Table 3.2: Allocation computation stepwise procedure example.

8
{es1 , e15 , e56 , e6 f }

6
{es1 , e12 , e23 , e3 f }

8
{es1 , e15 , e56 }

−∞
∅

8
{es4 , e45 }

12
{es1 , e15 }

10
{es4 }

−∞
∅

7
{es4 , e42 , e23 }

6
{es1 , e12 , e23 }

9
{es4 , e42 }

8
{es1 , e12 }

−∞
∅

14
{es1 }

12
∅

15
∅

{f}

6

8
{es1 , e15 , e56 , e6 f }

6
{es1 , e12 , e23 , e3 f }

8
{es1 , e15 , e56 }

−∞
∅

8
{es4 , e45 }

12
{es1 , e15 }

10
{es4 }

−∞
∅

7
{es4 , e42 , e23 }

6
{es1 , e12 , e23 }

9
{es4 , e42 }

8
{es1 , e12 }

−∞
∅

14
{es1 }

12
∅

15
∅

∅

f

3.6. REALIZATION & IMPLEMENTATION
121

Chapter 4
Applicability Extensions
The management of QoS metrics directly impacts the success of organizations
participating in e-commerce.

[CSM+ 04]

his section introduces design extensions to the complex service auction to
enable the applicability in service value networks in order to coordinate distributed activities in creating and provisioning complex services to customers. A
compensation transfer function is introduced in Section 5.1.2. The auction conduction is divided in a declaration phase and an execution phase in order to
incorporate ex-post information on provided QoS levels (monitoring information) into the monetary transfers which are distributed among participating service providers. Counteracting the absence of budget balance, Section 4.2 introduces the budget-balanced interoperability transfer function (ITF). By sacrificing
incentive compatibility to a certain degree, the design of the payment scheme incentivizes service providers to increase their services’ degree of interoperability.
Properties of the ITF are analyzed in detail in Section 6.2. As quality aspects are
gaining importance especially in the context of services, Section 4.3 introduces
and rule-based extension to the complex service auction which allows for the description and evaluation of complex QoS characteristics and their incorporation
in the allocation and pricing component of the basic mechanism.

T

124

4.1

CHAPTER 4. APPLICABILITY EXTENSIONS

Verification and Service Level Enforcement

In Section 2.1.3.3, the expressiveness of the complex service auction with respect
to complex QoS characteristics and their management has been introduced in
detail. From a computer science perspective, protocols and algorithms for distributed environments such as the Internet have been designed under the implicit assumption that participants report their information (e.g. the QoS of their
service offers) truthfully. This assumption only holds for predefined algorithms
and processes that produce a deterministic outcome but not in the context of selfinterested service providers that constantly seek to maximize their individual
utility while participating in distributed systems.
This section provides an extension for the complex service auction that enhances the transfer function (cp. Section 2.2.3.5) by a compensation function,
which on the one hand punishes service providers for untruthful announcements
about the QoS of their service offers and on the other hand compensates service
requesters for the utility loss they incur due to resulting non-performance.

4.1.1 Related Work
The assumption that service providers only announce attribute values that they
actually perform during execution is not realistic [NRTV07]. The basic assumption in traditional mechanism design theory is that agents can follow any of their
strategies no matter what their type is1 . Nevertheless, especially in algorithmic
mechanism design, settings are observed in which computer systems can gain extra information about the agents and their behavior that can be used in the mechanism. According to [NR01] the mechanism implementation can be divided into
two phases: a declaration phase and an execution phase.
Declaration phase In the declaration phase the service requester and the service
providers announce requests and offers according to the bidding language
introduced in Section 3.2. The declaration phase predominantly collects information objects exchanged according to the coordination protocol. These
information objects represent agents’ types which are directly reported to
the coordinator. This information which is explicitly announced by the
agent, is the only information available to the coordinator at this point of
time.
1 Nevertheless

it is obvious that the agents’ strategy space is limited due to technological and
physical restrictions

4.1. VERIFICATION AND SERVICE LEVEL ENFORCEMENT

125

Execution phase Based on the information gathered in the declaration phase, the
coordinator allocates a subset of service offers that together form the desired complex service instance. In the execution phase the service offers
that have been allocated by the mechanism embody the complex service instance, which is executed sequentially. During this phase the actual realized
output of each participant can be observed by the coordinator using monitoring techniques [SMS+ 02, PBB+ 04]. Required monitoring tasks can also
be outsourced by the coordinator in order to leverage external core competencies [Men02]. Such a scenario enables the coordinator to observe the
agents’ types with respect to reported QoS attributes and control the actual
outcome of offered services. Consequently, payments to allocated agents
are transferred after execution in order to incorporate ex-post information
about the services’ performances.

The utilization of the extra information about the agents that can be observed
ex-post in the execution phase enables the design of a penalty for deviating from
the announced attributes. That is an equivalent monetary penalty component
which enhances the transfer function in order to implement a threat based on a
punishment for lying about the offered QoS.

4.1.2 Compensation
Let alj be the announced attribute value for attribute type l of service j’s configuration. Furthermore let ãlj be the verified attribute value for attribute type l realized
by service j and monitored during execution. Analogously, A j and à j denote
announced and verified configurations of service j. Distinguishing between announced and verified attribute values, the overall utility may also differ. Recall
that U ∗ denotes the ex-ante overall utility of the allocated path f ∗ based on the
information available in the declaration phase. Furthermore, Ũ ∗s denotes the
ex-post overall utility that results from the complex service instance formed by
allocated service offers on a path f ∗ and based on the verified attribute values
ã1j , . . . , ãlj of all service offers j ∈ σ (s). According to the Compensation-and-Bonus
mechanism introduced in [NR01] a compensation function ∆tcomp,s is constructed
as follows:

(4.1)

∆tcomp,s := (U ∗ − Ũ ∗s )

126

CHAPTER 4. APPLICABILITY EXTENSIONS

The compensation function represents the overall utility gap that results from
the utility difference based on the announced attribute values and the verified
ones measured after execution. In other words ∆tcomp,s is the utility loss the whole
system incurs because of service provider s’s untruthful announcement(s). The
monetary equivalent to this utility gap represents the penalty payment the untruthful service provider has to bear for deviating from the announced attribute
values. This “negative consequence” can be interpreted as a contractual penalty
for not realizing specified service level agreements2 as defined in [SB04]. Based
on the design of the compensation function the transfer function is extended as
follows:

(4.2)

ts :=





∑ ∑

pij + ∆tcrit,s − ∆tcomp,s , if eij ∈ o

j∈σ(s) i∈τ ( j)


0,

otherwise

Example 4.1 [S ERVICE L EVEL V ERIFICATION AND E NFORCEMENT ]. This example illustrates the effect of untruthful announcements about QoS characteristics on the
whole system and the service requester. It further demonstrates how the compensation
function counteracts such behavior through imposing a penalty on the causer, which
represents the utility loss regarding the whole system while compensating the service
requester and retaining the previous level of overall utility.
Figure 4.1 shows a service value network with four service offers V = {1, 2, 3, 4} ∪
{s, f }. For simplicity it is assumed that each service provider owns a single service offer
within the network such that σ (s1 ) = {1}, τ (s2 ) = {2}, σ (s3 ) = {3} and σ (s4 ) = {4}.
There are two feasible paths from source to sink representing a complex service instance
f 1 = {es1 , e12 , e2 f } and f 2 = {es3 , e34 , e4 f }. Each service configuration is characterized by
a single attribute value aer of the attribute type error rate3 which is aggregated according
to Table 3.1. A value for error rate represents the average percentage of failures during
execution. Values on incoming edges pij represent price bids of service providers for the
corresponding service offer.
The analysis of the example scenario is divided into the declaration phase and the
execution phase:

2 For

the design of the verification payment scheme a risk-neutral service requester is assumed. In real-world scenarios a rather risk averse design of SLAs is observable, overcompensating
service requesters in case of non-performance of service providers.
3 Error rate describes the ratio of occurred number of failed operations during execution compared to the total number of operations executed by the service.

4.1. VERIFICATION AND SERVICE LEVEL ENFORCEMENT

ps1 = 10

1

p12 = 6

er
1

a = 0.1%

2

127

Caption

er
2

a = 0.5%

v

Service Offer
Composition
Relation

f

s

ps 4 = 1

3

4

a3er = 1.0%

a4er = 0.7%

p34 = 12

s

Source Node

f

Sink Node

Figure 4.1
Service value network with service offers characterized by error
rate quality attributes.
Declaration phase (ex-ante) Service providers announce prices and configurations of
the service offers they own (cp. Figure 4.1). The service requester announces a
er
lower boundary γer
B = 0.02 and an upper boundary γT = 0 which means that an
error rate equal or greater than 2% yields a utility of 0 and an error rate equal to
0% results in maximum utility of 1. The service requester’s willingness to pay for
a complex service with score 1 is reported as α = 50. Assuming a linear utility
characteristic with respect to error rates between the boundaries, the requester’s
score for a complex service depending on its QoS evolves as follows:


0.02−Aer
f


, if 0 < Aerf < 0.02

 0.02
S(A f ) = kAerf k = 1,
if Aerf = 0



0,
if Aerf ≥ 0.02

This leads to the following scores for paths f 1 and f 2 :

0.02 − max {0.001, 0.005}
= 0.75
0.02
0.02 − max {0.01, 0.007}
S(A f 2 ) =
= 0.5
0.02

S(A f 1 ) =

The overall utility caused by each allocation consequently is U f 1 = 50 · 0.75 − 16 =
21.5 and U f 2 = 50 · 0.5 − 13 = 12. As U f 1 > U f 2 the upper path is allocated
by o ( B). If transfers would be given in the declaration phase, service provider

128

CHAPTER 4. APPLICABILITY EXTENSIONS
s1
s1 received tex-ante
= 10 + (21.5 − 12) = 19.5 and service provider s2 received
s2
tex-ante = 6 + (21.5 − 12) = 15.5. This would lead to a service requester’s utility
R
of Uex-ante
= 50 · 0.75 − (19.5 + 15.5) = 2.5.

Execution phase (ex-post) After the completion of the declaration phase and the final
allocation based on the reported types, the complex service instance is executed
and the performance of each service component is verified using a monitoring service. The quality announced by service provider s1 for the service offer 1 can be
confirmed. In contrary, service component 2 produces a marginal failure during
execution which increases the announced error rate from 0.5% to 0.6%. The compensation function regarding service offer 2 evolves as:
∆tcomp,s2 = (U ∗ − Ũ ∗s2 )


0.02 − max {0.001, 0.006}
− 16 = 2.5
= 21.5 − 50 ×
0.02
Hence, the monetary equivalent to the utility loss caused by service provider s2
is 2.5. According to the extended transfer function (Equation 4.2), the ex-post
s2
transfer for service provider s2 including the penalty is tex-post
= 10 + (21.5 −
12) − 2.5 = 13. The decrease in transfer represents the monetary compensation for
the loss in quality which compensates the service requester. The service requester’s
R
utility is equal to the ex-ante situation as Uex-post
= 50 × 0.7 − (19.5 + 13) =
R
2.5 = Uex-ante .
The service level enforcement extension to the complex service auction satisfies Requirement 8. Incentives provided by the mechanism’s extension are central
to implement favorable properties with respect to the service providers’ multidimensional bids and their services’ true QoS characteristics. Such properties are
analyzed in detail in Section 5.1.2.

4.2

Achieving Budget Balance

Recall that the mechanism implementation of the complex service auction as
introduced in Section 3 consists of a transfer function that pays each service
provider z that owns allocated service offers the corresponding price bid and
the critical value ∆tcrit,z in addition. The critical value represents a monetary
equivalent to the provider’s utility contribution to the whole system such that
∗ . Price bids of each service offer that is allocated by the mech∆tcrit,z = U ∗ − U−
z
anism plus the corresponding critical value has to be payed by the service re-

4.2. ACHIEVING BUDGET BALANCE

129

quester to the service providers. A provider’s critical value compensates the individual contribution to the system which depends on the contributions of the
other participants. Hence, the payments, the service requester has to distribute
among service providers depend on multiple factors (e.g. the network topology).
In case the payments exceed the requester’s willingness to pay in the complex
service auction, the budget balance (cp. Requirement 4) cannot be achieved by
the mechanism.
Example 4.2 [A CHIEVING B UDGET B ALANCE ]. This example illustrates a nonbudget-balanced outcome of the complex service auction. Figure 4.2 shows a service value
network with service offers V = {1, 2, 3, 4, 5, 6} ∪ {s, f }. For simplicity it is assumed that
each service provider s1 , . . . , s6 only owns a single service within the network such that
σ (si ) = {i } with i = 1, . . . , 6. Furthermore it is assumed that the requester’s willingness
to pay is α = 12.
1
2

6

2

2
4

s

5

3

6

4

f

5
6

3

5

7

6

Figure 4.2
Non-budget-balanced outcome of the CSA.
The mechanism allocates the path f ∗ = {es1 , e14 , e4 f } as it yields the highest overall utility of U f ∗ = 12 − (2 + 2) = 8. According to the transfer function, each service provider that owns allocated service offers receives a payment consisting of the
corresponding price bid and the critical value such that t1 = 2 + (8 − 3) = 7 and
t4 = 2 + (8 − 4) = 6. The sum of transfers which are distributed among the service
providers exceeds the service requesters willingness to pay as U R = 12 − (7 + 6) = −1.
Thus, an amount of 1 unit has to be externally subsidized in order to obtain the efficient
allocation maximizing welfare.
This section introduces an extension to the complex service auction that restores the desideratum of budget balance (cp. Requirement 4) by sacrificing truthfulness to a certain degree. The extension is based on the design of a transfer
function – the Interoperability Transfer Function (ITF) – that limits overpayments

130

CHAPTER 4. APPLICABILITY EXTENSIONS

to satisfy budget balance constraints (cp. Section 2.2.3.5). The ITF implements
incentives for increasing services’ interoperability with adjacent offers to foster
the growth of agile service value networks with an increased level of feasible
complex service instantiations.

4.2.1 Related Work
In VCG-based mechanisms, the transfers are indeterministic and can be arbitrarily high [AT07]. These so called overpayments or a mechanism’s frugality is a central characteristic of a mechanism implementation, which is extensively analyzed
in mechanism design research especially in the context of graph-based implementations [ESS04, AT07, Tal03, KK05]. A frugality ratio that measures the payments
in a truthful mechanism compared to a non-truthful implementation is a ratio
that “characterizes the cost of insisting on truthfulness” [KK05]. Approaches to
predict overpayments that occur in truthful graph-based mechanisms have been
developed in [KN04] in the context of random graphs and in [KN05] for largescale networks.
Addressing this shortcoming of VCG-based mechanisms, an approximately
efficient and budget-balanced solution to overpayment issues in VCG-based combinatorial auctions is introduced in [PKE01] while focusing on solving linear
problems subject to budget balance that yield approximate incentive compatible
solutions. Another approach to counteract the loss of budget balance by sacrificing efficiency is introduced in [AT07] in the context of path auctions. In their work
they replace the efficient allocation function by a class of ”minimum functions”
that yield lower overpayments in certain scenarios. Nevertheless they show that
it is always possible to construct worse case scenarios in which minimum functions perform as bad as the efficient variant.

4.2.2 Interoperability Transfer
Let T denote the sum of all incoming edges to service offers V \ {v f }. Furthermore let τi be the number of incoming edges to service offer i such that
τ
∑i∈V \{v f } τi = T. The ratio ri = Ti denotes the incoming-edge-ratio for each node.
Recall, eui represents an interoperable connection of service i ∈ V with service
u ∈ V, meaning that service i is capable of interpreting service u’s output, i.e. service i is interoperable with service u. Thus, the more incoming edges to a service
offer, the higher its feasible interoperability with its predecessor services. Hence,

4.2. ACHIEVING BUDGET BALANCE

131

the incoming-edge-ratio ri represents the degree of interoperability of service i
with its predecessor services in comparison to all other services. Focusing on all
service offers owned by a service provider s, the ratio r s =
incoming-edge-ratio of service provider s.

∑i∈σ(s) τi
T

denotes the

Let ∆tcrit,s denote the critical value of service provider s. The idea to construct a transfer function that accounts for budget balance constraints is based
on the work in [PKE01] and focuses on choosing adequate discounts ∆s for each
service provider s ∈ S instead of paying every allocated service provider the critical value. The decision on how to choose adequate discounts is formulated as a
general optimization problem subject to budget balance constraints.

(4.3)

Lτ (∆, ∆tcrit,s ) =

∑ rs (∆tcrit,s − ∆s )
s∈S

Lτ represents the weighted distance function that measures the distance between the service providers’ critical values and computed discounts with respect to the incoming-edge-ratio. The goal is to distribute the surplus S∗ =
αS(A f ∗ ) − P f ∗ in a way that it minimizes the distance function Lτ . In other
words, the goal is to transfer discounts ∆s to service providers, which together
minimize the overall weighted distance ∑s∈S r s (∆tcrit,s − ∆s ) and do not exceed
the surplus S∗ . Minimizing the distance function Lτ subject to budget balance,
individual rationality and the critical values as upper boundaries leads to the
following special optimization problem:

(4.4)

min ∑ r s (∆tcrit,s − ∆s )
∆ s∈S

s.t.

∑ ∆ s ≤ S∗

(BB)

s∈S

∆s ≤ ∆tcrit,s , ∀s ∈ S
∆s ≥ 0, ∀s ∈ S
The Lagrangian problem consequently follows such that

z(λ) = min ∑ r s (∆tcrit,s − ∆s ) + λ( ∑ ∆s − S∗ )
∆ s∈S

s∈S

(CV)
(IR)

132

CHAPTER 4. APPLICABILITY EXTENSIONS
s.t. 0 ≤ ∆s ≤ ∆tcrit,s , ∀s ∈ S

The problem decomposes into smaller problems for each s.

min
(r s ∆tcrit,s ) − ∆s (λ − r s )
s
∆

s.t. 0 ≤ ∆s ≤ ∆tcrit,s , ∀s ∈ S
If the coefficient (λ − r s ) is negative, the expression is minimized by setting
∆s to the maximum value that does not violate the side condition which is ∆∗s =
∆tcrit,s . If the term (λ − r s ) is positive, the whole expression is minimized by
˜ s which is defined in the remainder
∆∗s = 0. If (λ − r s ) = 0, ∆∗s is set to a value ∆
of this section. Consequently the optimization problem implies finding a optimal
threshold parameter Cτ for λ such that



crit,s ,


∆t
˜ s,
∆∗s (Cτ ) = ∆



0,

(4.5)

if Cτ < r s
if Cτ = r s
otherwise

Based on the optimal solution ∆∗ , the complete interoperability transfer function evolves accordingly:

(4.6)

tITF,s :=



∑i∈τ ( j) ∑ j∈σ(s) pij + ∆tcrit,s ,




∑
˜s
i ∈τ ( j) ∑ j∈σ(s) pij + ∆ ,


∑i∈τ ( j) ∑ j∈σ(s) pij ,




0,

if eij ∈ o, Cτ < r s
if eij ∈ o, Cτ = r s
if eij ∈ o, Cτ > r s
otherwise

Service providers that have an incoming-edge ratio which equals the threshold (Cτ = r s ) and own service offers with allocated incoming edges, receive a part
of their critical value which depends on the number of service providers with
Cτ < r s , corresponding critical values and the number of service providers with
˜ s is defined as follows:
Cτ = r s . The value ∆

4.2. ACHIEVING BUDGET BALANCE

S∗ −

∆tcrit,s

∑
s∈S|Cτ

˜ s :=
∆

(4.7)

133

<r s

1

∑
s∈S|Cτ

=r s

4.2.3 Finding the Optimal Threshold Parameter
The threshold Cτ divides allocated service providers into two groups where one
gets a discount of ∆tcrit,s and the other 0. Let k denote the threshold index such
that if Cτ falls into the interval k such that Cτ ∈ [rτk+1 , rτk ) service providers 1, . . . k
(ordered increasingly based on their critical values) get their critical value while
service providers k + 1, . . . , I get no discount. Putting the solution ∆∗s (Cτ ) in the
Lagrangian problem z(Cτ ) leads to
(4.8)

I

z(Cτ , k ) =

(ri ∆tcrit,i ) + Cτ

∑

k

∑ ∆tcrit,i − S∗
i =1

i = k +1

!

The optimum is attained at
(4.9)

Cτ∗

k∗

= rk∗ +1 , ∑ ∆t

crit,i

i =1

∗

≤S ∧

k ∗ +1

∑

∆tcrit,i > S∗

i =1

Example 4.3 [A CHIEVING B UDGET B ALANCE (C ONTINUED )]. Recalling Example
4.2, this continuation illustrates how budget balance can be retained by implementing the
interoperability transfer function. In order to determine an optimal threshold parameter
Cτ , each service provider that owns allocated service offers is decreasingly ordered by
its incoming-edge-ratio r s . The number of possible edges within G is denoted by T =
10. Consequently, the incoming-edge-ratio r for service providers that own allocated
∑i∈σ(s ) τi

1
2
1
= 10
and r s4 = 10
. The vector of the ordered
service offers evolves as r s1 =
T
2 1
1
incoming-edge ratios is ( 15 , 10 ). Equation (4.9) is satisfied by Cτ∗ = 10
with k∗ = 2
∗
∗
which is the solution that satisfies the conditions ∑ik=1 ∆tcrit,i ≤ S∗ ∧ ∑ik=+1 1 ∆tcrit,i > S∗ .
˜ for service provider s1 is ∆
˜ s1 = 8−4 = 4. Payments for allocated service
The value ∆
1
ITF,s
1
offers evolve accordingly such that t
= 2 + 4 = 6 and t ITF,s4 = 2 + 4 = 6. As
U R = 12 − (6 + 6) = 0, the outcome of the extended complex service auction is budgetbalanced and does not have to be subsidized externally. It is important to notice that
the interoperability transfer function rewards service provider s4 for the high degree of
interoperability – i.e. the incoming-edge-ratio r s4 – which increases the variety of feasible
complex service compositions.

134

CHAPTER 4. APPLICABILITY EXTENSIONS

4.2.4 Summary
In summary, the ITF extension as a novel budget-balanced payment scheme
which satisfies Requirement 4 implements incentives for service providers to increase their services’ degree of interoperability which is shown in Section 6.2.2.
It is important to note that the incentives provided by the ITF are twofold:
First, the ITF limits strategic behavior of service providers which is shown in
Section 6.1. Second, the ITF rewards interoperability endeavors. Depending
on the design goals the payment scheme can be adjusted in order to calibrate
both effects. Introducing a calibration weight βITF ∈ [0; 1] and a threshold term
crit,s
r̃ s := βITF r s + (1 − βITF ) t ∆tcrit,s an adjustable interoperability transfer function
∑s∈S
evolves as follows:

(4.10)

t

ITF,s

:=



∑i∈τ ( j) ∑ j∈σ(s) pij + ∆tcrit,s ,




∑
˜ s,
p +∆
∑
i∈τ ( j)

j∈σ(s) ij



∑i∈τ ( j) ∑ j∈σ(s) pij ,




0,

if eij ∈ o, C̃τ < r̃ s
if eij ∈ o, C̃τ = r̃ s
if eij ∈ o, C̃τ ≥ r̃ s
otherwise

The computation of the optimal threshold parameter C̃τ is done analogously
to the procedure described in Section 4.2.3 accounting for r̃ s instead of r s . Thus,
βITF adjusts the transfer function with respect to both incentives. Higher values
for βITF result in stronger incentives for interoperability endeavors whereas lower
values provide stronger incentives to reduce strategic behavior.
With respect to the service level enforcement extension, the ITF can easily be
combined with the compensation function as introduced in Section 4.1. Service
providers that pass the threshold receive their critical value minus their compensation value. Note that in this case the computation of the optimal threshold
parameter has to be adjusted accordingly to assure budget balance.

4.3

Managing Service Quality

Recall that with the tremendous decrease of costs for the provision of highly scalable services, service providers shift from price to quality competition. QoS is
the key criterion to keep the business competitive as it has serious implications
on the provider and consumer side [Pap08]. Thus, an efficient management of

4.3. MANAGING SERVICE QUALITY

135

highly complex QoS characteristics is inevitable for service-oriented value creation in service value networks. In Section 3.2, the basic concept of QoS aggregation and evaluation has been described based on rather simple QoS attributes
such as response time, which are characterized by well-defined metrics to measure corresponding values.
In order to determine the overall score for a provider based on the scoring
function, the attribute values of the complex service have to be computed. The
type of operation for aggregating attribute value highly depends on the attribute
type. Basic quality of service attributes such as response time for example can
be aggregated with a sum operator. Table 3.1 shows different types of aggregation functions for multiple attribute types exemplarily. For example, the overall
throughput of a complex service that consists of multiple service components is
determined by the lowest throughput rate within the allocation and can therefore
be computed using a minimum operator.
Nevertheless, only considering basic quality of service attributes is not sufficient for dealing with complex non-functional service characteristics that express
rich semantic information. The auction mechanism must be capable of aggregating a broad range of descriptive service attributes that express multiple quality
aspects (e.g. the physical hosting location of a service and additional semantic information about the environment, a service’s usage policies or ownership rights)
. This section focuses on providing the conceptual foundations for a seamless
management of more sophisticated QoS characteristics, which require a semantic
understanding of their context and interrelations in order to measure and evaluate their particular occurrences.
To represent semantic knowledge about service quality attributes in an interoperable manner, ontologies are used to describe a conceptualization of service
characteristics and properties. The following definition is predominantly used in
the semantic Web community [SBF98].
Definition 4.1 [O NTOLOGY ]. An ontology is a formal explicit specification of a shared
conceptualization of a domain of interest.
In order to enable automatic processing and interpretation of explicit knowledge representations, adequate and machine-interpretable formalisms are used,
which are explained in the following section.

136

CHAPTER 4. APPLICABILITY EXTENSIONS

4.3.1 Knowledge Representation Formalisms
As a formalism to represent an ontology framework the Web Ontology Language
(OWL) is used. OWL is an ontology language standardized by the World Wide
Web Consortium (W3C) [MvH04] and is based on the description logic (DL) formalism [BCM+ 07]. Due to its close connection to DL it facilitates logical inferencing and allows to derive conclusions from an ontology that have not been stated
explicitly. As a brief introduction a review of some of the modeling constructs
of OWL using its DL-syntax is outlined here. The main elements of OWL are
individuals, properties that relate individuals to each other and classes that group
together individuals, which share some common characteristics. Classes as well
as properties can be put into subsumption hierarchies. Furthermore, OWL allows for describing classes in terms of complex class constructors that pose restrictions on the properties of a class. For example, the statement BigCity ⊑ ∃ isConnectedTo.Highway describes the class of big cities, which are connected to some
Highway. Subclass relationship can be expressed by a statement like BigCity ⊑
InterestingCity, saying that any big city is also interesting.
For the reader’s convenience, ontologies are illustrated in UML notation
where UML classes correspond to OWL concepts, UML associations to object properties, UML inheritance to sub-concept relations, UML dependencies
to OWL class instantiations and UML attributes to OWL datatype properties
[BVEL04].
To enable rule-like knowledge representation which is not supported by
the modeling primitives based on OWL-DL, the Semantic Web Rule Language
(SWRL) [HPSB+ 04] allows to extend OWL with Horn-like rules according to
first-order semantics. Additionally, SWRL provides an XML-based formalization,
which enables automatic processing of rule-based knowledge as an extension to
the OWL semantics. Furthermore SWRL allows for the implementation of algorithmic calculations such as mathematic operations and string comparison.

4.3.2 Semantic QoS Management
To foster a comprehensive management of QoS characteristics, the complex service auction is extended using concepts from Semantic Web research. Providing a broad contextual knowledge about attribute types, their conceptualization
and relations to other concepts in a machine-readable and interoperable manner, ontologies are used to capture relevant semantic information. Based on this
knowledge, individual attribute types can be expressed using a rule language

4.3. MANAGING SERVICE QUALITY

137

formalism. The following example demonstrates the expressiveness of a semantic approach towards the description of QoS characteristics and the expression of
individual requirements of requesters.
Example 4.4 [CSA WITH S EMANTIC Q O S M ANAGEMENT ]. For the reader’s convenience, the scenario is reduced to a minimal setting that is sufficient to illustrate the
strength of semantic service description and attribute aggregation. Figure 4.3 shows a
service value network with four service offers 1, 2, 3 and 4 and three feasible paths from
source to sink: f 1 = {es1 , e12 , e2 f }, f 2 = {es1 , e14 , e4 f } and f 3 = {es3 , e34 , e4 f }.

ps1 = 13

1
a1et = 1DES128

p12 = 16

a1ps = 0.9

Caption

2
v

a2et = 1RSA128

Service Offer

a2ps = 0.9

Composition
Relation

p14 = 17

s

3

ps 3 = 10

a3et = 1CFB128
a3ps = 0.9

f

s

Source Node

f

Sink Node

4
p34 = 20

a4et = 1RSA256
a4ps = 0.8

Figure 4.3
Service value network with semantic QoS characteristics.
For simplicity it is assumed that each service provider owns only a single service such
that σ (s1 ) = {1}, σ (s2 ) = {2}, σ (s3 ) = {3} and σ (s4 ) = {4}. Price values pij on the
edges represent price bids announced by service providers. Each service configuration
ps
A j consists of attribute values for encryption type aet
j and probability of success a j .
The attribute values in Figure 4.3 are assumed to be announced by each service provider
additionally to the corresponding price bid such that bij = ( A j pij ). Attribute values are
aggregated according to the aggregation operations in Table 3.1. Attribute values for
encryption type are derived from the concepts in the security algorithm ontology as
illustrated in Figure 4.4.
The security encryption ontology provides a brief conceptualization of encryption
types an their hierarchical classification in symmetric and asymmetric cipher methods.
Symmetric cipher methods are further divided into synchronous and self-synchronizing
stream ciphers and block cipher methods. Based on this semantic information about
different encryption types, the requester is capable of designing an individual attribute

138

CHAPTER 4. APPLICABILITY EXTENSIONS

EncryptionType
+hasKeyLength : int

SymmetricCipher

AsymmetricCipher

RSA
StreamCipher

BlockCipher

ECC

DES
SynchronousCipher

DSS

SelfSynchronizingCipher

TrippleDES
ElGamal
SFINKS

CFB
AES
Cramer-Shoup

ARC

Mosquito
Blowfish
Diffie-Hellman

Decim
IDEA

F-FCRS-8

Figure 4.4
Security encryption ontology.

type which incorporates the preferred encryption configuration. The following rules are
implementation-independently formulated in First-Order Logic (FOL) syntax.

(R1)

aie ←− EncryptionType( aet ), BlockCipher( aet ),

hasKeyLength( aet , k ), isGreaterOrEqual(k, 128)
(R2)

aie ←− EncryptionType( aet ), AsymmetricCipher( aet ),

hasKeyLength( aet , k ), isGreaterOrEqual(k, 256)

4.3. MANAGING SERVICE QUALITY

139

In this example the requester specifies an attribute type ie ∈ L representing individual encryption. This attribute type is defined by Rule (R1) and Rule (R2). If a single
rule fires, the boolean attribute value aie is set to true, meaning that the service offer
satisfies the individual encryption requirements expressed by the requester.
Assuming a requester’s maximum willingness to pay for a complex service with a
score of 1 is α = 100 and preferences for attribute types individual encryption and
probability of success are λie = 0.2 and λ ps = 0.8, the overall utility of each feasible
path evolves as follows

U f 1 = 100 × (0.2 × (1 ∧ 0) + 0.8 × (0.9 × 0.7)) − (13 + 16) = 21.4
U f 2 = 100 × (0.2 × (1 ∧ 1) + 0.8 × (0.9 × 0.8)) − (13 + 17) = 47.6
U f 3 = 100 × (0.2 × (0 ∧ 1) + 0.8 × (0.9 × 0.8)) − (10 + 20) = 27.6
As the complex service instance f 2 yields the highest overall utility, service offers 1
and 4 via edges es1 , e14 and e4 f are allocated by o ( B). Thus, service providers s1 and
s2 receive a transfer according to the transfer function in Equation (3.10) based on their
critical value.

ts1 = t1s1 = 13 + (47.6 − 27.6) = 33
ts4 = t4s4 = 17 + (47.6 − 21.4) = 43.2
Consequently the service requester’s utility evolves as

U R = 100 × (0.2 × (1 ∧ 1) + 0.8 × (0.9 × 0.8)) − (33 + 43.2) = 1.4
In summary, the integration of rule-based semantic description techniques allows for the specification, aggregation and management of highly complex QoS
characteristics which satisfies Requirement 7.

Part III
Evaluation

Chapter 5
Analytical Results
[...] the set of incentive-compatible direct-revelation mechanisms has simple
mathematical properties that often make it easy to characterize, because can be defined by
a set of linear inequalities.
[Mye88]

his chapter thoroughly analyzes the economic properties of the complex service auction and their extensions as introduced in Chapter 3. Section 5.1
analytically shows that the complex service auction with the service level enforcement extension implements a strategyproof social choice, i.e. reporting ones
true multidimensional type is an equilibrium in weakly dominant strategies. Focusing on cooperative behavior of adjacent service providers in service value networks, Section 5.2 studies the effect of interface customization and implicit cost
reductions for preceeding or succeeding services within service value networks.

T

5.1 Incentive Compatibility & Individual Rationality
Recalling Section 2.2.4, incentive compatibility is a valuable property to be
achieved in mechanism design. In decentralized environments such as service value networks with self-interested participants that have private information about their preferences for different outcomes, solving a global optimization problem fully depends on how participants can be incentivized to report
their private information to the auctioneer in a truthful manner. This information is needed to compute e.g. an allocative efficient outcome in such a setting.

144

CHAPTER 5. ANALYTICAL RESULTS

Hence, incentive compatibility can be seen as a necessary precondition in order to
achieve a welfare maximizing outcome in scenarios with incomplete information.
Another major beneficial result that derives from truthfulness is that it tremendously simplifies the strategy space of participants as they do not have to reason about strategies of other participants. Thus, incentive compatibility reduces
the participants’ strategy space and simplifies their decision problem to a single
weakly dominant strategy maximizing their individual utility.
The remainder of this section analytically shows that in the basic complex service auction (without the compensation function extension), bidding ones true
valuations for all offered services is a weakly dominant strategy for all participating service providers (Section 5.1.1). Based on these results, Section 5.1.2
shows that in the complex service auction with the service level enforcement
extension (cp. Section 4.1), bidding true valuations and true QoS characteristics
for all offered services is a weakly dominant strategy for all participating service
providers which satisfies Requirement 2. Based on the results regarding truthfulness it is briefly shown that service providers always end up with a payoff
equal to or greater than zero which satisfies individual rationality as stated in
Requirement 3.

5.1.1 One-Dimensional Bids in the Basic CSA
This section is concerned with strategic behavior in the basic complex service auction, i.e. the basic mechanism implementation without the compensation function extension which enables service level enforcement. The following analytical
evaluation of the mechanism implementation with respect to service providers’
bidding strategy considers price bids only in the first place. Thus, the providers’
strategy space is reduced to announcing prices for each incoming edge of each
service offer they own.
First, Corollary 5.1 shows that once a service provider is allocated – that is, the
service provider owns service offers that have at least one incoming edge which
is allocated by the mechanism – its payoff is independent of its bidding strategy.
This means that once a service provider is allocated it is indifferent between any
alternative bidding strategy within its strategy space.
Consequently, the only event that service providers can actively influence by
their bidding strategy is whether they are allocated by the mechanism or not.
Based on the results of Corollary 5.1, Theorem 5.1 considerers the cases in which
service providers intent to be allocated and derives the optimal bidding strategy:

5.1. INCENTIVE COMPATIBILITY & INDIVIDUAL RATIONALITY

145

Service providers act best (or at least equally good) by following a truth-telling
strategy, i.e. reporting their true valuations – which are assumed to be reflected
by corresponding internal costs – for each service offer is a weakly dominant
strategy for all service providers that participate in the complex service auction.
Corollary 5.1. For each service provider s ∈ S that participates in the complex service
auction, the transfer ts is independent of its price bid. More precisely this means that for
each service offer j ∈ V owned by s ∈ S with an incoming edge which is allocated by o
such that eij ∈ o with j ∈ σ (s) and i ∈ τ ( j), service provider s’s payoff is independent of
its price bid pij .
Proof 5.1 [C OROLLARY 5.1]. Let F−s denotes the set of all feasible paths from source
to sink in the reduced graph G−s without every service offer owned by service provider s
and corresponding incoming and outgoing edges. Let further f ∗ denote the path which
∗ be the utility of path f ∗ in
is allocated by o. Let U ∗ be the utility of path f ∗ . Let U−
s
−s
s
s
the reduced graph G−s . Let Ẽ denote the set of edges with Ẽ = {eij |eij ∈ o, j ∈ σ (s), i ∈
τ ( j)}. Distinguishing two possible cases, service provider s’s payoff π s evolves as follows.
1. Ẽs = ∅. Service provider s is not allocated. More precisely, none of the incoming
edges of service offers owned by service provider s are allocated by o.
It follows directly that in this case π s = 0 independent of s’s price bid.
2. Ẽs 6= ∅. Service provider s is allocated. More precisely, at least one of the incoming
edges of service offers owned by service provider s is allocated by o.
π s = ts − cs
πs =

∑ pij + (U ∗ − U−∗ s ) − ∑ cij
Ẽs

π

s

π

s

=

Ẽs

∑ pij + αS(A f ∗ ) − ∑

eij ∈o

Ẽs

(5.1)

= αS(A f ∗ ) −

∗
pij − U−
s

∑
eij |eij ∈o,eij

∗
pij − U−
s − ∑ cij

∈
/ Ẽs

Ẽs

− ∑ cij
Ẽs

This shows that for each service offer j owned by s that has an incoming edge eij
which is allocated by o – otherwise s does not receive a transfer – the corresponding profit
is independent of s’s price bid pij .

Theorem 5.1. For each service provider s ∈ S that participates in the complex service
auction, the price bidding strategy pij = cij (truth-telling) ∀i ∈ τ ( j), ∀ j ∈ σ (s) is a weakly
dominant strategy.

146

CHAPTER 5. ANALYTICAL RESULTS

Proof 5.1 [T HEOREM 5.1]. Corollary 5.1 shows that the transfer ts for each service
provider s ∈ S is independent of the price bid. Consequently, the only event that s can
proactively influence by its bidding strategy is whether its service offers are allocated
by o or not. Let Ẽs = {eij |eij ∈ o, j ∈ σ (s), i ∈ τ ( j)} denote the set of incoming edges
of service offers owned by service provider s that are allocated by o. Service provider
s wants incoming edges of service offers that s owns to be allocated by o (Ẽs 6= ∅) iff
π s > 0. Hence, service provider s wants the following equivalence1 to be fulfilled through
an adequate choice of its price bid.
Ẽs 6= ∅
(5.2)

⇐⇒ U ∗ > U−∗ s

⇐⇒ π s > 0

U ∗ − U−∗ s > 0 ⇐⇒

∑ ( pij − cij ) + (U ∗ − U−∗ s ) > 0
Ẽs

Equation (5.2) holds for pij = cij ∀ j ∈ σ (s), i ∈ τ ( j). According to Corollary 5.1, if
Ẽs 6= ∅, s is indifferent between any other solution that satisfies Equation (5.2) which
means that reporting true internal costs is a weakly dominant price bidding strategy for
service provider s.


5.1.2 Multidimensional Bids in the Extended CSA
The analytical evaluation of service providers’ bidding strategies in this section is
conducted analogously to the one-dimensional case. Nevertheless, the following
evaluation accounts for the complete strategy space of service providers, i.e. service providers announce multidimensional bids consisting of a price and QoS component for each incoming edge of every service offer they own within the service
value network. The analysis is based on the complex service auction mechanism
with the compensation function extension (cp. Section 4.1) which implements a
service level enforcement component.
Laying the groundwork for Theorem 5.2, Corollary 5.2 shows that once a service provider is allocated, its payoff is independent of its announced price and
corresponding attribute values which characterize guaranteed QoS. This means
that once a service provider is allocated it is indifferent between any alternative
bidding strategy within its strategy space with respect to all dimensions of its bid.
However, the service providers’ bid (price and attribute values) influences
the chance of being allocated by the mechanism. Based on the results of Corollary 5.2, Theorem 5.2 considerers the cases in which service providers intent to
1 Two

statements are equivalent as denoted by ⇐⇒ if and only if both statements yield the
same outcome for every possible interpretation.

5.1. INCENTIVE COMPATIBILITY & INDIVIDUAL RATIONALITY

147

be allocated and derives the optimal bidding strategy. Theorem 5.2 shows that
service providers act best (or at least equally good) by reporting their true multidimensional type, i.e. reporting their true valuations and guaranteed QoS for
each service offer regarding its predecessor is a weakly dominant strategy for all
service providers that participate in the extended complex service auction.
Corollary 5.2. For each service provider s ∈ S that participates in the complex service
auction with the compensation function extension (cp. Section 4.1), the transfer ts is
independent of all dimensions of s’s bids (configuration and price). This means that for
each service offer j ∈ V owned by s ∈ S that has an incoming edge which is allocated by o
such that eij ∈ o with j ∈ σ (s) and i ∈ τ ( j), service provider s’s payoff is independent of
all dimensions of its bid bij = ( A j , pij ).
Proof 5.2 [C OROLLARY 5.2]. Let F−s denote the set of all feasible paths from source to
sink in the reduced graph G−s without every service offer owned by service provider s
and corresponding incoming and outgoing edges. Let further f ∗ denote the path which
∗ be the utility of path f ∗
is allocated by o. Let U ∗ be the utility of path f ∗ . Let U−
s
−s
∗
s
in the reduced graph G−s . Let Ũ denote the overall utility of the allocated path f ∗
computed based on the verified attribute values ã1j , . . . , ã Lj of the verified configurations
à j of all service offers j ∈ σ (s). Let Ẽs denote the set of edges with Ẽs = {eij |eij ∈ o, j ∈
σ (s), i ∈ τ ( j)}. Distinguishing two possible cases, service provider s’s payoff π s evolves
as follows.

1. Ẽs = ∅. Service provider s is not allocated. More precisely, none of the incoming
edges of service offers owned by service provider s are allocated by o.
It follows directly that in this case π s = 0 independent of s’s price bid.
2. Ẽs 6= ∅. Service provider s is allocated. More precisely, at least one of the incoming
edges of service offers owned by service provider s is allocated by o.
π s = ts − cs
πs =

∑ pij + (U ∗ − U−∗ s ) − tcomp,s − ∑ cij
Ẽs

π

s

=

∑ pij + (U

∗

− U−∗ s ) − (U ∗

Ẽs
∗s

− Ũ ) − ∑ cij

Ẽs

π

s

=

∑ pij + (Ũ

Ẽs

∗s

− U−∗ s ) −

(5.3)

π

s

=

αS(Ãsf ∗ ) −

∑ cij
Ẽs

Ẽs

∑
eij |eij ∈o,eij ∈
/ Ẽs

∗
pij − U−
s − ∑ cij
Ẽs

148

CHAPTER 5. ANALYTICAL RESULTS

Equation (5.3) shows that for each service offer j ∈ V owned by s ∈ S that has an incoming
edge which is allocated by o such that eij ∈ o with j ∈ σ (s) and i ∈ τ ( j), service provider
s’s payoff is independent of all dimensions of its bid bij = ( A j , pij ).

Theorem 5.2. For each service provider s ∈ S that participates in the complex service
auction with the compensation function extension (cp. Section 4.1), the bidding strategy
bij = ( à j , cij ) with à j = ( ã1j , . . . , ã Lj ) – truth telling with respect to all dimensions of the
bid – ∀i ∈ τ ( j), ∀ j ∈ σ (s) is a weakly dominant strategy.
Proof 5.2 [T HEOREM 5.2]. Let F−s denote the set of all feasible paths from source to
sink in the reduced graph G−s without every service offer owned by service provider s
and corresponding incoming and outgoing edges. Let further f ∗ denote the path which
∗ be the utility of path f ∗
is allocated by o. Let U ∗ be the utility of path f ∗ . Let U−
s
−s
in the reduced graph G−s . Let Ũ ∗s denote the overall utility of the allocated path f ∗
computed based on the verified attribute values ã1j , . . . , ã Lj of the verified configurations
à j of all service offers j ∈ σ (s). Let Ẽs denote the set of edges with Ẽs = {eij |eij ∈ o, j ∈
σ (s), i ∈ τ ( j)}. Corollary 5.2 shows that the transfer ts for each service provider s ∈ S is
independent of all dimensions of its bid. In other words, s’s bid does not have an impact on
its transfer ts and its payoff π s respectively. Nevertheless, the bidding strategy influences
service provider s’s chance of being allocated by o. Thus, s wants to be allocated iff π s > 0.
Ẽs 6= ∅

⇐⇒ U ∗ > U−∗ s
U ∗ > U−∗ s

⇐⇒ π s > 0
⇐⇒

∑ pij + (Ũ ∗s − U−∗ s ) − ∑ cij > 0
Ẽs

(5.4)

U ∗ > U−∗ s

⇐⇒

∑ pij + Ũ ∗s > ∑
Ẽs

Ẽs
∗
cij + U−
s

Ẽs

Equation (5.4) holds for pij = cij and U ∗ = Ũ ∗s . According to Corollary 5.2, if Ẽs 6=
∅, s is indifferent between any other solution that satisfies Equation (5.4) which means
that reporting attribute values a1j , . . . , alj truthfully meaning that the announced values
equal the verified ones in the execution phase such that alj = ãlj ∀l ∈ L, ∀ j ∈ σ (s) and
consequently U ∗ = Ũ ∗s is a weakly dominant strategy.

The analytical proof in Section A.2 evaluates service providers’ bidding strategies from the perspective of the providers’ expected payoff which they intent to
maximize. Analogue to the previous result, it turns out that there exists a single
bidding strategy that maximizes service providers’ expected payoff.

5.1. INCENTIVE COMPATIBILITY & INDIVIDUAL RATIONALITY

149

5.1.3 Results & Implications
Theorem 5.2 shows that service providers act best (or at least as good as any other
alternative) by reporting their services’ configurations and internal costs truthfully which is a valuable mechanism property as it enables the computation of an
optimal welfare maximizing outcome although the scenario is predominated by
incomplete information. This property assures that although all service providers
act self-interested and therefore try to maximize their profit, their dominant strategy maximizes the system’s welfare and the requester receives a technically feasible instantiation of the desired complex service at a guaranteed service level2 . The
presence of a single beneficial strategy tremendously lowers strategic complexity
for service providers and fosters a trustful requester-provider-relationship. The
results at hand show that the extended complex service auction satisfies Requirement 2. It is straightforward to see that with the results of Theorem 5.2, participating service providers always end up with a payoff equal to or greater than
zero which satisfies individual rationality as stated in Requirement 3. In other
words, service providers have an incentive to participate in the complex service
auction without running into the risk of being worth of than their outside option.
Furthermore, it follows directly form Corollary A.1 that Requirement 1 is satisfied
through the social choice implemented by the complex service auction.
It is well-known in literature that incentive compatibility in VCG-based mechanisms may fail in repeated games [BS00]. Assuming that participants are able
to gather historic information about previous outcomes, deviation from truthtelling might be beneficial in certain situations and the theoretical results from
this section might not hold. However, in service value networks through a high
degree of alteration with respect to changing service providers, variable costs
and network topologies is observable. As outlined in Section 2.1.4, the complex
service auction is designed for scenarios with fast changing participants that together foster value creation which satisfies situational needs. Thus, each auction
setting is different from the preceding one which makes learning from past situations impossible and each game can therefore be treated as a one-shot game. For
a simulation-based analysis of collusion behavior in the complex service auction,
the interested reader is referred to [CvD09].

2 Despite

of service level agreement violations caused by events which are not under the control of service providers.

150

5.2

CHAPTER 5. ANALYTICAL RESULTS

Cooperation within the Value Chain

This section studies a special form of cooperation in the context of the complex
service auction in service value networks. Traditionally in social network research, the creation of links connecting players requires a cooperative process
such that both participants have to agree to a connection. Removing links, however, is a non-cooperative act as it can be done unilaterally by a single player
within the network. In the context of service value networks where service components’ input and outputs are plugged together realizing a value-added complex service, service providers have the strategic opportunity to customize their
service offers in a way that they are interoperable with predecessor services. This
form of establishing a feasible connection to another component within the network is – in contrary to traditional social network theory – unilateral and noncooperative. Predecessor services cannot control which successor service creates
a connection by postprocessing its output.

5.2.1 Related Work
In [JW96] the evolution of social and economic networks where self-interested
individuals form or sever links is analyzed. In [JW02] network formation is
founded upon players’ individual improvements resulting from changes in the
network topology. Traditionally, breaking relationships can be done unilaterally
while the formation of links requires consent from both players [JW96]. In [BG00],
however, links can be formed by individual decision under certain circumstances.
This is also the case in service value networks since service providers cannot influence which other services process their outputs.

5.2.2 A Model of Cooperation
In a service value network with four service offers a, b, y, z are two particular service offers y ∈ V and z ∈ V that are owned by two different service providers
sy ∈ S and sz ∈ S. Based on the topology of the Service Value Network y is the
predecessor of z connected by an edge eyz . Costs that service provider sz has to
bear for its service z being executed as a successor of service y are denoted by cyz .
Furthermore it is assumed that service provider sy has the strategic opportunity to invest an amount I in order to customize its service offer y in a way that
H to c L with c H > c L . As s
costs cyz of service provider sz are reduced from cyz
y
yz
yz
yz

5.2. COOPERATION WITHIN THE VALUE CHAIN
y

cyz

151

z

f

s

a

b

Figure 5.1
Cost dependency between service provider sy and sz .

is familiar with its internal processes and properties of its service offer y, proportionate investment costs I are less then the effect of cost reduction for sz such that
H − c L . Focusing on one-shot games, incorporating total fix costs for service
I < cyz
yz
customization in order to reduce variable costs caused by the preceeding service
is not reasonable. Therefore I constitutes proportionate investment costs as a fraction of the total fix costs for a particular auction conduction. The assumption is
that these proportionate investment costs are less than the reduction in variable
costs caused by the preceeding service.
Corollary 5.3 [C OOPERATION WITHIN THE VALUE C HAIN ]. Given two service
providers sy and sz that own service offers y and z with y being the predecessor service of z. Furthermore let Θyz be an enforceable ex-ante agreement that states that iff
services y and z are allocated such that eyz ∈ f ∗ then service provider sy is committed to
H to c L . Committing to an agreement Θ is
invest I in order to reduce costs cyz from cyz
yz
yz
H
L
an equilibrium in weakly dominant strategies if I ≤ cyz − cyz .
Proof 5.3 [C OROLLARY 5.3]. Let U ∗ H (eyz ) be the overall utility of the path allocated
H . Analogously let U ∗ L ( e ) be the overall utility of
by o that entails edge eyz and costs cyz
yz
L
∗ be the overall
the path allocated by o that entails edge eyz and costs cyz . Let further U−
sy
utility of the path allocated by o in the reduced graph without node y and all its incoming
and outgoing edges. Service offer i is an arbitrary predecessor of y.
The expected payoff of service provider sy under the assumption that there is no agreement Θyz evolves as follows
i
h
∗
∗H
∗
comp,sy
Esy = P(U ∗ H (eyz ) > U−
)
p
+
(U
−
U
)
−
∆t
−
c
iy
iy
sy
−sy
With the results of Theorem 5.2 that each service provider reports its type truthfully the
equation can be simplified to
E

sy

= P(U

∗H

(eyz ) >

U−∗ sy )

h

U

∗H

− U−∗ sy

i

152

CHAPTER 5. ANALYTICAL RESULTS

Analogously for service provider sz
i
h
∗
∗H
∗
Esz = P(U ∗ H (eyz ) > U−
)
U
−
U
sz
−sz
Assuming that sy and sz commit to the agreement Θyz expected payoffs evolve as follows
(5.5)
(5.6)

h
i
sy
∗
∗L
∗
)
U
−
U
−
I
EΘyz = P(U ∗ L (eyz ) > U−
sy
−sy
i
h
sz
∗L
∗
∗L
∗
EΘyz = P(U (eyz ) > U−sz ) U − U−sz

In order to be an equilibrium in weakly dominant strategies, the commitments θy and θz
to agreement Θyz must be a weakly dominant strategy for service provider sy and sz . The
strategy space of each service provider and corresponding expected payoffs are illustrated
as a normal form game in Table 5.1.
Table 5.1: Cooperation decision as a normal form game. θ denotes an ex-ante commitment to an agreement Θ whereas θ̄ states
the decision not to commit to an agreement Θ.

y,z

θ

θ̄

θ

sz
EΘyz , EΘ
yz

sy

E sy , E sz

θ̄

E sy , E sz

E sy , E sz

sy

sz
≥
The strategy θ is a weakly dominant strategy for each player if EΘyz ≥ Esy and EΘ
yz
E sz .
H > c L and the quasi-linearity of U it follows that
Based on the assumption that cyz
yz
∗
H
∗
L
U (eyz ) < U (eyz ). Consequently the probability of service offer y being allocated by o
∗ ) < P (U ∗ L ( e ) > U ∗ ).
increases if sy follows strategy θy such that P(U ∗ H (eyz ) > U−
yz
sy
−sy
If investment costs I for service provider y are lower (or at least equal) compared to the
H − c L for service provider z it can be derived that U ∗ H − I ≤ U ∗ L .
cost reduction cyz
yz
sy
Finally it can be concluded that EΘyz ≥ Esy .

As the service provider sz can only benefit from a cost reduction the same argumenta∗ ) < P (U ∗ L ( e ) > U ∗ ), U ∗ H < U ∗ L and directly to
tion leads to P(U ∗ H (eyz ) > U−
yz
sz
−sz
sz
s
z
EΘyz > E .

Example 5.1 [C OOPERATION WITHIN THE VALUE C HAIN ]. To illustrate Corollary
5.3 and its implications for cooperative behavior in service value networks, Example 2.1

5.2. COOPERATION WITHIN THE VALUE CHAIN

153

is consulted. For the reader’s convenience the complex service is reduced to the first two
business transactions, data verification and transaction processing. Figure 5.2 shows the
service value network with service offers and corresponding costs. Each feasible path from
s to f represents a possible instantiation of the payment processing complex service. Data
verification can be performed by either StrikeIron (sy ) and its service offer y or Duo Share
(s a ) offering service a. The execution of the actual monetary transaction can be done by
JETTIS (sz ) offering service z or service b offered by Net Billing (sb ).
y

8−x

z

2
f

s
1
a

10

b

Figure 5.2
Cooperation within the value chain of a payment processing
complex service.
A mandatory step for a transaction processing service is the credit assessment. As a
precondition, a transaction processing service has to check if the customer is credit worthy
in order to charge the corresponding account. The credit assessment has to be performed
at a central authority and generates variable costs each time the transaction processing
service is executed. The predecessor service that verifies the customer’s data has to consult
the same central authority to assure the correctness of processed data.
The provider of the data verification service has the strategic opportunity to customize
its internal process in a way that a credit assessment is done on the fly which is cheaper
than doing it in the second transaction. In other words if service provider sy agrees to bear
proportionate investment costs of I ∈ R+ with I ≤ x to customize its internal process in
order to enable credit assessment in case of eyz being allocated, service provider sz can
reduce its costs by x ∈ R+ .
To analyze the effect of such an agreement Θyz according to Corollary 5.3 two cases
are considered:
1. There is no conclusion to agreement Θyz such that x = 0
The top path f T consisting of service offer y and z such that f T = {esy , eyz , ez f } generates a welfare of U f T = α − 10 whereas the bottom path f B = {esa , eab , eb f } generates a welfare of U f B = α − 11. Consequently service offers y and z are allocated
by o such that f ∗ = {esy , eyz , ez f }. Each service provider that owns a service that is

154

CHAPTER 5. ANALYTICAL RESULTS
allocated receives its transfer. Service provider sy is payed tsy = 2 + (11 − 10) = 3
and sz gets tsz = 8 + (11 − 10) = 9. This leads to a payoff for provider sy of
π sy = 1 and for service provider sz of π sz = 1. The requester’s utility evolves as
U R = α − 12.

2. Both parties agree on Θyz such that costs for sz are reduced by x
In this case the top path f T consisting of service offer y and z such that f T =
{esy , eyz , ez f } generates a welfare of U f T = α − 10 + x whereas the bottom path
f B = {esa , eab , eb f } generates a welfare of U f B = α − 11. Analogue to the first
case, service offers y and z are allocated by o such that f ∗ = {esy , eyz , ez f }. Service
provider sy is payed tsy = 2 + (11 − 10 + x ) = 3 + x and sz gets tsz = 8 − x +
(11 − 10 + x ) = 9. This leads to a payoff for provider sy of π sy = 1 + x and for
service provider sz of π sz = 1. The requester’s utility evolves as U R = α − 12 − x.
The example shows that it is beneficial (or at least equally good) for adjacent service
sy
providers to commit to an agreement according to Corollary 5.3 as πcase 1 = 1 ≤
sy
sz
sz
πcase 2 = 1 + x − I and πcase
1 = 1 ≤ πcase 2 = 1.

Chapter 6
Numerical Results
In economic applications the analytical apparatus [...] diminishes the economic content
of the models.
[KV98]
his chapter analyzes properties of the complex service auction and their extensions as well as strategic behavior of service providers by means of a
simulation-based evaluation. In Section 6.1, the interoperability transfer function (ITF) is analyzed with respect to manipulation attempts of service providers
that deviate from their truth-telling strategy. The question is answered to what
degree bid manipulation is beneficial for service providers given different levels of competition in service value networks. Based on these results, Section 6.2
evaluates the incentives provided by the ITF which fosters interoperability endeavors of service providers, i.e. the ITF provides incentives for service providers
to customize their services’ interfaces to increase interoperability with adjacent
service components. Focusing on bundling and unbundling strategies of service providers, Section 6.3 analyzes strategic behavior by means of an agentbased simulation. Based on these results strategic recommendations for service
providers are derived depending on how they are situated within service value
networks.

T

6.1 Manipulation Robustness of the ITF Extension
This section considerers strategic behavior of service providers participating in
the complex service auction with the interoperability transfer function (ITF). Re-

156

CHAPTER 6. NUMERICAL RESULTS

call, in the basic complex service auction, allocated service providers are payed
their price bid plus their critical value compensating their contribution to the
whole system. This critical value is designed to implement a dominant strategy
equilibrium in which every service provider reports its multidimensional type
truthfully to the auctioneer according to Theorem 5.2.
Nevertheless, incentive compatibility comes at the price of losing budget
balance, i.e. the sum of service providers’ transfers may exceed the service requester’s willingness to pay which results in a negative budget that has to be
subsidized externally. As a possible remedy to retain budget balance, the ITF extending the basic complex service auction was introduced in Section 4.2. The ITF
distributes the available surplus – the difference between the service requester’s
willingness to pay and the sum of providers’ transfers – in a way that additionally to their bid, allocated providers are payed their critical value in the priority
of their degree of interoperability subject to budget balance. It is obvious that in
order to recover budget balance, incentive compatibility has to be sacrificed to
a certain degree. Incurring this trade-off, the set of possibly beneficial bidding
strategies of service providers increases and from a pure analytical perspective
Theorem 5.2 does not hold under the presence of the ITF extension. Although the
primary goal from an incentive engineering perspective of the ITF is to reward
interoperability endeavors, the design of the ITF gives a good indication that bid
manipulation is only beneficial to a certain level which strongly depends on the
level of competition [Jac92, RP76, Hur72].
This section analyzes strategic behavior of service providers in the complex
service auction with the ITF extension following a simulation-based approach
(cp. Section 2.3.2).

6.1.1 Simulation Model
To analyze the manipulation robustness of the complex service auction with the
ITF extension, a simulation is conducted as follows. A random service value
network topology is created with density 1.0 (complete graph) and – depending
on the degree of competition – with a predefined number of service offers and
candidate pools. For simplicity and without loss of generality it is assumed that
each service provider owns only a single service offer within the service value
network. The competition rate results from the number of alternative complex
service instances (number of feasible paths) without the participation of a single
service provider. The number of feasible paths depends on the number of service

6.1. MANIPULATION ROBUSTNESS OF THE ITF EXTENSION

157

offers within the network, the number of candidate pools and the density of the
graph, i.e. the ratio between the number of edges and the number of all possible edges in the graph. The ratio between the number of service offers and the
number of candidate pools is also responsible for the number of possible service
compositions.
Each problem set is characterized by a random network topology with random costs cij assigned to each incoming edge of service offers drawn from
U (0, 1.0). Furthermore, the requester’s willingness to pay α is analogously drawn
from U (0, 12 K )1 with K being the number of candidate pools.
For each problem set, a random service offer’s incoming edge eij is randomly
drawn. The bid price pij is manipulated stepwise from 50% to 150% in steps of
10% of the truth-telling price pij = cij . For each manipulation rate the auction
is conducted and the service provider’s utilities for the deviation and the truthtelling strategies are computed based on the critical value transfer function and
the ITF. Figure 6.1 depicts the stepwise procedure of the simulation.
Generation of random topology. Assignment of random edge costs and requester’s willingness to pay.

Random selection of a service
offer. Random selection of an
incoming edge eij

Deviation from truth-telling
strategy by manipulation rate
mr

Computation of absolute
utility for truth-telling and
deviation strategies

pij = cij (1 + mr )
Increase of manipulation rate

Figure 6.1
Simulation model for the evaluation of manipulation robustness
using the ITF.
As the number of variable parameters and their interdependencies are high,
heavy statistical noise is likely to be generated. To counteract the high volatility of the simulation model, a large number of problem sets of 5000 is evaluated
for each degree of manipulation and the mean results are reported. In order to
identify the degree of manipulation for which a deviation from the truth-telling
strategy is beneficial for service providers, the statistical significance is tested using a one-tailed matched-pairs t-test analyzing the alternative hypothesis that
service providers benefit from manipulation, that is, the mean difference in utility is greater than zero. The large size of analyzed problem sets for each obser11K
2

denotes the mean price of a complex service in a network with K candidate pools and
internal costs of service providers drawn from U (0, 1.0) under the presence of truth-revelation.

158

CHAPTER 6. NUMERICAL RESULTS

vation assures robustness of the t-test to violations of the normality assumption
[SB92, BS99, Ram80].

6.1.2 Results
Participating in the complex service auction with the ITF extension, service
providers’ strategies and corresponding outcomes are illustrated in Figure 6.2.
The decision tree evaluates possible bidding strategies in comparison to a truthtelling strategy. Focusing on a single service provider, two fundamental cases
must be considered in order to evaluate the result of different strategies:
1. Having followed a truth-telling strategy, the service provider s would have
been allocated by o.
In this case, overstating the true valuation by announcing a price p̃ij > cij
leads to a payoff π̃ s ≥ π s if the service providers stays allocated and to a
payoff π̃ s < π s if it is dropped out of the allocation. The monotonicity of
the allocation function assures that the service provider still gets allocated
by understating the true valuation such that p̃ij < cij which leads to a payoff
π̃ s ≤ π s .
2. Having followed a truth-telling strategy, the service provider s would not
have been allocated by o.
In this case, by overstating the true valuation announcing a price p̃ij > cij ,
the service provider is not allocated due to monotonicity of the allocation
function which leads to a payoff π̃ s = π s . Understating the true valuation
results in a payoff π̃ s < π s if the service provider gets allocated and to a
payoff π̃ s = π s if it is not allocated.

The effect of a bid manipulation strategy of service providers is highly dependent on the level of competition in the service value network as this increases the
risk of dropping out of the allocation by overstating ones true valuation. As market size increases, participants become price takers and strategic considerations
converge towards a truth-telling strategy [Jac92, RP76, Hur72]. In the complex
service auction, the level of competition results from the number of alternative

6.1. MANIPULATION ROBUSTNESS OF THE ITF EXTENSION

p̃ij > cij

eij ∈ o
m

m
p̃ij > cij

π̃ s ≥ π s

eij 6∈ o

π̃ s < π s

m

s
p̃ij < cij

eij ∈ o

m

eij ∈ o
eij 6∈ o

eij 6∈ o
s
p̃ij < cij

159

π̃ s ≤ π s
π̃ s = π s

eij ∈ o

π̃ s < π s

eij 6∈ o

π̃ s = π s

m

Figure 6.2
Decision tree of service providers.
paths in the absence of a single service provider. Therefore a good indication for
the level of competition can be derived from the number of feasible paths in the
network2 . The lower the level of competition, the higher the benefit for service
providers that deviate from their truth-telling strategy.
Table 6.1 shows the utility of a singe manipulating service provider in a low
competition setting with 12 service offers in 4 candidate pools. Understating
one’s true valuation results in a negative utility gain compared to a truth-telling
strategy. However, service providers that overstate their true valuation significantly benefit from a deviation up to 100% of their true valuation.

2 Based

on the service value network model in Section 2.1.4, the number of feasible paths
depends on the number of candidate pools and service offers per candidate pool. Assuming an
 |V \{v ,v }| K
s f
, where K denotes
equal number of service offers per pool, the number of paths is
K
the number of candidate pools.

160

CHAPTER 6. NUMERICAL RESULTS
Table 6.1: Utility for a single manipulating service provider with
12 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-50%

0.0423

0.5865

0.0793

-0.0209

-0.6871

0.1022

-40%

0.0562

0.7789

0.0506

-0.0009

-0.0308

0.0714

-30%

0.0631

0.8741

0.0334

0.0113

0.3645

0.0478

-20%

0.0693

0.9603

0.0136

0.0194

0.6763

0.0264

-10%

0.0715

0.9904

0.0050

0.0250

0.8795

0.0144

0%

0.0722

1.0000

0.0000

0.0302

1.0000

0.0000

10%

0.0715

0.9906

0.0050

0.0317

1.0688***

0.0125

20%

0.0705

0.9771

0.0097

0.0327

1.0968***

0.0199

30%

0.0703

0.9738

0.0102

0.0393

1.1380***

0.0283

40%

0.0696

0.9638

0.0137

0.0384

1.1776***

0.0355

50%

0.0673

0.9320

0.0261

0.0379

1.1774***

0.0435

60%

0.0640

0.8870

0.0383

0.0384

1.1016***

0.0445

70%

0.0627

0.8691

0.0424

0.0377

1.0866***

0.0486

80%

0.0603

0.8354

0.0508

0.0355

1.0535***

0.0449

90%

0.0596

0.8251

0.0521

0.0362

1.0233*

0.0475

100%

0.0591

0.8181

0.0533

0.0351

1.0581***

0.0508

110%

0.0578

0.8006

0.0560

0.0378

1.0091

0.0537

120%

0.0554

0.7670

0.0632

0.0354

0.9652

0.0524

130%

0.0550

0.7613

0.0639

0.0314

0.9824

0.0543

140%

0.0534

0.7395

0.0672

0.0317

0.9529

0.0576

150%

0.0526

0.7285

0.0685

0.0344

0.9557

0.0581

A marginal increase in the level of competition decreases the number of beneficial manipulation strategies. Table 6.2 shows the simulation results in a setting
with 16 service offers in 4 candidate pools. The utility of a single manipulating

6.1. MANIPULATION ROBUSTNESS OF THE ITF EXTENSION

161

service provider is analyzed with respect to its manipulation rate. In this settings,
deviation from truth-telling is only significantly beneficial – at a level of p = 0.05 –
up to a manipulation rate of 60%. It is also noticeable that the mean utility gains
of manipulation strategies compared to a truth-telling strategy are smaller and
less favorable in comparison to the previous setting.
Table 6.2: Utility for a single manipulating service provider with
16 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-50%

0.0171

0.4002

0.0757

-0.0081

-0.3140

0.0845

-40%

0.0300

0.7035

0.0465

0.0072

0.2799

0.0546

-30%

0.0383

0.8983

0.0217

0.0158

0.6344

0.0315

-20%

0.0413

0.9687

0.0095

0.0209

0.8354

0.0176

-10%

0.0424

0.9954

0.0027

0.0234

0.9331

0.0083

0%

0.0426

1.0000

0.0000

0.0248

1.0000

0.0000

10%

0.0425

0.9980

0.0013

0.0263

1.0453***

0.0070

20%

0.0420

0.9858

0.0055

0.0274

1.0659***

0.0131

30%

0.0403

0.9466

0.0144

0.0276

1.0334***

0.0213

40%

0.0402

0.9434

0.0149

0.0283

1.0562***

0.0246

50%

0.0394

0.9244

0.0180

0.0271

1.0570***

0.0282

60%

0.0382

0.8974

0.0227

0.0281

1.0256*

0.0309

70%

0.0373

0.8757

0.0261

0.0267

1.0170

0.0325

80%

0.0359

0.8418

0.0315

0.0268

0.9777

0.0376

90%

0.0352

0.8259

0.0339

0.0268

0.9607

0.0391

100%

0.0348

0.8168

0.0348

0.0276

0.9411

0.0395

110%

0.0329

0.7724

0.0414

0.0254

0.8877

0.0383

120%

0.0320

0.7504

0.0437

0.0245

0.8816

0.0412

130%

0.0314

0.7376

0.0463

0.0240

0.8616

0.0420

140%

0.0305

0.7153

0.0487

0.0246

0.8350

0.0444

162

CHAPTER 6. NUMERICAL RESULTS
Table 6.2: Utility for a single manipulating service provider with
16 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

150%

0.0299

0.7012

0.0506

0.0234

0.8274

0.0440

In the setting with 20 service offers in 4 candidate pools as shown in Table
6.3, service providers do not significantly gain from deviation of more than 20%.
Although, the complex service auction with the ITF extension is not incentive
compatible in a strict theoretical sense, in settings with relatively low competition
(e.g. 28 service offers in 4 candidate pools), service providers cannot significantly
benefit from deviation from reporting their true valuation as shown in Table 6.4,
i.e. the truth-telling strategy is a best (or equally good) strategy compared to any
manipulation strategy.
Table 6.3: Utility for a single manipulating service provider with
20 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-50%

0.0025

0.1122

0.0630

-0.0111

-0.7315

0.0741

-40%

0.0107

0.4870

0.0425

0.0003

0.0187

0.0495

-30%

0.0173

0.7854

0.0231

0.0090

0.5533

0.0292

-20%

0.0208

0.9444

0.0089

0.0137

0.8251

0.0146

-10%

0.0219

0.9916

0.0020

0.0150

0.9434

0.0063

0%

0.0220

1.0000

0.0000

0.0167

1.0000

0.0000

10%

0.0219

0.9920

0.0017

0.0169

1.0298***

0.0059

20%

0.0215

0.9748

0.0051

0.0168

1.0227***

0.0086

30%

0.0205

0.9300

0.0108

0.0157

0.9929

0.0111

6.1. MANIPULATION ROBUSTNESS OF THE ITF EXTENSION

163

Table 6.3: Utility for a single manipulating service provider with
20 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

40%

0.0195

0.8849

0.0156

0.0150

0.9266

0.0143

50%

0.0191

0.8662

0.0169

0.0149

0.9129

0.0163

60%

0.0189

0.8562

0.0176

0.0150

0.8881

0.0166

70%

0.0185

0.8387

0.0197

0.0148

0.8794

0.0187

80%

0.0183

0.8324

0.0201

0.0153

0.8847

0.0201

90%

0.0182

0.8246

0.0207

0.0149

0.8776

0.0218

100%

0.0179

0.8125

0.0217

0.0149

0.8526

0.0220

110%

0.0176

0.7988

0.0235

0.0148

0.8480

0.0234

120%

0.0174

0.7888

0.0243

0.0154

0.8303

0.0266

130%

0.0168

0.7602

0.0270

0.0139

0.7904

0.0270

140%

0.0165

0.7474

0.0285

0.0139

0.7947

0.0293

150%

0.0163

0.7397

0.0293

0.0139

0.7869

0.0279

Table 6.4: Utility for a single manipulating service provider with
28 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-50%

0.0000

0.0005

0.0501

-0.0048

-0.4739

0.0540

-40%

0.0081

0.6271

0.0247

0.0037

0.3617

0.0305

-30%

0.0103

0.8014

0.0152

0.0069

0.6498

0.0191

-20%

0.0119

0.9275

0.0070

0.0090

0.8521

0.0100

-10%

0.0127

0.9908

0.0014

0.0097

0.9500

0.0042

164

CHAPTER 6. NUMERICAL RESULTS
Table 6.4: Utility for a single manipulating service provider with
28 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

0%

0.0129

1.0000

0.0000

0.0101

1.0000

0.0000

10%

0.0127

0.9873

0.0018

0.0108

1.0044

0.0029

20%

0.0122

0.9489

0.0058

0.0101

0.9681

0.0063

30%

0.0120

0.9315

0.0069

0.0107

0.9546

0.0080

40%

0.0119

0.9240

0.0072

0.0099

0.9526

0.0084

50%

0.0116

0.9059

0.0088

0.0098

0.9350

0.0103

60%

0.0113

0.8799

0.0110

0.0099

0.9054

0.0123

70%

0.0109

0.8455

0.0133

0.0098

0.8773

0.0141

80%

0.0106

0.8232

0.0146

0.0094

0.8464

0.0144

90%

0.0104

0.8083

0.0154

0.0092

0.8546

0.0163

100%

0.0099

0.7667

0.0181

0.0087

0.7969

0.0187

110%

0.0099

0.7667

0.0181

0.0088

0.8045

0.0183

120%

0.0095

0.7410

0.0199

0.0087

0.7596

0.0212

130%

0.0093

0.7208

0.0216

0.0081

0.7390

0.0229

140%

0.0091

0.7089

0.0223

0.0083

0.7360

0.0228

150%

0.0089

0.6937

0.0231

0.0082

0.7289

0.0224

Providing an overview over multiple settings with different levels of competition, Figure 6.3 illustrates the relative utility gain following a manipulation
strategy compared to truth-telling.
More detailed results of the simulation-based analysis with respect to different
competition scenarios can be found in Section A.4.

6.1. MANIPULATION ROBUSTNESS OF THE ITF EXTENSION

165

Figure 6.3
Utility for a single manipulating service provider in different
competition scenarios. ITF_|Ṽ |_K denotes the setting with |Ṽ |
service offers in K candidate pools, where |Ṽ | = V \ {vs , v f }.

6.1.3 Implications
In Section 6.1, strategic behavior of service providers has been analyzed in the
complex service auction with the interoperability transfer in comparison to the
complex service auction with the critical value transfer.
As shown analytically in Section 5.1, the complex service auction with critical
value transfer implements a truth-telling equilibrium in weakly dominant strategies, i.e. service providers cannot benefit from misreporting their true valuation.
This is a valuable property for a mechanism and the implemented social choice as
it assures truthful behavior of all participants which allows for an efficient allocation that maximizes welfare among service providers and the service requester. It
furthermore reduces the strategy space of beneficial strategies to a single weakly
dominant strategy independent of the strategies of all other participants. This
implies that service providers do not have to reason about the behavior of other
participants in the complex service auction.
Incentive compatibility comes at the price of budget balance. As a remedy for
this shortcoming, the ITF has been introduced in Section 4.2. The ITF sacrifices

166

CHAPTER 6. NUMERICAL RESULTS

incentive compatibility and efficiency to a certain degree in order to retain budget
balance. The ITF furthermore rewards service providers that offer highly interoperable services within the service value network, which increases the number
of feasible service compositions that can be offered to the requester. Thus, the
ITF implements incentives to increase a services’ interoperability and therefore
fosters the growth of vital and more agile service value networks. This property
is analyzed in detail in Section 6.2.
Using the complex service auction with the critical value transfer as a benchmark, the robustness of the complex service auction with the ITF extension has
been analyzed with respect to bid manipulation (deviation from the truth-telling
strategy). The simulation-based results show that in scenarios with a low level
of competition, implementing the ITF extension opens up strategic behavior to a
certain degree. Service providers can significantly benefit from misreporting their
true valuation. Nevertheless, in settings with a slightly higher level of competition (e.g. 20 service offers in 4 candidate pools), the set of beneficial manipulation
strategies is decreased tremendously. Although the complex service auction with
the ITF extension is not incentive compatible in a strict analytical sense, service
providers cannot significantly benefit from misreporting their true valuation in
settings with a still relatively low level of competition (e.g. cp. results in Table
A.5 in a setting with 28 service providers in 4 candidate pools).
As the attraction of service value networks underlays network externalities,
the value that service requesters gain from initiating a complex service auction
highly depends on the number of participating service providers and the number
of feasible complex service instances that can be provided through the network.
Hence, especially in an early growing stage of a service value network, it might be
desirable for platform providers to implement a mechanism that rewards service
providers for offering multiple services with a high degree of interoperability,
such as the complex service auction with the ITF extension does. Especially in
settings with a low level of competition, critical values of service providers can
be relatively high and unpredictable for the platform provider. Hence, a budgetbalanced variant might be favorable in such an early stage as well. Reaching
a critical mass of participants the network’s inherent competition increases and
critical values of service providers tremendously decrease. Assuring complete
truthful behavior of service provider, the complex service auction with the critical value transfer might be beneficial for both service providers and the service
requester. Service providers do not have to reason about the other participants’
behavior and the service requester trustfully receives a tailored complex service

6.2. INCENTIVIZING INTEROPERABILITY ENDEAVORS

167

instance. This variant always assures a welfare maximizing solution accounting
for the providers’ and the requester’s side.

6.2 Incentivizing Interoperability Endeavors
The interoperability transfer function (ITF) is designed as a remedy to overcome
the lack of budget balance in the complex service auction. The design goal of
the ITF is on the one hand to reduce strategic behavior of service providers with
respect to beneficial deviation from the truth-telling strategy as evaluated in Section 6.1. On the other hand the design of the ITF targets to incentivize service
providers to increase their services’ degree of interoperability, i.e. to increase the
capability of their offered services to communicate and function with other services within the service value network. A higher degree of interoperability increases the potential of a service value network to satisfy different customers’
needs and to provide a huge variety of feasible complex service instances to requesters. Increasing customers’ choice leads to a rapid growth of demand and addresses the long tail of business [And06](cp. Section 2.1.4.3). These implications
are especially important for service value networks in their early stage of development as it attracts various customers which leads to a growth of rich candidate
pools by attracting service providers to participate in value creation (the effect of
network externalities is well-known in literature [SV99, FK07, LM94, KS85]).
To study the effect of the ITF on the network’s degree of interoperability,
the work at hand follows the research method of an agent-based simulation as
outlined in Section 2.3.2. As a suitable benchmark to evaluate incentives implemented by the ITF, an Equal Transfer Function (ETF) is consulted that distributes
the system’s surplus equally among all allocated service providers [PKE01]3 . The
ETF represents a neutral payment scheme as it equally distributes the same surplus as the ITF. The goal of this evaluation is to analyze if and to what degree
increasing the interoperability degree of service offers within a service value network is beneficial for service providers in the complex service auction with the
ITF compared to the complex service auction with the ETF. This leads to the following hypotheses:
Hypothesis 6.1. The overall interoperability degree of a service value network increases
by establishing the ITF compared to the ETF.
3 The

equal transfer function that serves as a benchmark is similar to the k-pricing scheme in
[Sto09, Sch07] with parameter selection k = 1

168

CHAPTER 6. NUMERICAL RESULTS

Hypothesis 6.2. The interoperability degree of allocated service offers increases using the
ITF compared to the ETF.
Hypothesis 6.3. The interoperability degree of non-allocated service offers increases using the ITF compared to the ETF.

6.2.1 Simulation Model
According to the design of the ITF, allocated service providers can gain by increasing their degree of interoperability as this increases their chance of receiving their critical value as a discount in addition. Nevertheless, in the complex
service auction with the ETF it might also be beneficial to increase one’s degree
of interoperability. Focusing on non-allocated service offers, by building additional connections to predecessor services proactively, service providers face the
opportunity to change the network’s topology and augment the chance of being
allocated. It is unclear which effect dominates in settings with different levels of
competition and different proportionate investment costs.
Each service provider is assumed to have a set of strategies representing the
degree of its service’s interoperability that the service provider intents to realize
depending on how it is situated within the network4 . This means that depending on the number of predecessor services, service providers can decide on how
many edges to predecessor services they want to establish. Recall, an edge between two adjacent services denotes the capability of interpreting each others
inputs and outputs, i.e. both services are interoperable and therefore can be iteratively combined within a complex service instance.
Each agent’s5 strategy space is determined by all feasible degrees of interoperability (ID) of its service offer represented by its number of incoming edges. E.g.
if a service offer has 4 predecessor services within the service value network and
the initial number of incoming edges is 2, the service provider’s strategy space is
{2, 3, 4}.
For each extra edge built additionally to the initial number of incoming edges
the service provider is charged proportionate investment costs (IVC) no matter
if the service is being allocated or not6 . Proportionate investment costs are cal4 For

simplicity it is assumed that each service provider owns only a single service within the
network
5 In the context of the agent-based simulation, the terms service provider and agent are used
interchangeably.
6 It is important to note that the complex service auction is conducted as a one-shot game
which has to be considered when evaluating specific properties. Therefore, accounting for full

6.2. INCENTIVIZING INTEROPERABILITY ENDEAVORS

169

culated as a fraction of the internal costs for executing the particular service depending on the predecessor service. It is assumed that internal costs for contextdependent execution reflect the degree of similarity of both services’ interfaces
(e.g. low internal costs indicate a high degree of interfaces’ compatibility). Hence,
investment costs for reprogramming a service’s interface in order to work seamlessly with another service component behave accordingly.
Analogue to Section 6.1.1, each problem set is characterized by a random network topology with random costs cij assigned to each incoming edge of service
offers drawn from U (0, 1.0). Furthermore, the requester’s willingness to pay α is
analogously drawn from U (0, 12 K ) with K being the number of candidate pools.
The evaluation is conducted by means of an agent-based simulation based on
a simple form of a Q-Learning model [WD92]. In contrary to more sophisticated
variants of Q-learning models, the simulation model at hand only considers a
single state which reduces the parameter complexity and therefore simplifies the
calibration of the simulation. Simplifying the simulation model reduces the number of assumptions which allows for a better generalization of results.
Each agent maintains a fitness table which keeps track of the “successfulness”
of each action such that frik represents the fitness of agent i for action k in simulation round r. The fitness for each chosen action is updated based upon the
resulting “reward” (represented by the agent’s utility urik ). Balancing past and
present experiences, the learning parameter β ∈ [0; 1] determines to which degree past and present feedback is incorporated into the fitness update. Thus, the
fitness update evolves as follows:

(6.1)

frik = βfrik−1 + (1 − β)urik

Each action is selected based on a softmax selection method [SB99], i.e. each action is randomly chosen based on the probability Pikr that results from the action’s
fitness relative to the sum of all actions’ fitness such that

(6.2)

Pikr

frik
=
∑k frik

investment costs that are necessary to reprogram a service’s interface in order to enable interoperability with certain other services results in prohibitively high costs which hinders a feasible
one-shot game analysis.

170

CHAPTER 6. NUMERICAL RESULTS

The simulation is conducted as depicted in Figure 6.4. The simulation process
is divided into an exploration phase and a simultaneous exploitation phase.
Exploration Phase
Strategy selection for a single node i
based on probability

Pikr =

Computation of
allocation and
transfers

fikr

∑

Fitness update for node i based on
past and present information

fikr = β (fikr−1 ) + (1 − β )uikr

fr
k ik
∀r ∈ R
∀i ∈ V ∖ { v s , v f }

Simultaneous Exploitation Phase
All nodes choose a strategy
based on

Pikr =

fikr

∑

Calculation of
allocation and transfer
to each node based on
each requester type

Calculation of mean transfer and
update of fitness for all nodes

fikr = β (fikr−1 ) + (1 − β )uikr

r
k ik

f

∀r ∈ R

Figure 6.4
Simulation model for the evaluation of interoperability
incentives using the ITF.

Exploration Phase In this phase each agent explores the solution space in a constant environment where only a single agent learns simultaneously. Starting based on an initial fitness table with equal probabilities for every action,
each agent adapts its individual best action given the other agents do not
change their decisions. The exploration phase is conducted 100 rounds 7
for each agent i ∈ V \ {vs , v f }.
Simultaneous Exploitation Phase In order to determine the most promising action for each agent dependent on the decision taken by every other agent,
in the exploitation phase every agent learns its best action simultaneously
based on the experiences gained from the exploration phase. The simultaneous exploration phase is conducted 100 rounds. 7
7 The

number of required rounds in order to achieve a convergence of the fitness values for
each action has been analyzed by means of a sensitivity analysis.

6.2. INCENTIVIZING INTEROPERABILITY ENDEAVORS

171

As the number of observations is relatively high (N = 50) and the data is normally distributed which has been tested by means of a Kolmogorov-Smirnov test,
stated hypothesis are tested using a one-tailed matched-pairs t-test. With respect
to the overall network, allocated, and non-allocated service offers, the alternative
hypothesis that the interoperability degree of a service value network increases
by establishing the ITF compared to the ETF is analyzed, i.e. the mean difference
in interoperability degrees is greater than zero.

6.2.2 Results
Recall, the complex service auction with the interoperability transfer function
(ITF) is designed to incentivize service providers to increase their services’ degree
of interoperability. In order to evaluate this property, the ITF is benchmarked
against an equal transfer function (ETF) which distributes the system’s surplus
among all allocated service providers equally.
Table 6.5 and Figure 6.5 show a comparison of the ITF and the ETF with respect to resulting interoperability degrees (ID) at different levels of proportionate
investment costs (IVC) for 20 service offers in 4 candidate pools.
Table 6.5: Interoperability degrees (ID) at different levels of proportionate investment cost (IVC) for 20 service offers in 4 candidate pools. ID denotes the overall interoperability degree, ID_A denotes the interoperability degree of all allocated service offers, and ID_NA denotes the interoperability degree of all nonallocated service offers. * denotes significance at the level of
p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

IVC

ID

ID_A

ID_NA

ID

ID_A

0%

0.6665

0.7571

0.6438

0.6766***

0.7711*** 0.6530***

10%

0.4595

0.6025

0.4238

0.4891***

0.6710*** 0.4436***

20%

0.3676

0.4811

0.3392

0.3963***

0.5780*** 0.3509***

30%

0.3343

0.4201

0.3129

0.3544***

0.4934*** 0.3196***

40%

0.3199

0.3838

0.3040

0.3347***

0.4474*** 0.3065***

50%

0.3201

0.3831

0.3043

0.3321***

0.4394*** 0.3053*

60%

0.3147

0.3576

0.3039

0.3218***

0.3899*** 0.3048**

70%

0.3118

0.3355

0.3059

0.3164***

0.3616*** 0.3051*

ID_NA

172

CHAPTER 6. NUMERICAL RESULTS
Table 6.5: Interoperability degrees (ID) at different levels of proportionate investment cost (IVC) for 20 service offers in 4 candidate pools. ID denotes the overall interoperability degree, ID_A denotes the interoperability degree of all allocated service offers, and ID_NA denotes the interoperability degree of all nonallocated service offers. * denotes significance at the level of
p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

IVC

ID

ID_A

ID_NA

ID

80%

0.3145

0.3612

0.3029

0.3196*** 0.3854*** 0.3032

90%

0.3097

0.3407

0.3019

0.3133*** 0.3616*** 0.3013

100% 0.3111

0.3617

0.2985

0.3137*** 0.3772*** 0.2979

110% 0.3101

0.3542

0.2990

0.3113*** 0.3614*** 0.2988

120% 0.3150

0.3789

0.2990

0.3159*** 0.3841*** 0.2989

130% 0.3084

0.3749

0.2918

0.3110*** 0.3877*** 0.2918

140% 0.3114

0.3504

0.3017

0.3122*** 0.3537*** 0.3018

150% 0.3091

0.3431

0.3006

0.3101*** 0.3479**

0.3007

160% 0.3101

0.3407

0.3025

0.3111**

0.3469**

0.3022

170% 0.3076

0.3416

0.2991

0.3080*

0.3437*

0.2991

180% 0.3115

0.3563

0.3003

0.3076*

0.3505

0.2969

190% 0.3126

0.3539

0.3022

0.3126

0.3541

0.3022

200% 0.3098

0.3598

0.2973

0.3101

0.3613

0.2973

ID_A

ID_NA

In general, it is observable that an increase of proportionate investment costs results
in a decrease of interoperability degrees with respect to both transfer functions. Investment costs are obviously a disincentive for increasing ones services’ degree of
interoperability and therefore counteract the incentive schema provided by the
ITF. Despite of the primary incentives provided by the transfer function, service
providers might also have an incentive to increase their degree of interoperability
independent of the design of the transfer function as establishing more relations
to other services allows for proactively changing the initial topology of the service
value network. By doing so, service providers face the opportunity to be better
situated within the network and increase the likelihood of being allocated. Thus,
proportionate investment costs also disincentivize interoperability endeavors un-

6.2. INCENTIVIZING INTEROPERABILITY ENDEAVORS

173

Figure 6.5
Interoperability degrees (ID) at different levels of proportionate
investment Cost (IVC) for 20 service offers in 4 candidate pools.
ID denotes the overall interoperability degree, ID_A denotes the
interoperability degree of all allocated service offers, and ID_NA
denotes the interoperability degree of all non-allocated service
offers.

der the presence of a “neutral” transfer function such as the ETF which results in
a decrease of interoperability degrees with respect to both transfer functions.
Furthermore the degree of interoperability is higher for allocated service offers than
for non-allocated services offers. The reason for this phenomenon is based on the
fact that service offers that are initially more interoperable with other services
face a higher likelihood of being allocated than service offers with a low degree
of interoperability. Hence, independent of the design of the transfer function,
allocated services yield a higher degree of interoperability than non-allocated
services. Nevertheless the difference in interoperability between allocated and
non-allocated services decreases as proportionate investment costs increase. Due
to the fact that investment costs are a disincentive for being interoperable, each
service’s interoperability degree is pushed down towards the initial density of
the service value network.

174

CHAPTER 6. NUMERICAL RESULTS

In the setting with 20 service offers in 4 candidate pools (cp. Table 6.5), Hypothesis 6.1 is supported significantly until a level of proportionate investment
costs of 180%. Distinguishing between allocated and non-allocated service offers,
Hypothesis 6.2 is supported until 170% investment costs and Hypothesis 6.3 is
significantly supported up to 70% proportionate investment costs. The difference
in the levels of investment costs until each hypothesis is supported bases on two
effects. First, allocated services are primarily incentivized by the construction of the ITF
whereas non-allocated services only benefit from a higher degree of interoperability if they are allocated in the changed topology. Hence, for service providers
that own non-allocated services, the effect of the implemented incentive is compensated
earlier by the disincentive provided through the investment costs. The second effect for
the different support levels of each hypothesis is based on the fact that there are
more discrete degrees of interoperability for the overall network than for a subset
of service offers. This means that as allocated service offers are rare, if a single service’s degree of interoperability decreases, the overall degree of interoperability
for all allocated services drops rapidly.
Looking at different levels of competition in the service value network, Table
6.6 shows a comparison of the ITF and the ETF with respect to resulting interoperability degrees at different levels of proportionate investment costs for 32 service
offers in 4 candidate pools.
Table 6.6: Interoperability degrees (ID) at different levels of proportionate investment cost (IVC) for 32 service offers in 4 candidate pools. ID denotes the overall interoperability degree, ID_A denotes the interoperability degree of all allocated service offers, and ID_NA denotes the interoperability degree of all nonallocated service offers. * denotes significance at the level of
p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

IVC

ID

ID_A

ID_NA

ID

0%

0.6118

0.7298

0.5949

0.6189*** 0.7369*** 0.6020***

50%

0.2026

0.2474

0.1962

0.2051*** 0.2642*** 0.1966*

100% 0.2015

0.2453

0.1952

0.2017*** 0.2472**

0.1952*

150% 0.2016

0.2427

0.1957

0.2016*

0.2433*

0.1957

200% 0.2004

0.2369

0.1952

0.2004

0.2369

0.1952

ID_A

ID_NA

6.2. INCENTIVIZING INTEROPERABILITY ENDEAVORS

175

Compared to the previous setting, the overall incentive provided by the ITF
to increase interoperability is weakened. At a level of 150% proportionate investment costs, Hypothesis 6.1 and 6.2 are only supported at a level of p = 0.1
whereas Hypothesis 6.3 is not supported at all. A higher level of competition
decreases critical values of service providers. Thus, increasing ones degree of interoperability to obtain ones critical value is less favorable in highly competitive
settings.

6.2.3 Implications
In Section 6.2 the interoperability transfer function (ITF) is analyzed with respect
to its design to incentivize service providers to increase their services’ degree of
interoperability. The evaluation is conducted by means of an agent-based simulation comparing the complex service auction with the ITF extension and the
ITF with an equal transfer function (ETF) that distributes the available surplus
equally among service providers that own allocated service offers within the service value network.
Summarizing the results in Section 6.2.2, the ITF extension incentivizes service
providers – those which own allocated (cp. Hypothesis 6.2) and non-allocated
(cp. Hypothesis 6.3) service offers – to increase their services’ degree of interoperability as stated by Hypothesis 6.1. That is, the design of the ITF implements
incentives to undertake endeavors to customize service interfaces which enables
communication and data transfer with multiple adjacent service components. Of
course, proportionate investment costs that service providers have to bear for this
customization process function as a disincentive counteracting interoperability
endeavors. In general, in service value networks with a low level of competition and only few interrelated service offers, the ITF extension appears to be a
promising approach to foster the growth of service value networks’ variety in
an early stage of development and to increase the multitude of feasible complex
service instances that can be offered to customers. An increase of variety and
interoperability leverages network externalities [SV99, FK07, LM94, KS85] and
attracts customers which in turn attracts more service providers to participate in
the complex service auction.

176

6.3

CHAPTER 6. NUMERICAL RESULTS

Bundling Strategies of Service Providers

Recall, in Section 5.1 it has been shown that under the assumption of rationality,
service providers act best (or at least equally good) by revealing their true multidimensional type which reduces their bidding strategy space to a single strategy.
Broadening service providers’ strategic horizon, it might be beneficial under certain circumstances to form coalitions and offer services in a bundled fashion. This
section focuses on strategies of service providers with focus on opportunities to
form bundled offers with other providers depending on how they are situated
within service value networks.
Since a service provider’s offer is only successful if one of its edges is allocated,
service providers tend to find strategies to improve their situation. Two options
are mainly distinguished, unbundling vs. bundling. Service providers can decide
on either offering services on their own with a certain degree of interoperability
to preceeding offers. Such a strategy is referred to as unbundling strategy. On
the other hand service providers can also provide bundled services together with
service providers that own services in adjacent candidate pools (either preceeding
or succeeding), i.e. two service providers from different candidate pools combine
their offers to a single service which aggregates both service characteristics. It is
furthermore assumed that a combined service offer results in lower internal costs
due to synergy effects that can be leverage through bundled offers. This strategy
is referred to as bundling strategy. There are mainly two contrary effects and it is
unclear which effect dominates in what setting.
Competing in quality through differentiation and flexibility It is certainly just
reasonable to follow an unbundling strategy if a provider’s service offers
expose significantly lower prices (due to lower internal costs) or significantly better QoS characteristics than competing offers. Additionally, unbundled services offer more differentiated and specialized functionality
which increases their flexible integration into different complex services,
and thus, increase the number of possible combinations with other services
from other candidate pools.
Competing in price through cost reduction On the other hand, it might be advantageous for service providers to cut costs through forming bundled offers collaboratively, i.e. combining their service offers to a service bundle
which offers the functionality of both services in an integrated manner.
In that case internal costs of the bundled services are likely to be lower
compared to the sum of internal costs of two single offers. In the case of

6.3. BUNDLING STRATEGIES OF SERVICE PROVIDERS

177

bundling, an aggregation of attribute values defining the service’s configuration is done according to aggregation operations in Table 3.1. Nevertheless, bundling service offers results in a reduction of the degree of interoperability, i.e. a merge of service offers prunes incoming edges to preceeding
services which decreases the number of complex service instances the bundled offer is part of.
It is unclear which strategy is beneficial for service providers with respect to
how their service offers are situated within the service value network. Even for
service offers that are competitive in price and attractive for the service requesters
– i.e. they are allocated solely – forming a bundled offer with a less competitive
service offer may be mutually beneficial for both partners. The following example
illustrates the phenomenon where a bundling strategy is mutually beneficial for
an allocated and a non-allocated service provider at the same time even though
there is no reduction of internal costs due to bundling synergies assumed:
Example 6.1 [B ENEFICIAL B UNDLING S TRATEGY ]. Figure 6.6 depicts the service
value network from an initial ex-ante perspective. Without loss of generality it is assumed
that service providers only announce price bids (no QoS) and each service provider only
owns a single service offer within the service value network. Consequently there are four
service providers sy , sz , s a , sb that own service offers y, z, a, b. Numbers on incoming edges
to each node represent price bids placed by service providers8 .
0.1

y

0.3

z

0.2
f

s
0.1
0.1

a

0.9

b

Figure 6.6
Beneficial bundling strategy for allocated and non-allocated
service providers (ex-ante case).
According to the CSA mechanism, the path f ∗ = {esa , eaz , ez f } is allocated as it yields
the overall lowest price of 0.2 and therefore maximizes welfare. The “second-best” path
f 2 = {esy , eyb , ez f } yields an overall price of 0.3. According to the CSA’s transfer function, payments are given to allocated service providers such that tsa = 0.1 + (0.3 − 0.2) =
0.2 and tsz = 0.1 + (0.3 − 0.2) = 0.2.
8 Note

that according to Theorem 5.2 it is a dominant strategy equilibrium in the CSA that
service providers report their valuations truthfully, that is, they announce their internal costs.

178

CHAPTER 6. NUMERICAL RESULTS

Focusing on the ex-post case depicted in Figure 6.7, service providers sy and sz have
agreed on offering their service offers y and z as a bundle yz. As it is assumed that it is
not possible to realize a cost reduction following a bundling strategy, internal costs for
offering the single services add up to 0.4 for service offer yz.
yz
0.4
f

s
0.1

a

0.9

b

Figure 6.7
Beneficial bundling strategy for allocated and not allocated
service providers (ex-post case).
According to the CSA mechanism, the path f ∗ = {esyz , ez f } is allocated which results
in a price of 0.4 whereas the other path f 2 = {esa , eab , eb f } yields a price of 1.0. It is assumed that service providers sy and sz divide their payoff according to their contribution
to the alliance which means the ratio of their internal costs determines their share. Consequently payments to service providers evolve es follows: tsy = 43 (0.4 + (1.0 − 0.4)) =
0.75 and tsz = 14 (0.4 + (1.0 − 0.4)) = 0.25.
The example at hand shows that although if there is no cost reduction due to synergy
effects when following a bundling strategy it might be beneficial for allocated and nonallocated service providers to jointly offer a bundled solution. In this scenario the effect of
reducing the network’s density (meaning cutting edges by merging service offerings) also
affects the number of feasible complex service instances and the composition outcome.
Both fundamental strategies imply advantageous and disadvantageous effects and it is unclear which effect dominates: lower costs to increase the likelihood of being part of the allocation by offering bundled services at a lower price
but at the same time a decrease in interoperability which reduces the number of
possible service combinations that entail the bundled offer, and thus, reducing the
likelihood to be part of the allocation. In contrary an unbundling strategy increase
differentiation and specialization but disables opportunities to realize synergy effects. It is proposed that the question whether or not bundling or unbundling is
the better strategy to follow depends on the service provider’s individual strategic strength. Thus, it is distinguished in service providers that are part of the
allocation and those which are not. The following hypotheses are derived:

6.3. BUNDLING STRATEGIES OF SERVICE PROVIDERS

179

Hypothesis 6.4. Service offers which are not allocated have a higher likelihood of being
allocated by choosing a bundling strategy instead of an unbundling strategy.
Hypothesis 6.5. For service offers which are not allocated, a bundling strategy leads to
a higher expected payoff than an unbundling strategy.
Hypothesis 6.6. Allocated service offers have a higher likelihood of staying allocated by
following an unbundling strategy instead of a bundling strategy.
Hypothesis 6.7. For service offers that are allocated, an unbundling strategy leads to a
higher payoff than following a bundling strategy.
The terms likelihood or probability and expected payoff are used with respect to
the limited set of observations. Therefore the likelihood or probability of an event
refers to the relative frequency of the occurrences of that particular event. Analogously, the term expected payoff refers to the relative frequency times the mean
payment observed.

6.3.1 Simulation Model
The stated hypotheses are studied following a simulation approach. The problem is modeled as an n-person game in which each node represents a service
offer. Without loss of generality it is assumed that service providers only own
a single service offer within the network. Each service offer is characterized by
an attribute value for the types encryption and response time. Dependent on the
network topology each service provider faces the decision of choosing an action k
which is either to offer a service on its own, i.e. an unbundling strategy which is denoted by k = u, or to form a bundled offer with one of its successors, i.e. a bundling
strategy which is denoted by k = b. Thus, in each simulation round r ∈ R each
node i ∈ V \ {vs , v f } has to choose an action k ∈ Ki . The service provider’s utility
uik resulting from the action chosen is dependent on the topology of the network,
the service requester’s scoring function, and all other service offers within the
network including their quality and price. For each topology all these factors are
stochastic. As such, the node’s action decision does not solely control the payoff. Thus, the decision problem of the nodes is comparable to an n-armed bandit
problem. Since reinforcement learning has proven to cope with such a model-free
situation, a simple form of a reinforcement learning algorithm is applied in the
present approach. Each node i assigns a fitness value frik to each possible action

180

CHAPTER 6. NUMERICAL RESULTS

k ∈ Ki . The fitness of the chosen action k is updated at the end of the period
according to the update rule with the learning rate β ∈ [0; 1].:

(6.3)

frik = βfrik−1 + (1 − β)urik

Actions are chosen according to a probability choice rule based on each fitness
propensity.

(6.4)

Pikr =

frik
∑k frik

The action’s propensity is calculated as its fitness weighted by the sum of all
fitness values corresponding to the node’s actions.
Analogue to the simulation model in Section 6.2.1, the conduction of the simulation is divided in two phases: an exploration phase and a simultaneous exploitation
phase. Figure 6.8 displays the simulation phases and the steps of each phase. Each
phase consists of a certain number of rounds r ∈ R. Each round in the single exploration phase consists of 3 steps. In the first step a single node i chooses an
action k with propensity Pikr out of its action set. In the second step, the allocation
is computed as well as the mean payoffs for all allocated nodes based on all requester types (different requester types are explained in detail in Section 6.3.2). It
is important to notice that, depending on the requesters’ scoring functions, allocated service offers and corresponding payoffs differ. In the third step, the fitness
value of the chosen action is updated based on the mean payoff computed based
on all requester types.
After having trained all nodes, the simultaneous exploitation phase starts in
order to evaluate settings with simultaneous decisions. Analogue to the exploration phase, each round of the simultaneous exploitation phase runs through
three steps. In the first step, all nodes simultaneously choose a strategy based
on Pik . Note, that in the training phase it is just one node choosing the strategy.
Only if bilateral bundling decisions match, service offers are merged to a single
node forming a bundled offer. The allocation and the mean payoffs based on all
requester types are computed in the second step. Each service provider is assigned a numerical value indicating its market power within the service value
network. In case two service offers are merged to a bundled offer which is allocated, resulting payoff is distributed based on the market power ratio of both

6.3. BUNDLING STRATEGIES OF SERVICE PROVIDERS

181

Exploration Phase
Strategy selection for a single node i
based on probability

Pikr =

fikr

∑

Computation of
allocation and
transfers based on all
different requester
types

fr
k ik

Fitness update for node i based on
past and present information

fikr = β (fikr−1 ) + (1 − β )uikr

∀r ∈ R
∀i ∈ V ∖ { v s , v f }

Simultaneous Exploitation Phase
All nodes choose a strategy
based on

Pikr =

fikr

∑

r
k ik

f

Calculation of
allocation and transfer
to each node based on
all different requester
types

and matching decision are accepted

Calculation of mean transfer and
update of fitness for all nodes

fikr = β (fikr−1 ) + (1 − β )uikr

∀r ∈ R

Figure 6.8
Simulation model for the evaluation of bundling and
unbundling strategies of service providers.

service providers. The last step is again to update the fitness values of all nodes
based on the mean payoff.
The data of the simultaneous exploitation phase is analyzed with respect to
every possible event that may occur during the conduction of the complex service
auction. Table 6.7 shows each possible event that is analyzed with respect to its
relative frequency of occurrence (which can be interpreted as the likelihood of the
event’s realization) and its expected payoff, i.e. the corresponding mean payoffs
received times the event’s likelihood of occurrence.
The stated hypothesis are tested using a Wilcoxon signed-rank test as the
number of observations is relatively small (N = 30) and the data is not normally
distributed which was tested by means of a Kolmogorov-Smirnov test. The data
is based on the mean relative frequencies of each event and corresponding expected payoffs over all service providers.

182

CHAPTER 6. NUMERICAL RESULTS
Table 6.7: Analyzed events for the evaluation of bundling and
unbundling strategies of service providers with respect to their
relative frequency of occurrence and the corresponding expected
payoffs. The set Ẽs denotes the set of edges with Ẽs = {eij |eij ∈
o, j ∈ σ (s), i ∈ τ ( j)}, i.e. the set of allocated edges that belong to
service provider s’s service offers.

Metric
Relative
Frequency
Expected
Payoff

Ẽt
Ẽt 6= ∅
Ẽt = ∅
Ẽt 6= ∅
Ẽt = ∅

Ẽt+1

k=b

k=u

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

P( Ẽt+1 6= ∅|k = b, Ẽt 6= ∅)
P( Ẽt+1 = ∅|k = b, Ẽt 6= ∅)
P( Ẽt+1 6= ∅|k = b, Ẽt = ∅)
P( Ẽt+1 = ∅|k = b, Ẽt = ∅)

P( Ẽt+1 6= ∅|k = u, Ẽt 6= ∅)
P( Ẽt+1 = ∅|k = u, Ẽt 6= ∅)
P( Ẽt+1 6= ∅|k = u, Ẽt = ∅)
P( Ẽt+1 = ∅|k = u, Ẽt = ∅)

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

E( Ẽt+1 6= ∅|k = b, Ẽt 6= ∅)
E( Ẽt+1 = ∅|k = b, Ẽt 6= ∅)
E( Ẽt+1 6= ∅|k = b, Ẽt = ∅)
E( Ẽt+1 = ∅|k = b, Ẽt = ∅)

E( Ẽt+1 6= ∅|k = u, Ẽt 6= ∅)
E( Ẽt+1 = ∅|k = u, Ẽt 6= ∅)
E( Ẽt+1 6= ∅|k = u, Ẽt = ∅)
E( Ẽt+1 = ∅|k = u, Ẽt = ∅)

6.3.2 Simulation Settings
As introduced in Section 6.3 there are two fundamental strategic alternatives service providers have to face: Focusing on differentiation and the provision of flexible service offers that are of highly specialized by following an unbundling strategy
or focusing on cost reduction due to synergy effects in order to compete in price
by following a bundling strategy.
To evaluate the success of both strategies and how advantageous and disadvantageous effects of both strategies dominate under which conditions, five different representative types of services requesters are simulated that have different
preferences over different QoS attributes and prices of the complex service. Each
of these five standard subjects represents a homogenous group of requesters9 .
As the results are dependent on the level of competition, multiple scenarios
with different numbers of service offers and candidate pools are evaluated. Each
scenario has been evaluated with 30 different problems sets, i.e. 30 randomly generated topologies based on the parameters outlines in Table 6.8. The exploration
phase as well as the simultaneous exploitation phase are conducted 500 times10 .
Each service offer is characterized by attribute values for the types response
time and encryption. Attribute values for the type response time are uniformly
9 An alternative approach is the simulation of service requesters with randomly chosen prefer-

ences. Nevertheless, this results in heavy statistical noise and hinders the convergence of service
providers’ fitness in an appropriate number of exploration and exploitation rounds.
10 A sensitivity analysis has shown that after 500 rounds with a learning rate of β = 0.1, which
avoids stagnation in local optima, the agents’ fitness converges to a single best action.

6.3. BUNDLING STRATEGIES OF SERVICE PROVIDERS

183

Table 6.8: Simulation settings for the evaluation of bundling and
unbundling strategies of service providers.
Parameter

Value

Exploration phase
Exploitation phase
Learning rate β

# rounds
# rounds

500
500
0.1

Service offers

#
Response time (art
j )
et
Encryption (a j )
Costs (cij )
Market power mp

varied
∈ U (0, 1.0)
∈ {0, 1}
∈ U (0, 1.0)
∈ U (0, 1.0)

Service requesters

#
α
Type A
Type B
Type C
Type D
Type E

5

1
2K
λrt =

0.3, λet = 0.7
λrt = 0.4, λet = 0.6
λrt = 0.5, λet = 0.5
λrt = 0.6, λet = 0.4
λrt = 0.7, λet = 0.3

distributed over the interval [0, 0.1]. Encryption values are also randomly chosen
and can be either FALSE or TRUE indicated by 0 and 1. Internal costs of service
offers on each incoming edge are drawn from a uniform distribution over the
interval [0, 0.1].

6.3.3 Results & Implications
For the assessment two different situations for a service provider’s service offer
are distinguished: it either is part of the allocation or it is not for the case that
the service is solely offered. In both cases, the service provider can decide on
the u or the b strategy which can result in either allocation or non allocation. As
such, there are eight possible results. The probability of ending up in either of
these states is the conditional probability of the described preconditions. These
conditional probabilities are derived through the mean relative frequencies (over
all service providers) of each event within the simulation. Table 6.7 displays the
possible states, the conditional probabilities of these states as well as the expected
payoff in these states.
As the number of effects is manifold, the analysis of protruding observations,
their interpretation, and implications are structured as follows:

184

CHAPTER 6. NUMERICAL RESULTS

• Analysis within a single competition and cost reduction scenario
• Analysis across different levels of cost reduction and competition
• Bird’s eye analysis regarding the overall provider surplus
Analysis within a single competition and cost reduction scenario – Focusing on
a single competition and cost reduction scenario, Table 6.9 shows the results in a
setting with 20 service offers in 4 candidate pools with no cost reduction due to
synergy effects.
Table 6.9: Evaluation of bundling and unbundling strategies of
service providers with 20 service offers in 4 candidate pools and
0% cost reduction due to synergy effects. Relative frequency of
possible events and corresponding expected payoffs of service
providers are analyzed in t + 1 for bundling and unbundling
strategies depending on the allocation in t. * denotes significance
at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01. Only
tested results that correspond to stated hypothesis are indicated.
Metric
Relative
Frequency
Expected
Payoff

Ẽt
Ẽt 6= ∅
Ẽt = ∅
Ẽt 6= ∅
Ẽt = ∅

k=b

k=u

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

0.4707
0.5293
0.1904***
0.8095

0.7269***
0.2730
0.0355
0.9645

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

0.2834
0.0000
0.1009***
0.0000

0.4013***
0.0000
0.0193
0.0000

Ẽt+1

The results show that service offers which are not allocated have a significantly higher likelihood of being allocated by choosing a bundling strategy instead of an unbundling strategy which supports Hypothesis 6.4. Also with respect to expected payoffs, for service offers which are not allocated, a bundling
strategy leads to a significantly higher expected payoff than an unbundling strategy which supports Hypothesis 6.5. The fact, that these service offers are not
allocated initially indicates that they are either not pricewise competitive or that
their QoS characteristics are not sufficiently valuable for the service requesters
(or both). Thus, by combining their offers with more attractive components – although bearing the loss of interoperability as edges to adjacent service offers are
pruned – less competitive service providers increase their chance of being allocated and manage to increase their payoff at the same time (cp. Hypothesis 6.4
and 6.5).

6.3. BUNDLING STRATEGIES OF SERVICE PROVIDERS

185

Service providers that are initially capable of competing successfully within
the service value networks, i.e. their unbundled service offers are pricewise attractive and expose valuable characteristics for the requesters, have a higher
chance of staying allocated by following an unbundling strategy instead of a
bundling strategy. Thus, Hypothesis 6.6 is supported. Also with respect to the expected payoff, an unbundling strategy is beneficial for allocated service providers
and outperforms a bundling strategy significantly which supports Hypothesis
6.7.
Summarizing the results, Figure 6.9 shows the corresponding decision tree
for service providers participating in the complex service auction with respect to
bundling and unbundling strategies in a setting with a low level of competition
and no cost reduction due to bundling synergies.
Analysis across different levels of cost reduction and competition – On average,
the results show that cost reduction due to synergy effects realized through a bundling
strategy increase the likelihood of being allocated in more competitive scenarios. This
effect is not observable in a setting with 20 service offers in 4 candidate pools as
the relatively low level of competition requires a tremendous cost reduction to
outperform other substitute service offers (cp. Table 6.9 and Table 6.10).
Table 6.10: Evaluation of bundling and unbundling strategies of
service providers with 20 service offers in 4 candidate pools and
50% cost reduction due to synergy effects. Relative frequency
of possible events and corresponding expected payoffs of service providers are analyzed in t + 1 for bundling and unbundling
strategies depending on the allocation in t. * denotes significance
at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01. Only
tested results that correspond to stated hypothesis are indicated.
Metric
Relative
Frequency
Expected
Payoff

Ẽt
Ẽt 6= ∅
Ẽt = ∅
Ẽt 6= ∅
Ẽt = ∅

k=b

k=u

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

0.5035
0.4965
0.1851***
0.8148

0.7068***
0.2931
0.0328
0.9672

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

0.2519
0.0000
0.0698***
0.0000

0.3940***
0.0000
0.0157
0.0000

Ẽt+1

In other words, the spread between dominant and dominated service
providers is larger in settings with a low level of competition which makes ef-

186

CHAPTER 6. NUMERICAL RESULTS

Ẽt+1 6= ∅

P( Ẽt+1 6= ∅|k = b, Ẽt 6= ∅) = 0.4707
E( Ẽt+1 6= ∅|k = b, Ẽt 6= ∅) = 0.2834

m
k=b
Ẽt+1 = ∅

...

Ẽt+1 6= ∅

P( Ẽt+1 6= ∅|k = u, Ẽt 6= ∅) = 0.7269***

s
k=u

Ẽt 6= ∅

E( Ẽt+1 6= ∅|k = u, Ẽt 6= ∅) = 0.4013***

m
Ẽt+1 = ∅

...

Ẽt+1 6= ∅

P( Ẽt+1 6= ∅|k = b, Ẽt = ∅) = 0.1904***

m

E( Ẽt+1 6= ∅|k = b, Ẽt = ∅) = 0.1009***

m

Ẽt = ∅

k=b
Ẽt+1 = ∅

...

Ẽt+1 6= ∅

P( Ẽt+1 6= ∅|k = u, Ẽt = ∅) = 0.0355

s
k=u

E( Ẽt+1 6= ∅|k = u, Ẽt = ∅) = 0.0193

m
Ẽt+1 = ∅

...

Figure 6.9
Relative frequencies and expected payoffs of bundling and
unbundling strategies with 20 service offers in 4 candidate pools
and no cost reduction due to synergy effects. Nodes indicated
by m denote a decision triggered by the mechanism and s a
decision by the service provider.

6.3. BUNDLING STRATEGIES OF SERVICE PROVIDERS

187

forts to increase a service offer’s attractiveness harder than in high competition
settings. In settings with an increased level of competition (e.g. 28 service offers in 4 candidate pools) the effect is significantly observable as a cost reduction
of 50% is sufficient to make previously dominated service providers pricewise
attractive for the requesters as bundled offers. For a comparison of the results,
Table 6.11 shows a setting with an increased level of competition and no cost reduction whereas Table 6.12 shows results assuming a 50% cost reduction for a
bundling strategy.
Table 6.11: Evaluation of bundling and unbundling strategies of
service providers with 28 service offers in 4 candidate pools and
0% cost reduction due to synergy effects. Relative frequency of
possible events and corresponding expected payoffs of service
providers are analyzed in t + 1 for bundling and unbundling
strategies depending on the allocation in t. * denotes significance
at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01. Only
tested results that correspond to stated hypothesis are indicated.
Metric
Relative
Frequency
Expected
Payoff

Ẽt
Ẽt 6= ∅
Ẽt = ∅
Ẽt 6= ∅
Ẽt = ∅

k=b

k=u

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

0.3947
0.6053
0.0502**
0.9497

0.9398***
0.0601
0.0129
0.9871

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

0.1553
0.0000
0.0199
0.0000

0.4248***
0.0000
0.0052
0.0000

Ẽt+1

As shown in Theorem 5.2 it is a weakly dominant strategy for service
providers to bid truthfully which implies that reducing costs results in a reduced
price which service providers charge for their offerings. Nevertheless, Corollary
5.2 shows that in case of being part of the allocation, the service providers’ payoff
is independent of their bids which means that in contrary to an increased likelihood to become allocated, a cost reduction does not influence the agents payoff.
In contrary to e.g. a setting with 20 service offers in 4 candidate pools and
no cost reduction, Hypothesis 6.5 is not supported in settings with a high level of competition and no cost reduction as illustrated in Table 6.11. With an increase of the
number of service offers, interrelations and feasible complex services, a bundling
strategy results in a tremendous loss of interoperability. The more preceeding and
succeeding service offers and the higher the number of interrelations between services, the higher the loss of interoperability incurred through a merge of single
offers within a service value network. In the setting with 28 service offers in 4

188

CHAPTER 6. NUMERICAL RESULTS
Table 6.12: Evaluation of bundling and unbundling strategies of
service providers with 28 service offers in 4 candidate pools and
50% cost reduction due to synergy effects. Relative frequency
of possible events and corresponding expected payoffs of service providers are analyzed in t + 1 for bundling and unbundling
strategies depending on the allocation in t. * denotes significance
at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01. Only
tested results that correspond to stated hypothesis are indicated.
Metric
Relative
Frequency
Expected
Payoff

Ẽt
Ẽt 6= ∅
Ẽt = ∅
Ẽt 6= ∅
Ẽt = ∅

k=b

k=u

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

0.4396
0.5604
0.1127***
0.8872

0.9275***
0.0725
0.0128
0.9872

Ẽt+1 6= ∅
Ẽt+1 = ∅
Ẽt+1 6= ∅
Ẽt+1 = ∅

0.1274
0.0000
0.0509***
0.0000

0.4556***
0.0000
0.0040
0.0000

Ẽt+1

candidate pools and no cost reduction for bundled services, the likelihood to get
allocated is still higher when following a bundling strategy (supported at a significance level of p = 0.05). Nevertheless, the expected payoff that results from
that strategy is not significantly better than for the case of unbundling. Thus, in
case the service providers’ services are not allocated solely given a high level of competition and given there are no synergy effects that reduce costs for bundled offers, they are
indifferent between a bundling and an unbundling strategy. As a result of the higher
level of competition, critical values for service providers are generally lower and
especially in the case of bundling, both service providers have to share their payoff according to their market power which again decreases payments in case of
getting allocated.
Bird’s eye analysis regarding the overall provider surplus – Recall, in the simulation model, service providers maintain a fitness table for each bundling and unbundling strategy. Fitness values indicate the “successfulness” of feasible strategies based on the payoff received when choosing a particular strategy (e.g. higher
fitness values indicate beneficial strategies). Thus, fitness values for each strategy
are closely related to the payments gained as a feedback to the actions triggered
by service providers. Mean fitness values over all service providers for each problem set are depicted in Figure 6.10 and Figure 6.11 in scenarios with different
levels of competition and different levels of cost reduction.

189

1.0

1.0

6.3. BUNDLING STRATEGIES OF SERVICE PROVIDERS

0.8
0.6
0.0

0.2

0.4

Fitness
0.0

0.2

0.4

Fitness

0.6

0.8

unbundling
bundling

0

5

10

15

20

25

30

unbundling

Topology

bundling
Strategy

1.0

1.0

(a) No cost reduction due to bundling synergies with 20 service offers in 4
candidate pools.

0.8
0.6
0.0

0.2

0.4

Fitness
0.0

0.2

0.4

Fitness

0.6

0.8

unbundling
bundling

0

5

10

15
Topology

20

25

30

unbundling

bundling
Strategy

(b) 50% cost reduction due to bundling synergies with 20 service offers in 4
candidate pools.
Figure 6.10
Strategy fitness in different cost reduction scenarios with 20
service offers in 4 candidate pools.

1.0

CHAPTER 6. NUMERICAL RESULTS

1.0

190

0.8
0.6
0.0

0.2

0.4

Fitness
0.0

0.2

0.4

Fitness

0.6

0.8

unbundling
bundling

0

5

10

15

20

25

30

unbundling

Topology

bundling
Strategy

1.0

1.0

(a) 0% cost reduction due to bundling synergies with 28 service offers in 4
candidate pools.

0.8
0.6
0.0

0.2

0.4

Fitness
0.0

0.2

0.4

Fitness

0.6

0.8

unbundling
bundling

0

5

10

15
Topology

20

25

30

unbundling

bundling
Strategy

(b) 50% cost reduction due to bundling synergies with 28 service offers in 4
candidate pools.
Figure 6.11
Strategy fitness in different cost reduction scenarios with 28
service offers in 4 candidate pools.

6.3. BUNDLING STRATEGIES OF SERVICE PROVIDERS

191

In general, bundling strategies seem to outperform unbundling strategies regarding their fitness values. Nevertheless, this is only true for the collectivity
of service providers. It is important to notice that there are less allocated service offers than non-allocated services and service providers that own services
within each group valuate each strategy differently. As already shown, following
an unbundling strategy is in general not beneficial for providers that offer less
competitive services which is true for the majority of participants. Hence, fitness
values for an unbundling strategy for service providers that offer less competitive services are close to zero which in turn strongly decreases the mean fitness
for that strategy.
A fundamental effect is observable when comparing scenarios with no cost
reduction due to missing synergies as illustrated in Figure 6.10a and with large
synergy effects as depicted in Figure 6.10b. The higher the synergy effects realized through bundled offers, the lower the mean fitness value for that strategy.
Recall, fitness value are closely related to the payments gained by following a particular strategy. Thus, a decrease in the mean fitness value for the bundling strategy reflects the fact that service providers receive lower payments when realizing
synergy effects. Synergy effects reduce costs for service provision. A reduction of
costs is directly reflected in the bid prices as shown in Theorem 5.2. Consequently,
by simultaneously realizing synergy effects and reducing costs, service providers
run into a stronger price competition which is constantly decreasing their payoffs. Looking at service providers as a collectivity, realizing synergy effects by
offering bundled solutions decreases the overall provider surplus.

6.3.4 Strategic Recommendations
Based on the results described in Section 6.3.3, the following coarse-grained
strategic recommendations regarding single service offers and bundled forms are
derived.
For less competitive service offers, a bundling strategy leads to a significantly higher
expected payoff than an unbundling strategy and increases the likelihood of being allocated if synergy effects can be realized. Less competitive means that these service
offers are either not pricewise competitive or that their QoS characteristics are
not sufficiently valuable for the service requesters (or both). Thus, by combining their offers with more attractive components – although bearing the loss of
interoperability as edges to adjacent service offers are pruned – less competitive

192

CHAPTER 6. NUMERICAL RESULTS

service providers increase their chance of being allocated and manage to increase
their payoff at the same time.
Service providers that are initially capable of competing successfully within the service value network have a higher chance of staying allocated and also face a higher expected payoff by following an unbundling strategy instead of a bundling strategy even
though synergy effects lie idle. In this case, a loss of interoperability through the
merge with another service offer even if compensated by a reduction of costs is
not advantageous as it increases the risk of being less favorable from a requester
perspective.

Part IV
Finale

Chapter 7
Conclusion & Outlook
This explosion of large-scale e-commerce poses new computational challenges that stem
from the need to understand incentives. Because individuals and organizations that own
and operate networked computers and systems are autonomous, they will generally act
to maximize their own self-interest – a notion that is absent from traditional algorithm
design.
[FPP09]
oncluding the work at hand, this chapter points out the key contributions
in Section 7.1 followed by an elaboration of open questions and future research directions that are closely related to this work in Section 7.2. Section 7.3
briefly outlines research streams and future challenges that complement the topics addressed in the work at hand.

C

7.1 Contribution
The key objective of this work is to design a mechanism that enables the coordination of value generation in service value networks which requires that it is
on the one hand theoretically sound and on the other hand applicable in the context of electronic services and their composition. It is a well-known result from
algorithmic or computational mechanism design [NR01, DJP03] and market engineering [WHN03, Neu04] that these theoretical and practical goals are oftentimes
conflicting which requires reasonable solutions regarding these trade-offs to satisfy the requirements upon a suitable mechanism in a certain domain. Addressing these challenges and satisfying detailed requirements derived from a thor-

196

CHAPTER 7. CONCLUSION & OUTLOOK

ough environmental analysis, the work at hand extends the body of research on
mechanisms for trading combinatorial entities in distributed environments with
special focus on sequential compositions of service components in service value
networks. The fact that service compositions only generate value for requesters
that expose a feasible order of their service components imposes novel challenges
on an adequate coordination mechanism.
A thorough mechanism design requires an in-depth understanding of the economic and technical environment, i.e. the trading objects, the market participants,
and the characteristics of the surrounding environment. Hence, the intention of
the following research question is to lay the groundwork for the design, implementation and evaluation of an adequate mechanism that enables the trade of
composite services in service value networks.

Research Question 1 ≺ E NVIRONMENTAL A NALYSIS ≻ . What are
the characteristics of service value networks and complex services, and
what are resulting economic and applicability requirements upon a mechanism to coordinate value creation?
Addressing this question, characteristics and definition of tangibles, intangibles and services are developed and discussed in Section 2.1.1. This discussion
is followed by an analysis of different types of services categorized by a service
decomposition model in Section 2.1.2. Especially complex services constituting the
final outcome of the value creation process in service value networks through
the realization of a sequence of modularized service offers is in the focus of this
analysis. The concept of traditional services, e-services, software services, Web services
and related technical concepts such as service-oriented architectures are analyzed
and their key characteristics are outlined in Section 2.1.3. Based on these results, a
clear understanding of service value networks is provided in Section 2.1.4 by defining their characteristics, their structure, and their components, and by filling the
lack of definitions in current related literature. The discussion about service value
networks which embody the trading environment subject to the work at hand
is followed by an analysis of economic and applicability requirements upon an
adequate mechanism for coordinating value creation in service value networks
in Section 2.2.4.1. Based on these requirements, current approaches which are
closely related to this work are analyzed and existing research gaps are identified
in Section 2.2.4.2. In summary, the environmental analysis and resulting requirement analysis serves as a starting point for further research.

7.1. CONTRIBUTION

197

Research Question 2 focuses on the core contribution: The development of an
adequate multidimensional and scalable auction mechanism to coordinate value
creation in service value networks.

Research Question 2 ≺ M ECHANISM D ESIGN ≻ . How can a scalable,
multidimensional auction mechanism for allocating and pricing of complex services in service value networks be designed that limits strategic
behavior of service providers?
The question is addressed by the development of an abstract model of service
value networks that captures the key characteristics and components in a comprehensive manner in Section 3.1. As part of the mechanism, a bidding language is
provided that enables the specification of multidimensional service offers and
service requests in Section 3.2. To allow for the expression of the service requester’s preferences for different QoS characteristics and prices of complex services, the specification of a scoring function is developed. Finally, the core mechanism – the Complex Service Auction (CSA) – consisting of an allocation and transfer function which implements valuable properties that are analyzed in detail in
the evaluation part, is introduced in Section 3.3. A process model and an adequate architecture of the CSA from a technical perspective are presented in Section 3.5. Focusing on a computational tractable implementation of the auction
mechanism, an algorithm is presented in Section 3.6 that solves the winner determination problem in polynomial time regarding the number of service offers and
feasible service compositions.
Focusing on the applicability of the proposed auction model in real-world
scenarios such as a Web-based intermediation service, Research Question 3 states
additional requirements and addresses the challenge of developing necessary extensions to the core mechanism in order to be applicable in practical settings.

Research Question 3 ≺ A PPLICABILITY E XTENSIONS ≻ . How can an
auction mechanism be extended to support complex QoS characteristics
and service level enforcement? How can the pricing scheme be modified in
order to achieve budget balance and incentivize interoperability endeavors
of service providers?

198

CHAPTER 7. CONCLUSION & OUTLOOK

In order to provide trust and assurance of service quality, service level enforcement is an inevitable applicability aspect. In Section 4.1, the mechanism
is enriched by a compensation function which incorporates ex-post information
about each service’s performance in order to impose penalties if necessary. The
compensation function provides valuable economic properties which are analyzed in detail in the evaluation part. Addressing the challenge of supporting
complex QoS characteristics, a common conceptualization of quality attributes
and their description, aggregation and enforcement from an economic and technical perspective is provided. The auction mechanism is extended in order to
support complex QoS characteristics by means of rule-based semantic concepts and
a toolbox of adequate aggregation operations in Section 4.3.
Another central requirement upon a mechanism from an economic perspective is budget balance which is an important property for a mechanism in order
to be sustainable in the long-run as a continuous external subsidization is neither
reasonable nor profitable for e.g. a platform provider and its business model. It
is well-known from impossibility results in mechanism design that the achievement of certain combinations of economic desiderata is not possible. Addressing
the second part of Research Question 3, an extended transfer function – the Interoperability Transfer Function (ITF) – is developed in Section 6.2 which restores
budget balance by sacrificing incentive compatibility to a certain extent and at the
same time incentivizes service providers to increase their services’ degree of interoperability, i.e. to increase the capability of their offered services to communicate and
function with other services within the service value network which is shown
addressing Question 4.

Research Question 4 ≺ E VALUATION ≻ . How can an auction mechanism be analytically and numerically evaluated regarding its economic
properties as well as cooperation and bundling strategies of service
providers?
Focusing on central economic properties of a mechanism and the implemented social choice function, Research Question 4 is firstly addressed in Chapter
5 by an analytical evaluation which shows that the complex service auction implements a social choice function that is incentive compatible and individual rational
for service providers (Section 5.1). The mechanism is strategyproof with respect
to all dimensions of service providers’ bids, i.e. the truthful announcement of private information on QoS attributes and valuations of offered services is an equi-

7.1. CONTRIBUTION

199

librium in dominant strategies. Consequently, if the service requester announces
its accurate preferences for different outcomes, the social choice is allocative efficient as it is shown in Section A.3. Based on a model of cooperation provided in
Section 5.2, it is further shown that there exist mutually beneficial ex-ante agreements between service providers that face the opportunity to customize their service offers in order to reduce internal costs.
Following a numerical research method in Chapter 6, the extended budgetbalanced transfer function ITF is firstly evaluated with respect to its robustness
against misreporting of service providers by means of simulation-based analysis
in Section 6.1. The question is more precisely: To what degree is it beneficial for
service providers to deviate from their true valuation? Results show that even
in settings with a low level of competition strategic behavior of service providers
is tremendously limited as a deviation from a truth-telling strategy is not significantly beneficial. Despite of the incentives that limit service providers’ strategic
behavior, the ITF rewards service providers to increase their services’ degree of
interoperability. This property is elaborated in detail in Section 6.2 by means of
an agent-based simulation. Compared to an equal transfer function which distributes available surplus equally among allocated service providers, it is shown
that the ITF extension implements incentives to foster a higher overall degree of interoperability in settings with a low level of competition and up to a certain level
of proportionate investment costs for customization.
Focusing on cooperation models in the form of offering bundled services, the
question arises whether it is beneficial to offer bundled services which decreases
flexibility but leverages synergy effects or if it is beneficial to offer single highly
specialized services that are more flexibly composable into various complex service instances. By means of an agent-based simulation with reinforcement learning, this question is addressed in Section 6.3. More precisely there are two main
strategies analyzed: Competing in quality through differentiation and flexibility and competing in price through bundling synergies as cost reduction. Results show that in general service providers that own services within the service
value network which are highly competitive, i.e. they are likely to be allocated,
act best by following an unbundling strategy. In contrary, for service providers
with less competitive service offers it is beneficial to form bundled service offers
while leveraging synergy effects.

200

7.2

CHAPTER 7. CONCLUSION & OUTLOOK

Open Questions

Based on the above mentioned results, there is a number of possible future
research directions and open questions which are briefly addressed in the
remainder of this section.
Allocation computation in the context of sophisticated control logic
The allocation function of the complex service auction computes the “shortest”
path in graphs and is therefore only capable of allocating rudimentary flow logic
in the form of sequential compositions whereas e.g. AND-states have to be split
up in separate statecharts and different auction processes. Such an approach is
sufficient for the allocation of more granular service components that are iteratively composed into a complex service.
However, more sophisticated flow logic increases the complexity of finding
feasible allocations that embody a flawless instantiation of a complex service
from a technical perspective. This leads directly to the questions of how more complex control logic (e.g. AND-states, loops, branches, conditional flows) can be covered by
an allocation function? However, a more complex allocation problem that results
from a more powerful control logic of complex services directly leads to an
increase of computational complexity with respect to solving the winner determination problem while assuring feasible solutions from a technical perspective.
This hinders the satisfaction of Requirement 5 which stresses the importance of
computational tractable algorithms to solve the winner determination problem in
polynomial time for the application in online systems. Addressing this challenge,
heuristics might be a reasonable approach to solve the allocation problem in
the context of complex services that expose highly sophisticated control logic.
Nevertheless, in the absence of an optimal solution, the central Requirement 1 of
allocative efficiency is not fully satisfied depending on the degree of optimality
of the heuristic allocation algorithm. In case the mechanism is designed to
foster an incentive compatible social choice, a suboptimal solution of the winner
determination problem becomes critical from an economic perspective. The
heuristic has to satisfy certain properties such as monotonicity – i.e. an allocated
participant in the complex service auction cannot drop out of the allocation by
decreasing its bid price – in order to retain truthfulness [MN08a, NS06].
Allocation and pricing of people services

7.2. OPEN QUESTIONS

201

Hybrid complex services that involve electronic and human activities impose
new challenges from an economic and organizational perspective. So far,
micro-task markets such as Amazon’s Mechanical Turk1 provide a platform to
leverage the power of human intelligence – the so called crowdsourcing – for
highly specialized tasks such as image recognition. A pool of human individuals
encapsulated by well-defined interfaces can be integrated in hybrid processes.
A seamless integration of human work force in automated compositions of
multiple services opens up further research questions to be addressed in the
future. How can people services sufficiently be described and integrated into service
value networks and the coordination of value creation? The challenges that arise
from the service characteristic C 2.5 describing the fuzzyness of input and
output parameters and capabilities are partly addressed by the high degree of
standardization and specified description languages (e.g. WSDL, WS-BPEL)
which are common sense. Nevertheless, in the context of people services, these
challenges arise anew as human work force is hardly parameterizable and the
scope, capabilities and quality of the output vary widely. Thus, incorporating
human activities in automated processes requires well-specified task descriptions [KCS08]. As inputs and outputs have to be carefully described the issue of
quality assurance becomes even more crucial. The question arises of how these
activities can be monitored in order to compute compensation transfers and apply service
level enforcement mechanisms.
Allocation and pricing of highly complex application services
As introduced in Section 2.1.4.3, a trend towards simplification is observable
that enables an agile composition of highly specialized services that expose
puristic interfaces and descriptions e.g. as in RESTful architectures based on the
CRUD paradigm2 . Nevertheless as outlined in Section 2.1.2.3, complex services
consist of service components that can themselves be a utility, elementary or
complex service (analogue to the recursive specification in WS-BPEL). As the
granularity of service components decreases, the complexity of their interfaces
and necessary descriptions grows which implies new challenges for the mechanism. As a result of the increased interface complexity and the semantic of
input and output values, the computational complexity of the algorithm that
solves the respective winner determination problem augments as well. This
conflicts with the requirement of computation tractability which is inevitable for
a mechanism in order to be realized in online systems. Furthermore, investment
1 http://mturk.com/
2 CRUD

stands for the persistent functions create, read, update, and delete.

202

CHAPTER 7. CONCLUSION & OUTLOOK

costs for the customization of service offers’ interfaces fostering a higher degree
of interoperability rise which results in more static and less multifaceted service
value networks. More complex service descriptions and interfaces also impact
the elicitation and expression of preferences for different QoS levels. Service
requesters have to specify their preferences for different outcomes regarding the
complex service’s attributes which leads to the question of how service consumers
can be supported by tools and concepts to enable the elicitation and expression of
preferences for complex multidimensional QoS characteristics.
Multi-layered markets for utility and complex services
Service components that are traded in e.g. the complex service auction require
low level resource services (utility services) to enable their deployment and assure scalability during run-time. Focusing on the infrastructure layer, it is also
reasonable to trade utility services themselves independent from mechanisms to
allocate and price complex services. Nevertheless, utility services expose different characteristics and therefore impose different requirements upon suitable
market mechanisms [Neu04]. There are several market mechanisms for the trade
of utility services proposed in literature [Sto09, Sch07]. Combining the trade of
utility and complex services as depicted in Figure 7.1, the question arises of how
a multi-layered market can be designed in order to enable a seamless allocation and pricing of complex services and corresponding utility service which are required by the layer
above.

7.3

Complementary Research

Besides research directions closely related to the work at hand as illustrated in
Section 7.2, this section points out research questions which are partly complementary to this work and therefore possibly enrich certain aspects.
Alternative design goals and business models for platform providers
The design of the complex service auction mechanisms implements a social
choice that is allocative efficient, i.e. it maximizes welfare. Although this is a
commonly desired design goal that has valuable implications for all participants,
there are alternative design desiderata that are favorable for certain stake holders.
From the perspective of a platform provider that offers an intermediation service
to e.g. a service value network, a revenue maximizing social choice is certainly

7.3. COMPLEMENTARY RESEARCH

203

Complex Service Auction
Abstract
Composition

binding

Service

allocation

Resource

binding

binding

Service

allocation

Service

allocation

Resource

allocation

Resource

Resource

Resource Market

Figure 7.1
Multi-layered market for complex services and resources.

beneficial compared to an optimal solution from a utilitarian point of view if
e.g. the intermediary receives a fraction of the each service provider’s revenue.
Research that deals with auction formats which are designed to maximize the
revenue for e.g. the seller of an economic entity is well-known in literature as
optimal auction design [Mye81]. Focusing on procurement scenarios where price
and quality matters, optimal buying mechanisms that intent to maximize the
buyer’s expected payoff are evaluated in [CIoWM93, AC05]. Looking at optimal
auction designs and revenue models for platform providers, the question of how
to design a successful business model for providers of intermediation services arises.
The structure of “traditional” business model types might not be sufficient in
order to address the requirements that result from highly agile and distributed
environments such as service value networks [MWL+ 06]. Recall that a mechanism in order to be sustainable in the long-run must satisfy the economic design
desideratum of budget balance (cp. Desideratum 2.4) in order to avoid the need
for external subsidization as well as the desideratum of individual rationality
(cp. Desideratum 2.3) to provide incentives to participate in the market. In

204

CHAPTER 7. CONCLUSION & OUTLOOK

this regard, revenue models for platform providers that stipulate for charging
participation fees may violate individual rationality and (strong) budget balance.
However, in certain cases it might be reasonable for a e.g. a public institution
to subsidize an efficient market. Nevertheless, such implications of the revenue
model on economic properties of a mechanism implementation must be carefully
analyzed and considered when constructing and structuring novel business
models.
Preference elicitation
It is a typical assumption in game theory and especially mechanism design
research that market participants know their true valuations. However, elicitation of preferences especially in multidimensional settings (e.g. preferences for
different QoS levels of multiple service attributes and their semantics) embodies
a complex task for service providers and requesters. In combinatorial settings
(cp. the complex service auction), participants must be capable of expressing
preferences for different combinations of e.g. service components. This is a
crucial task as it implicitly requires the comparison of a large set of alternative
combinations. Although preference elicitation embodies a prerequisite of any
market-based approach, research in this area is still in its infancy [SNP+ 05]. For
instance, prominent approaches for the elicitation of preferences – e.g. in the
context of services – are conjoint analysis [GR71, LT64] and analytical hierarchical
processing [Saa80, Saa08]. A major shortcoming of these approaches is that they
become infeasible in settings with large sets of attributes which are common in
e.g. service markets.
Automated bidding
Having suitably determined the true valuations for the trading object, a bidding
strategy must be developed in order to successfully participate in the market.
With preference elicitation as a prerequisite, developing such a bidding strategy
and efficiently communicating it to the market is another complex task to be
solved by participants. In order to support users in evaluating and expressing
a beneficial bidding strategy, tools for automated bidding are a promising approach to overcome complexity and effort [MMW06, Tes01]. Another advantage
of facilitating tools to interact with markets is that there is no need to constantly
monitor market activities and incorporate information in the bidding strategy as
this information can be processed and interpreted by automatic bidding agents.
Although these tools can simplify market interaction, participants want to keep

7.4. FINAL REMARKS

205

control over their strategy and resulting actions. Hence, hybrid models are
more practical as they still hide complexity and simplify the trading process but
also allow for a manual interaction triggered by the user which might also be
necessary for legal reasons. Another success factor of automatic trading agents
is the parameter selection and their customization for the application in different
market mechanisms that impose different requirements upon beneficial strategies. Addressing these challenges, strategies for bidding agents are developed
that successfully perform in multiple settings and market mechanisms [Bor09].
Reputation mechanisms
Another class of mechanisms that enable coordination of distributed activities in
a broader sense are reputation mechanisms. Using feedback information, reputation mechanisms aim at building trust in environments with self-interested participants [BKO02]. Reputation mechanisms aggregate trading histories of e.g. service providers and requesters and compute a metric which indicates the trustworthiness of market participants. This information can be incorporated in the
allocation and pricing procedure providing additional characteristics of the trading parties. For example, the lower the reputation of a service provider, the less
likely is the allocation of services offered by this service provider. Although it
is well-known in literature that reputation mechanisms have proven to perform
well in distributed systems in the absence of a central instance such as in peerto-peer networks [WV03], it is an interesting question of how such reputation
components can be designed and realized additionally to a central market mechanism. Challenges that arise in this context are e.g. how to make truthful revelation of reputation information an optimal strategy market participants [JF03].
For a detailed survey on state-of-the-art trust and reputation systems for service
provision via electronic networks, the interested reader is referred to [JIB07].

7.4 Final Remarks
Services become a central component of value creation in today’s society. Novel
technical, economic, and organizational challenges arise from their unique nature
as services’ provision and consumption coincide in time [Hil77]. Recognizing
and understanding the importance of an efficient design, production, and provision of services under the presence of their special characteristics is inevitable
for individuals and the society to compete in today’s global economy. Especially
rapid service innovation driven by the power of modularity that is inherent in the

206

CHAPTER 7. CONCLUSION & OUTLOOK

concept of services [BC00] embodies the success factor in service-centric environments. However, when composing distributed service activities, the question of
an efficient form of coordination comes to light and turns out to be fundamental
to govern distributed value creation. As complex services are living artifacts that
generally exist under the ownership of different economic entities which are selfinterested in nature, system-wide goals are hard to achieve as they mostly collide
with individual objectives and are therefore not intrinsically pursued [Par01].
The approach of mechanism design [Hur73, Mye88] – and the revelation principle [Gib73, Mye82] as the central possibility result – considers economic problems in situations where individuals’ private information and actions are hard
to monitor. The main objective is to design mechanisms that provide incentives
for individuals to “share information and exert efforts” [Mye88] which implements a social choice that constitutes a system-wide solution. Hence, although
individuals (e.g. service owners) seek to maximize their utility based on their private information about their preferences for different outcomes, they inevitably
contribute to the achievement of a global goal.
Following the approach of mechanism design, this work provided an auction mechanism which enables the trade of composite services in service value
networks. The mechanism constitutes an equilibrium in which truth-revelation
of private multidimensional types is a weakly dominant strategy for all service
providers and implements a social choice that maximizes the utility across all
participants. The mechanism exposes valuable properties as it is not beneficial
for individuals to lie about their private information, neither on their services’
QoS characteristics nor on corresponding private valuations. Furthermore, participation is voluntary and beneficial for service providers and the mechanism
results in an allocation which is optimal and constitutes a system-wide welfare
maximizing solution.
The work at hand shows that mechanism design in combination with technical, computational, and applicability considerations is a promising approach to
efficiently govern distributed service activities in agile and fast changing environments such as service value networks. However, open questions and complementary research directions constitute further challenges that need to be mastered in
an integrated manner in order to leverage the power of algorithmic mechanism
design and to move the results at hand from theory to practice, to innovation.

Appendix A
Appendix
A.1 Formal Notation
Table A.1: Notation of abstract model and mechanism implementation.
Notation

Meaning

G = (V, E)

Service Value Network

V \ { v s , v f } = { v1 , . . . , v N }

N Service offers/services/nodes with i, j ∈ V are arbitrary
services

vs , v f ∈ V

Source and sink node

E = {eij |i, j ∈ V }

Technical feasible combinations of services

f ∈F

Feasible path from source to sink that is an instantiation
of a complex service f

S = { s1 , . . . , s Q }

Q Service providers

σ:S→V

Ownership function

A j = { a1j , . . . , a Lj }

Configuration of service j with alj is the attribute value of
type l ∈ L

cij

Interoperability costs of service j as a successor of service
i

A f = (A1f , . . . , A Lf )

Configuration of complex service f with Alf is the attribute value of type l ∈ L

S : A → [0; 1]

Scoring function of service requester

208

APPENDIX A. APPENDIX
Table A.1: Notation of abstract model and mechanism implementation.

Notation

Meaning

Λ = ( λ1 , . . . , λ L )

Preference structure of service requester with λl is the
weight for attribute type l ∈ L

Γ = (γ1B , γ1T , . . . , γBL , γTL )

Preference boundaries of service requester with γlB is the
lower and γTl is the upper boundary for attribute type l ∈
L

α

Willingness to pay of service requester for a complex service f with S(A f ) = 1

A.2

Incentive Compatibility

Proof A.1 [T HEOREM 5.2]. 1 Let F−s denotes the set of all feasible paths from source
to sink in the reduced graph G−s without every service offer owned by service provider s
and corresponding incoming and outgoing edges. Let further f ∗ denote the path which is
∗ be the utility of path f ∗ in the
allocated by o. Let U ∗ be the utility of path f ∗ . Let U−
s
−s
∗
s
∗
reduced graph G−s . Let Ũ denote the overall utility of the allocated path f computed
based on the verified attribute values ã1j , . . . , ã Lj of the verified configurations à j of all
service offers j ∈ σ (s). Let Ẽs denote the set of edges with Ẽs = {eij |eij ∈ o, j ∈ σ (s), i ∈
τ ( j)}. Service provider s wants to maximize its expected payoff:
s

∗

E(π ) = P(U >

U−∗ s )

∗
E(π s ) = P(U ∗ > U−
s)
∗
E(π s ) = P(U ∗ > U−
s)

"

∑ pij + (U

"

∑ pij +

∗

− U−∗ s ) − ∆tcomp,s

Ẽs

" Ẽ

− ∑ cij
Ẽs

#

(U ∗ − U−∗ s ) − (U ∗ − Ũ ∗s ) − ∑ cij

s



∗s
∗
p
+
Ũ
−
U
ij
∑
−s − ∑ cij
Ẽs

Ẽs

#




Ẽs

#

This leads to two possible cases:
1. If s’s payoff π s is positive, it wants to maximize the probability of being allocated
which leads to the problem statement
max

pij ,A j | j∈σ(s),i ∈τ ( j)
1 This

∗
P(U ∗ > U−
s)

proof is based on the argumentation in [MMV94]

A.3. ALLOCATIVE EFFICIENCY

st.

"

∑ pij +

209
#

Ũ ∗s − U−∗ s − ∑ cij > 0



Ẽs

Ẽs

∗ .
From the side condition it follows directly that ∑ Ẽs pij + Ũ ∗s − ∑ Ẽs cij > U−
s
Hence, P(·) is maximized by setting pij = cij and A j = Ã j , ∀ j ∈ σ (s), i ∈ τ ( j)
as this leads to U ∗ = Ũ ∗s and finally to P(·) = 1.

2. If s’s payoff π s is negative, it wants to minimize the probability of being allocated
which leads to the problem statement
min

pij ,A j | j∈σ(s),i ∈τ ( j)

st.

"

∑ pij +
Ẽs

∗
P(U ∗ > U−
s)

#

Ũ ∗s − U−∗ s − ∑ cij < 0



Ẽs

Symmetrically to the first case, it follows directly from the side condition that
∗ . Hence, P (·) is minimized by setting p = c and
∑ Ẽs pij + Ũ ∗s − ∑ Ẽs cij < U−
ij
ij
s
A j = Ã j , ∀ j ∈ σ (s), i ∈ τ ( j) as this leads to U ∗ = Ũ ∗s and finally to P(·) = 0.
In any case one solution that maximizes the expected payoff E(π s ) of service provider
s is pij = cij and A j = Ã j , ∀ j ∈ σ (s), i ∈ τ ( j). This solution is the truth-telling strategy as s reveals its true multidimensional type. Although truth-telling is not the only
solution, service provider s does not benefit from deviation as its strategy does not influence its payoff as shown in Corollary 5.2 which makes truth-telling with respect to the
multidimensional types of service providers (configuration and price) a weakly dominant
strategy.


A.3 Allocative Efficiency
This section briefly shows that under the assumption of the absence of strategic
behavior of the service requester, the complex service auction always leads to a
welfare maximizing outcome:
Corollary A.1 [W ELFARE M AXIMIZATION ]. The allocation function according to


(3.8) argmax f ∈ F αS(A f ) − P f is efficient as it maximizes the system’s welfare with
α representing the requester’s maximal willingness to pay, S(A f ) its score for the configuration of the complex service f and P f the sum of all price bids of service providers
that own service offers that have incoming edges on the path f .

210

APPENDIX A. APPENDIX

Proof A.1 [C OROLLARY A.1]. Let U R = αS(A f ) − T f denote the service requester’s
utility with α represents the requester’s maximal willingness to pay, S(A f ) the requester’s score for the configuration of the complex service f and T f the sum of all transfer
payments to allocated providers according to (4.2). Furthermore let U s = ts − cs be the
utility of service provider s ∈ S. The system’s welfare W f based on an allocated path f is
the sum of consumer (requester) and providers’ surplus such that
Wf = U R +

∑ Us
s∈S

W f = αS(A f ) − T f +

∑ (ts − cs )
s∈S

W f = αS(A f ) − T f + T f −

∑ cs
s∈S

W f = αS(A f ) −

∑c

s

s∈S

Based on the results of Theorem 5.2 truth-telling with respect to configuration and price
is a weakly dominant strategy for all service providers so it can be directly concluded that
W f ∗ = αS(Ã f ∗ ) − P f ∗



A.4

Manipulation Robustness
Table A.2: Utility for a single manipulating service provider with
12 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-50%

0.0423

0.5865

0.0793

-0.0209

-0.6871

0.1022

-45%

0.0506

0.7007

0.0634

-0.0113

-0.3802

0.0860

-40%

0.0562

0.7789

0.0506

-0.0009

-0.0308

0.0714

-35%

0.0604

0.8359

0.0413

0.0055

0.1809

0.0596

-30%

0.0631

0.8741

0.0334

0.0113

0.3645

0.0478

A.4. MANIPULATION ROBUSTNESS

211

Table A.2: Utility for a single manipulating service provider with
12 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-25%

0.0656

0.9092

0.0275

0.0158

0.5254

0.0394

-20%

0.0693

0.9603

0.0136

0.0194

0.6763

0.0264

-15%

0.0702

0.9724

0.0103

0.0235

0.7919

0.0196

-10%

0.0715

0.9904

0.0050

0.0250

0.8795

0.0144

-5%

0.0721

0.9981

0.0015

0.0291

0.9477

0.0066

0%

0.0722

1.0000

0.0000

0.0302

1.0000

0.0000

5%

0.0721

0.9982

0.0012

0.0326

1.0378***

0.0075

10%

0.0715

0.9906

0.0050

0.0317

1.0688***

0.0125

15%

0.0711

0.9847

0.0074

0.0302

1.1036***

0.0148

20%

0.0705

0.9771

0.0097

0.0327

1.0968***

0.0199

25%

0.0704

0.9750

0.0100

0.0365

1.1194***

0.0238

30%

0.0703

0.9738

0.0102

0.0393

1.1380***

0.0283

35%

0.0702

0.9721

0.0109

0.0397

1.1700***

0.0328

40%

0.0696

0.9638

0.0137

0.0384

1.1776***

0.0355

45%

0.0690

0.9554

0.0184

0.0422

1.1672***

0.0402

50%

0.0673

0.9320

0.0261

0.0379

1.1774***

0.0435

55%

0.0664

0.9201

0.0304

0.0383

1.1507***

0.0455

60%

0.0640

0.8870

0.0383

0.0384

1.1016***

0.0445

65%

0.0636

0.8806

0.0388

0.0390

1.0768***

0.0480

70%

0.0627

0.8691

0.0424

0.0377

1.0866***

0.0486

75%

0.0605

0.8381

0.0504

0.0364

1.0366**

0.0438

80%

0.0603

0.8354

0.0508

0.0355

1.0535***

0.0449

85%

0.0602

0.8335

0.0511

0.0365

1.0537***

0.0470

90%

0.0596

0.8251

0.0521

0.0362

1.0233*

0.0475

212

APPENDIX A. APPENDIX
Table A.2: Utility for a single manipulating service provider with
12 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

95%

0.0592

0.8206

0.0529

0.0366

1.0422***

0.0489

100%

0.0591

0.8181

0.0533

0.0351

1.0581***

0.0508

105%

0.0580

0.8039

0.0557

0.0362

1.0204

0.0534

110%

0.0578

0.8006

0.0560

0.0378

1.0091

0.0537

115%

0.0566

0.7838

0.0605

0.0352

1.0146

0.0518

120%

0.0554

0.7670

0.0632

0.0354

0.9652

0.0524

125%

0.0552

0.7641

0.0634

0.0366

0.9901

0.0549

130%

0.0550

0.7613

0.0639

0.0314

0.9824

0.0543

135%

0.0540

0.7484

0.0660

0.0349

0.9504

0.0548

140%

0.0534

0.7395

0.0672

0.0317

0.9529

0.0576

145%

0.0534

0.7395

0.0672

0.0371

0.9328

0.0566

150%

0.0526

0.7285

0.0685

0.0344

0.9557

0.0581

Table A.3: Utility for a single manipulating service provider with
16 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-50%

0.0171

0.4002

0.0757

-0.0081

-0.3140

0.0845

-45%

0.0247

0.5793

0.0597

0.0020

0.0757

0.0678

-40%

0.0300

0.7035

0.0465

0.0072

0.2799

0.0546

-35%

0.0340

0.7977

0.0361

0.0107

0.4300

0.0439

-30%

0.0383

0.8983

0.0217

0.0158

0.6344

0.0315

A.4. MANIPULATION ROBUSTNESS

213

Table A.3: Utility for a single manipulating service provider with
16 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-25%

0.0397

0.9310

0.0163

0.0181

0.7444

0.0234

-20%

0.0413

0.9687

0.0095

0.0209

0.8354

0.0176

-15%

0.0418

0.9814

0.0067

0.0247

0.9011

0.0138

-10%

0.0424

0.9954

0.0027

0.0234

0.9331

0.0083

-5%

0.0426

0.9988

0.0010

0.0252

0.9748

0.0044

0%

0.0426

1.0000

0.0000

0.0248

1.0000

0.0000

5%

0.0425

0.9981

0.0012

0.0265

1.0175***

0.0046

10%

0.0425

0.9980

0.0013

0.0263

1.0453***

0.0070

15%

0.0423

0.9927

0.0035

0.0273

1.0557***

0.0102

20%

0.0420

0.9858

0.0055

0.0274

1.0659***

0.0131

25%

0.0415

0.9744

0.0082

0.0277

1.0570***

0.0157

30%

0.0403

0.9466

0.0144

0.0276

1.0334***

0.0213

35%

0.0402

0.9444

0.0148

0.0266

1.0529***

0.0228

40%

0.0402

0.9434

0.0149

0.0283

1.0562***

0.0246

45%

0.0399

0.9361

0.0162

0.0291

1.0416***

0.0259

50%

0.0394

0.9244

0.0180

0.0271

1.0570***

0.0282

55%

0.0387

0.9079

0.0212

0.0272

1.0326**

0.0304

60%

0.0382

0.8974

0.0227

0.0281

1.0256*

0.0309

65%

0.0377

0.8839

0.0252

0.0272

1.0037

0.0307

70%

0.0373

0.8757

0.0261

0.0267

1.0170

0.0325

75%

0.0367

0.8623

0.0288

0.0277

0.9994

0.0331

80%

0.0359

0.8418

0.0315

0.0268

0.9777

0.0376

85%

0.0355

0.8333

0.0330

0.0262

0.9778

0.0366

90%

0.0352

0.8259

0.0339

0.0268

0.9607

0.0391

214

APPENDIX A. APPENDIX
Table A.3: Utility for a single manipulating service provider with
16 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

95%

0.0350

0.8204

0.0344

0.0274

0.9673

0.0372

100%

0.0348

0.8168

0.0348

0.0276

0.9411

0.0395

105%

0.0335

0.7854

0.0405

0.0266

0.9083

0.0372

110%

0.0329

0.7724

0.0414

0.0254

0.8877

0.0383

115%

0.0324

0.7599

0.0430

0.0239

0.8655

0.0404

120%

0.0320

0.7504

0.0437

0.0245

0.8816

0.0412

125%

0.0314

0.7376

0.0463

0.0237

0.8639

0.0403

130%

0.0314

0.7376

0.0463

0.0240

0.8616

0.0420

135%

0.0306

0.7191

0.0485

0.0238

0.8278

0.0443

140%

0.0305

0.7153

0.0487

0.0246

0.8350

0.0444

145%

0.0305

0.7153

0.0487

0.0245

0.8290

0.0434

150%

0.0299

0.7012

0.0506

0.0234

0.8274

0.0440

Table A.4: Utility for a single manipulating service provider with
20 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-50%

0.0025

0.1122

0.0630

-0.0111

-0.7315

0.0741

-45%

0.0075

0.3412

0.0502

-0.0032

-0.1944

0.0588

-40%

0.0107

0.4870

0.0425

0.0003

0.0187

0.0495

-35%

0.0147

0.6651

0.0316

0.0065

0.3905

0.0373

-30%

0.0173

0.7854

0.0231

0.0090

0.5533

0.0292

A.4. MANIPULATION ROBUSTNESS

215

Table A.4: Utility for a single manipulating service provider with
20 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-25%

0.0194

0.8822

0.0155

0.0129

0.7391

0.0208

-20%

0.0208

0.9444

0.0089

0.0137

0.8251

0.0146

-15%

0.0212

0.9621

0.0068

0.0135

0.8736

0.0102

-10%

0.0219

0.9916

0.0020

0.0150

0.9434

0.0063

-5%

0.0220

0.9958

0.0011

0.0161

0.9756

0.0031

0%

0.0220

1.0000

0.0000

0.0167

1.0000

0.0000

5%

0.0220

0.9965

0.0009

0.0156

1.0155***

0.0027

10%

0.0219

0.9920

0.0017

0.0169

1.0298***

0.0059

15%

0.0217

0.9855

0.0032

0.0160

1.0339***

0.0074

20%

0.0215

0.9748

0.0051

0.0168

1.0227***

0.0086

25%

0.0210

0.9543

0.0079

0.0168

0.9996

0.0107

30%

0.0205

0.9300

0.0108

0.0157

0.9929

0.0111

35%

0.0199

0.9050

0.0135

0.0152

0.9629

0.0131

40%

0.0195

0.8849

0.0156

0.0150

0.9266

0.0143

45%

0.0192

0.8691

0.0167

0.0151

0.9063

0.0156

50%

0.0191

0.8662

0.0169

0.0149

0.9129

0.0163

55%

0.0190

0.8604

0.0173

0.0152

0.9012

0.0168

60%

0.0189

0.8562

0.0176

0.0150

0.8881

0.0166

65%

0.0188

0.8536

0.0177

0.0150

0.9143

0.0185

70%

0.0185

0.8387

0.0197

0.0148

0.8794

0.0187

75%

0.0184

0.8350

0.0200

0.0152

0.8847

0.0211

80%

0.0183

0.8324

0.0201

0.0153

0.8847

0.0201

85%

0.0183

0.8295

0.0204

0.0152

0.8771

0.0207

90%

0.0182

0.8246

0.0207

0.0149

0.8776

0.0218

216

APPENDIX A. APPENDIX
Table A.4: Utility for a single manipulating service provider with
20 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

95%

0.0181

0.8198

0.0211

0.0143

0.8751

0.0231

100%

0.0179

0.8125

0.0217

0.0149

0.8526

0.0220

105%

0.0178

0.8075

0.0222

0.0147

0.8461

0.0224

110%

0.0176

0.7988

0.0235

0.0148

0.8480

0.0234

115%

0.0175

0.7925

0.0241

0.0143

0.8359

0.0254

120%

0.0174

0.7888

0.0243

0.0154

0.8303

0.0266

125%

0.0173

0.7856

0.0245

0.0146

0.8280

0.0238

130%

0.0168

0.7602

0.0270

0.0139

0.7904

0.0270

135%

0.0165

0.7487

0.0284

0.0136

0.7826

0.0286

140%

0.0165

0.7474

0.0285

0.0139

0.7947

0.0293

145%

0.0165

0.7474

0.0285

0.0141

0.7801

0.0291

150%

0.0163

0.7397

0.0293

0.0139

0.7869

0.0279

Table A.5: Utility for a single manipulating service provider with
28 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-50%

0.0000

0.0005

0.0501

-0.0048

-0.4739

0.0540

-45%

0.0046

0.3551

0.0371

0.0005

0.0468

0.0411

-40%

0.0081

0.6271

0.0247

0.0037

0.3617

0.0305

-35%

0.0091

0.7086

0.0208

0.0054

0.5255

0.0243

-30%

0.0103

0.8014

0.0152

0.0069

0.6498

0.0191

A.4. MANIPULATION ROBUSTNESS

217

Table A.5: Utility for a single manipulating service provider with
28 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.
Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

-25%

0.0113

0.8765

0.0112

0.0076

0.7570

0.0142

-20%

0.0119

0.9275

0.0070

0.0090

0.8521

0.0100

-15%

0.0124

0.9681

0.0042

0.0095

0.9224

0.0066

-10%

0.0127

0.9908

0.0014

0.0097

0.9500

0.0042

-5%

0.0128

0.9972

0.0007

0.0106

0.9837

0.0023

0%

0.0129

1.0000

0.0000

0.0101

1.0000

0.0000

5%

0.0128

0.9959

0.0009

0.0106

1.0080***

0.0019

10%

0.0127

0.9873

0.0018

0.0108

1.0044

0.0029

15%

0.0124

0.9625

0.0047

0.0104

0.9845

0.0058

20%

0.0122

0.9489

0.0058

0.0101

0.9681

0.0063

25%

0.0121

0.9393

0.0064

0.0101

0.9587

0.0071

30%

0.0120

0.9315

0.0069

0.0107

0.9546

0.0080

35%

0.0119

0.9268

0.0071

0.0106

0.9563

0.0080

40%

0.0119

0.9240

0.0072

0.0099

0.9526

0.0084

45%

0.0117

0.9133

0.0082

0.0098

0.9396

0.0093

50%

0.0116

0.9059

0.0088

0.0098

0.9350

0.0103

55%

0.0116

0.9022

0.0090

0.0098

0.9432

0.0100

60%

0.0113

0.8799

0.0110

0.0099

0.9054

0.0123

65%

0.0111

0.8628

0.0122

0.0095

0.8963

0.0137

70%

0.0109

0.8455

0.0133

0.0098

0.8773

0.0141

75%

0.0107

0.8294

0.0142

0.0095

0.8635

0.0145

80%

0.0106

0.8232

0.0146

0.0094

0.8464

0.0144

85%

0.0104

0.8115

0.0152

0.0094

0.8522

0.0164

90%

0.0104

0.8083

0.0154

0.0092

0.8546

0.0163

218

APPENDIX A. APPENDIX
Table A.5: Utility for a single manipulating service provider with
28 service offers in 4 candidate pools. abs denotes the mean absolute utility and rel the ratio of means of the utility with manipulation and the utility following a truth-telling strategy. sd is the
standard deviation of the mean absolute utility. * denotes significance at the level of p = 0.1, ** at p = 0.05, and *** at p = 0.01.

A.5

Critical Value Transfer

Interoperability Transfer

Manipulation Rate

abs

rel

sd

abs

rel

sd

95%

0.0101

0.7858

0.0169

0.0091

0.8210

0.0167

100%

0.0099

0.7667

0.0181

0.0087

0.7969

0.0187

105%

0.0099

0.7667

0.0181

0.0091

0.8050

0.0190

110%

0.0099

0.7667

0.0181

0.0088

0.8045

0.0183

115%

0.0097

0.7556

0.0190

0.0090

0.7827

0.0190

120%

0.0095

0.7410

0.0199

0.0087

0.7596

0.0212

125%

0.0095

0.7360

0.0201

0.0086

0.7604

0.0202

130%

0.0093

0.7208

0.0216

0.0081

0.7390

0.0229

135%

0.0093

0.7208

0.0216

0.0086

0.7696

0.0220

140%

0.0091

0.7089

0.0223

0.0083

0.7360

0.0228

145%

0.0090

0.7031

0.0226

0.0081

0.7336

0.0232

150%

0.0089

0.6937

0.0231

0.0082

0.7289

0.0224

Bundling Strategies

219

1.0

1.0

A.5. BUNDLING STRATEGIES

0.8
0.6
0.0

0.2

0.4

Fitness
0.0

0.2

0.4

Fitness

0.6

0.8

unbundling
bundling

0

5

10

15

20

25

30

unbundling

Topology

bundling
Strategy

1.0

1.0

(a) 0% cost reduction due to bundling synergies with 32 service offers in 4
candidate pools.

0.8
0.6
0.0

0.2

0.4

Fitness
0.0

0.2

0.4

Fitness

0.6

0.8

unbundling
bundling

0

5

10

15
Topology

20

25

30

unbundling

bundling
Strategy

(b) 50% cost reduction due to bundling synergies with 32 service offers in 4
candidate pools.
Figure A.1
Strategy fitness in different cost reduction scenarios with 32
service offers in 4 candidate pools.

References
[AAA+ 07] Alexandre Alves, Assaf Arkin, Sid Askary, Charlton Barreto, Ben Bloch, Francisco Curbera, Mark Ford, Yaron
Goland, Alejandro Guízar, Neelakantan Kartha, Canyang Kevin
Liu, Rania Khalaf, Dieter König, Mike Marin, Vinkesh
Mehta, Satish Thatte, Danny van der Rijn, Prasad Yendluri, and Alex Yiu. Web Service Business Process Execution Language (WS-BPEL). Technical report, OASIS, 4 2007.
http://docs.oasis-open.org/wsbpel/.
[AB91] B.R. Allen and A.C. Boynton. Information Architecture: In
Search of Efficient Flexibility. MIS Quarterly, 15(4):435–445,
1991.
[AB08] Ben Adida and Mark Birbeck. Resource Description Framework - in - attributes.
Technical report, W3C, 10 2008.
http://www.w3.org/TR/xhtml-rdfa-primer/.
[ABC+ 02] Eric Armstrong, Stephanie Bodoff, Debbie Carson, Maydene
Fisher, Dale Green, and Kim Haase. The Java Web Services Tutorial. Addison-Wesley, 2002.
[AC05] J. Asker and E. Cantillon. Optimal Procurement When Both
Price and Quality Matter. Technical report, 2005.
[AC08] J. Asker and E. Cantillon. Properties of Scoring Auctions. The
RAND Journal of Economics, 39(1):69–85, 2008.
[ACD+ 04] A. Andrieux, K. Czajkowski, A. Dan, K. Keahey, H. Ludwig,
J. Pruyne, J. Rofrano, S. Tuecke, and M. Xu. Web Services Agreement Specification (WS-Agreement). In Global Grid Forum, 2004.
[ACSV04] A. AuYoung, B.N. Chun, A.C. Snoeren, and A. Vahdat. Resource
Allocation in Federated Distributed Computing Infrastructures.

222

REFERENCES
In Proceedings of the 1st Workshop on Operating System and Architectural Support for the On-demand IT InfraStructure, 2004.
[AGB+ 04] Daniel Austin, W. W. Grainger, Abbie Barbir, Christopher Ferris, and Sharad Garg.
Web Services Architecture Requirements.
Technical report, W3C, 2 2004.
http://www.w3.org/TR/wsa-reqs/.
[Ama08] Amazon.
Blog.

Amazon
Web
report,
Amazon,

Services
Technical
5
2008.
http://aws.typepad.com/aws/2008/05/lots-of-bits.html.

[And06] C. Anderson. The Long Tail: How Endless Choice is Creating Unlimited Demand. Random House Business Books, 2006.
[AT07] Aaron Archer and Eva Tardos. Frugal Path Mechanisms. ACM
Transactions on Algorithms, 3(1):3, 2007.
[BBL99] Y. Bakos, E. Brynjolfsson, and D. Lichtman. Shared Information
Goods. The Journal of Law and Economics, 42(1):117–156, 1999.
[BBS08] B. Blau, C. Block, and J. Stösser. How to trade Electronic Services? – Current Status and Open Questions. In Proceedings of
the Joint Conference of the INFORMS section on Group Decision and
Negotiation, the EURO Working Group on Decision and Negotiation
Support, and the EURO Working Group on Decision Support Systems, 2008.
[BBT09] James Broberg, Rajkumar Buyya, and Zahir Tari. MetaCDN:
Harnessing Storage Clouds for High Performance Content Delivery. Journal of Network and Computer Applications, In Press,
Corrected Proof, 2009.
[BC00] C.Y. Baldwin and K.B. Clark. Design Rules: Volume 1: The Power
of Modularity. Mit Press Cambridge, MA, 2000.
[BCC+ 04] Don Box, Erik Christensen, Francisco Curbera, Donald Ferguson, Jeffrey Frey, Marc Hadley, Chris Kaler, David Langworthy, Frank Leymann, Brad Lovering, Steve Lucco, Steve
Millet, Nirmal Mukhi, Mark Nottingham, David Orchard,
John Shewchuk, Eugene Sindambiwe, Tony Storey, Sanjiva Weerawarana, and Steve Winkler. Web Services Ad-

REFERENCES

223
dressing (WS-Addressing).

Technical report, W3C, 8 2004.
http://www.w3.org/Submission/ws-addressing/.
[BCM+ 07] F. Baader, D. Calvanese, D.L. McGuinness, D. Nardi, and P.F.
Patel-Schneider. The Description Logic Handbook. Cambridge
University Press New York, NY, USA, 2007.
[BCM09] B. Blau, T. Conte, and T. Meinl. Coordinating Service Composition. In Proceedings of the 17th European Conference on Information
Systems, 2009.
[BDBD+ 00] Gabe Beged-Dov, Dan Brickley, Rael Dornfest, Ian Davis,
Leigh Dodds, Jonathan Eisenzopf, David Galbraith, R.V. Guha,
Ken MacLeod, Eric Miller, Aaron Swartz, and Eric van der
Vlist. RDF Site Summary (RSS) 1.0. Technical report, 2000.
http://purl.org/rss/1.0/spec/.
[BDF+ 03] P. Barham, B. Dragovic, K. Fraser, S. Hand, T. Harris, A. Ho,
R. Neugebauer, I. Pratt, and A. Warfield. Xen and the Art of Virtualization. ACM SIGOPS Operating Systems Review, 37(5):164–
177, 2003.
[BEA08] BEA. Revised Statistics of Gross Domestic Product by Industry,
2004-2006. Technical report, BEA (Bureau of Economic Analysis), 2008.
[BEK+ 00] Don Box, David Ehnebuske, Gopal Kakivaya, Andrew Layman,
Noah Mendelsohn, Henrik Frystyk Nielsen, Satish Thatte, and
Dave Winer. Web Services Architecture Requirements. Technical report, W3C, 5 2000. http://www.w3.org/TR/soap/.
[Ben38] J. Bentham. An Introduction to the Principles of Morals and
Legislation. The Works of Jeremy Bentham, 43, 1838.
[BFHZ97] M.J. Bitner, W.T. Faranda, A.R. Hubbert, and V.A. Zeithaml.
Customer Contributions and Roles in Service Delivery. International Journal of Service Industry Management, 8(3):193–205, 1997.
[BG00] V. Bala and S. Goyal. A Noncooperative Model of Network Formation. Econometrica, pages 1181–1229, 2000.
[BK05] M. Bichler and J. Kalagnanam. Configurable Offers and Winner
Determination in Multi-Attribute Auctions. European Journal of
Operational Research, 160(2):380–394, 2005.

224

REFERENCES
[BKCvD09] B. Blau, J. Krämer, T. Conte, and C. van Dinther. Service Value
Networks. In Proceedings of the 11th IEEE Conference on Commerce
and Enterprise Computing (CEC 2009), 2009.
[BKO02] G. Bolton, E. Katok, and A. Ockenfels. How Effective are Online
Reputation Mechanisms. Discussion Papers on Strategic Interaction, 25:2002–25, 2002.
[BLFM98] T. Berners-Lee, R. Fielding, and L. Masinter. RFC2396: Uniform
Resource Identifiers (URI): Generic Syntax. RFC Editor United
States, 1998.
[BLH09] B. Blau, S. Lamparter, and S. Haak. remash! - Blueprints for
RESTful Situational Web Applications. In Proceedings of the 2nd
Workshop on Mashups, Enterprise Mashups and Lightweight Composition on the Web (MEM 2009), 2009.
[BLNW08] B. Blau, S. Lamparter, D. Neumann, and C. Weinhardt. Planning
and pricing of service mashups. In 10th IEEE Joint Conference on
E-Commerce Technology (CEC 2008) and Enterprise Computing, ECommerce and E-Services (EEE 2008), 21-24 July 2008, Washington,
D.C., USA, 2008.
[BNWM08] B. Blau, D. Neumann, C. Weinhardt, and W. Michalk. Provisioning of service mashup topologies. In Proceedings of the 16th
European Conference on Information Systems, ECIS 2008, 2008.
[Bon02] E. Bonabeau. Agent-Based Modeling: Methods And Techniques
for Simulating Human Systems. In National Academy of Sciences,
volume 99, pages 7280–7287. National Acad Sciences, 2002.
[Bor09] Nikolay Borissov. Q-Strategy: Automated Bidding and Convergence in Computational Markets. In 21st Innovative Applications of Artificial Intelligence (IAAI) Conference collocated with IJCAI, July 2009.
[BP91] L.L. Berry and A. Parasuraman. Marketing Services: Competing
Through Quality. Free Press, 1991.
[BPSM+ 06] Tim Bray, Jean Paoli, C. M. Sperberg-McQueen, Eve Maler, and
François Yergeau. Extensible Markup Language (XML). Technical report, W3C, 8 2006. http://www.w3.org/XML/.

REFERENCES

225

[BR04] R. Bianchini and R. Rajamony. Power and Energy Management
for Server Systems. Computer, 37(11):68–76, 2004.
[Bra97] F. Branco. The Design of Multidimensional Auctions. RAND
Journal of Economics, 28(1):63–81, 1997.
[BS99] P.D. Bridge and S.S. Sawilowsky. Increasing PhysiciansŠ Awareness of the Impact of Statistics on Research Outcomes Comparative Power of the T-Test and Wilcoxon Rank-Sum Test in
Small Samples Applied Research. Journal of Clinical Epidemiology, 52(3):229–235, 1999.
[BS00] K. Binmore and J. Swierzbinski. Treasury Auctions: Uniform or
Discriminatory? Review of Economic Design, 5(4):387–410, 2000.
[BS08] B. Blau and B. Schnizler. Description languages and mechanisms for trading service objects in grid markets. In Martin
Bichler, Thomas Hess, Helmut Krcmar, Ulrike Lechner, Florian Matthes, Arnold Picot, Benjamin Speitkamp, and Petra
Wolf, editors, Multikonferenz Wirtschaftsinformatik, MKWI 2008,
München, 26.2.2008 - 28.2.2008, Proceedings. GITO-Verlag 2008
Berlin, 2 2008.
[Bur04] M. Burner. Service Orientation and Its Role in Your Connected
Systems Strategy. Microsoft White Paper, July, 2004.
[BvDC+ 09] Benjamin Blau, Clemens van Dinther, Tobias Conte, Yongchun
Xu, and Christof Weinhardt. How to Coordinate Value Generation in Service Networks? – A Mechanism Design Approach.
(forthcoming), Journal of Business and Information Systems Engineering (Wirtschaftsinformatik), Special Issue Internet of Services,
2009.
[BvDCW09] Benjamin Blau, Clemens van Dinther, Tobias Conte, and
Christof Weinhardt. A Multidimensional Procurement Auction
for Trading Composite Services. Electronic Commerce Research
and Applications, Special Issue on Emerging Economic, Strategic and
Technical Issues in Online Auctions and Electronic Market Mechanisms (submitted), 2009.
[BVEL04] S. Brockmans, R. Volz, A. Eberhart, and P. Loffler. Visual Modeling of OWL DL Ontologies Using UML. Lecture Notes in Computer Science, pages 198–213, 2004.

226

REFERENCES
[CAT+ 01] Jeffrey S. Chase, Darrell C. Anderson, Prachi N. Thakar,
Amin M. Vahdat, and Ronald P. Doyle. Managing Energy and
Server Resources in Hosting Centers. SIGOPS Oper. Syst. Rev.,
35(5):103–116, 2001.
[CBSvD09] T. Conte, B. Blau, G. Satzger, and C. van Dinther. Enabling service networks through contribution-based value distribution. In
Proceedings of the 15th Americas Conference on Information Systems,
2009.
[CCMW01] Erik Christensen, Francisco Curbera, Greg Meredith,
and Sanjiva Weerawarana.
Web Service Description
Language (WSDL) 1.1.
Technical report, W3C, 3 2001.
http://www.w3.org/TR/wsdl/.
[CHvRR04] Luc Clement, Andrew Hately, Claus von Riegen, and
Tony Rogers.
Universal Description, Discovery, and Integration (UDDI).
Technical report, OASIS, 10 2004.
https://http://uddi.org/pubs/.
[CIoWM93] Y.K. Che, Social Systems Research Institute, and University
of Wisconsin-Madison. Design Competition Through Multidimensional Auctions. RAND Journal of Economics, 24:668–668,
1993.
[Cla71] E.H. Clarke. Multipart Pricing of Public Goods. Public Choice,
11(1):17–33, 1971.
[CNLP05] Martin Chapter, Eric Newcomer, Mark Little, and Greg
Pavlik.
Web Services Coordination Framework (WS-CF).
Technical report, OASIS, Public Review Draft, 10 2005.
http://www.oasis-open.org/committees/ws-caf/.
[Cro06] D. Crockford. JSON: The Fat-Free Alternative To XML. In Proceedings of XML, 2006.
[CSM+ 04] J. Cardoso, A. Sheth, J. Miller, J. Arnold, and K. Kochut. Quality of Service for Workflows and Web Service Processes. Web
Semantics: Science, Services and Agents on the World Wide Web,
1(3):281–308, 2004.

REFERENCES

227

[CvD09] T. Conte C. van Dinther, B. Blau. Strategic Behavior in Service
Networks under Price and Service Level Competition. In Proceedings of the 9th International Conference on Business Informatics,
2009.
[Dev98] J.F. Devlin. Adding Value to Service Offerings: The Case of
UK Retail Financial Services. European Journal of Marketing,
32(11):1091–1109, 1998.
[Dij59] EW Dijkstra. A Note on Two Problems in Connexion With
Graphs. Numerische Mathematik, 1(1):269–271, 1959.
[DJP03] RK Dash, NR Jennings, and DC Parkes. ComputationalMechanism Design: A Call to Arms. IEEE Intelligent Systems,
18(6):40–47, 2003.
[DLP03] A. Dan, H. Ludwig, and G. Pacifici. Web Service Differentiation
with Service Level Agreements. White Paper, IBM Corporation, 3
2003.
[DM93] W.H. Davidow and M.S. Malone. The Virtual Corporation:
Structuring and Revitalizing The Corporation for the 21st Century.
HarperBusiness, 1993.
[DSBF01] G. Da Silveira, D. Borenstein, and F.S. Fogliatto. Mass Customization: Literature Review and Research Directions. International Journal of Production Economics, 72(1):1–13, 2001.
[DVVfMSiES03] S. De Vries, R.V. Vohra, Center for Mathematical Studies in Economics, and Management Science. Combinatorial Auctions: A
Survey. INFORMS Journal on Computing, 15(3):284–309, 2003.
[EOS07] B. Edelman, M. Ostrovsky, and M. Schwarz. Internet Advertising and the Generalized Second-Price Auction: Selling Billions of Dollars Worth of Keywords. American Economic Review,
97(1):242–259, 2007.
[Eso01] M. Eso. An Iterative Online Auction for Airline Seats. IMA
Volumes In Mathematics And Its Applications, 127:45–58, 2001.
[ESS04] E. Elkind, A. Sahai, and K. Steiglitz. Frugality in Path Auctions. In Proceedings of the fifteenth annual ACM-SIAM symposium
on Discrete algorithms, pages 701–709. Society for Industrial and
Applied Mathematics Philadelphia, PA, USA, 2004.

228

REFERENCES
[Eva91] J.S. Evans. Strategic Flexibility for High Technology Manoeuvres: A Conceptual Framework. Journal of Management Studies,
28(1):69–89, 1991.
[EWL06] Yagil Engel, Michael P. Wellman, and Kevin M. Lochner. Bid Expressiveness and Clearing Algorithms in Multiattribute Double
Auctions. In Proceedings of the 7th ACM Conference on Electronic
Commerce, pages 110–119. ACM, 2006.
[FCSS05] Michael Feldman, John Chuang, Ion Stoica, and Scott Shenker.
Hidden-Action in Multi-Hop Routing. In Proceedings of the 6th
ACM Conference on Electronic commerce, pages 117–126. ACM,
2005.
[FGM+ 99] R. Fielding, J. Gettys, J. Mogul, H. Frystyk, L. Masinter, P. Leach,
and T. Berners-Lee. RFC2616: Hypertext Transfer Protocol–
HTTP/1.1. RFC Editor United States, 1999.
[Fie00] Roy Thomas Fielding. Architectural Styles and the Design of
Network-based Software Architectures. PhD thesis, University of
California, Irvine, 2000.
[FK07] J. Farrell and P. Klemperer. Coordination and Lock-In: Competition with Switching Costs and Network Effects. Handbook of
Industrial Organization, page 1967, 2007.
[FKNT02] I. Foster, C. Kesselman, J.M. Nick, and S. Tuecke. Grid Services
for Distributed System Integration. COMPUTER, pages 37–46,
2002.
[FL07] Joel Farrell and Holger Lausen. Semantic Annotations for
WSDL and XML Schema. Technical report, W3C, 8 2007.
http://www.w3.org/TR/sawsdl/.
[FPP09] Joan Feigenbaum, David C. Parkes, and David M. Pennock.
Computational Challenges in E-commerce. Communications of
the ACM, 52(1):70–74, 2009.
[FRS06] Joan Feigenbaum, Vijay Ramachandran, and Michael Schapira.
Incentive-Compatible Interdomain Routing. In Proceedings of the
7th ACM Conference on Electronic Commerce, pages 130–139, 2006.
[Fuc68] V.R. Fuchs. The Service Economy. Natl Bureau of Economic Res,
1968.

REFERENCES

229

[Gad92] J. Gadrey. L’économie des Services. 1992.
[Gad00] J. Gadrey. The Characterization of Goods and Services: An Alternative Approach. Review of Income and Wealth, 46(3):369–387,
2000.
[Gal73] J.R. Galbraith. Designing Complex Organizations. AddisonWesley Longman Publishing Co., Inc. Boston, MA, USA, 1973.
[Gib73] Allan Gibbard. Manipulation of Voting Schemes: A General
Result. Econometrica, 41(4):587–601, July 1973.
[Gib92] R. Gibbons. Game Theory for Applied Economists. Princeton University Press Princeton, 1992.
[GL78] Jerry R. Green and Jean-Jacques Laffont. Incentives in Public Decision – Making, Studies in Public Economics. North–Holland Publishing Company, Boston, 1978.
[GNC+ 04] Steve Graham, Peter Niblett, Dave Chappell, Amy Lewis,
Nataraj Nagaratnam, Jay Parikh, Sanjay Patil, Shivajee
Samdarshi, Igor Sedukhin, David Snelling, Steve Tuecke,
William Vambenepe, and Bill Weihl. Web Services Notification (WS-Notification). Technical report, OASIS, 5 2004.
http://www.oasis-open.org/committees/wsn/.
[GR71] P.E. Green and V.R. Rao. Conjoint Measurement for Quantifying
Judgmental Data. Journal of Marketing Research, pages 355–363,
1971.
[Gri92] Z. Griliches. Output Measurement in the Service Sectors, Studies in Income and Wealth. 56, 1992.
[Gro73] Theodore Groves. Incentives in Teams. Econometrica, 41(4):617–
631, 1973.
[GS06] J. Gebauer and F. Schober. Information System Flexibility and
the Cost Efficiency of Business Processes. Journal of the Association for Information Systems, 7(3):122–147, 2006.
[GSB+ 02] S. Graham, S. Simeonov, T. Boubez, D. Davis, G. Daniels,
Y. Nakamura, and R. Neyama. Building Web services with Java.
Sams, 2002.

230

REFERENCES
[GW97] F. Gallouj and O. Weinstein. Innovation in Services. Research
Policy, 26(4-5):537–556, 1997.
[Had06] Marc J. Hadley.
Web Application Description Language
(WADL). Technical report, Sun Microsystems Inc., 11 2006.
https://wadl.dev.java.net/.
[Hil77] T.P. Hill. On Goods and Services. Review of Income and Wealth,
23(4):315–338, 1977.
[Hil99] T.P. Hill. Tangibles, Intangibles and Services: A New Taxonomy
for the Classification of Output. Canadian Journal of Economics,
32:426–446, 1999.
[HN96] D. Harel and A. Naamad. The STATEMATE Semantics of Statecharts. ACM Transactions on Software Engineering and Methodology, 5(4):293–333, 1996.
[HPSB+ 04] Ian Horrocks, Peter F. Patel-Schneider, Harold Boley, Said
Tabet, Benjamin Grosof, and Mike Dean.
Semantic Web
Rule Language (SWRL).
Technical report, W3C, 5 2004.
http://www.w3.org/Submission/SWRL/.
[HS01] J. Hershberger and S. Suri. Vickrey Prices and Shortest Paths:
What Is an Edge Worth? In Foundations of Computer Science,
2001. Proceedings. 42nd IEEE Symposium on, pages 252–259, 2001.
[Hur72] L. Hurwicz. On Informationally Decentralized Systems/Decision And Organization. Radner, R., CB McGuire. In Honor of J.
Marschak, 1972.
[Hur73] L. Hurwicz. The Design of Mechanisms for Resource Allocation.
American Economic Review, 63(2):1–30, 1973.
[HW90] L. Hurwicz and M. Walker. On the Generic Nonoptimality of
Dominant-Strategy Allocation Mechanisms: A General Theorem that Includes Pure Exchange Economies. Econometrica: Journal of the Econometric Society, pages 683–704, 1990.
[IL04] M. Iansiti and R. Levien. Strategy as Ecology. Harvard Business
Review, 82(3):68–81, 2004.
[Jac92] M.O. Jackson. Incentive Compatibility and Competitive Allocations. Economics Letters, 40:299–302, 1992.

REFERENCES

231

[Jac03] M.O. Jackson. Efficiency and Information Aggregation in Auctions With Costly Information. Review of Economic Design,
8(2):121, 2003.
[JF03] R. Jurca and B. Faltings. An Incentive Compatible Reputation
Mechanism. In Proceedings of the IEEE International Conference on
E-Commerce, pages 285–292, 2003.
[Jhi06] A. Jhingran. Enterprise Information Mashups: Integrating Information, Simply. In Proceedings of the 32nd International Conference on Very Large Data Bases, pages 3–4. VLDB Endowment,
2006.
[JIB07] A. Jøsang, R. Ismail, and C. Boyd. A Survey of Trust and Reputation Systems for Online Service Provision. Decision Support
Systems, 43(2):618–644, 2007.
[JMS02] L. Jin, V. Machiraju, and A. Sahai. Analysis on Service Level
Agreement of Web Services. HP, 6 2002.
[JW96] M.O. Jackson and A. Wolinsky. A Strategic Model of Social
and Economic Networks. Journal of economic Theory, 71(1):44–74,
1996.
[JW02] M.O. Jackson and A. Watts. The Evolution of Social and Economic Networks. Journal of Economic Theory, 106(2):265–295,
2002.
[KCS08] A. Kittur, E.H. Chi, and B. Suh. Crowdsourcing User Studies
with Mechanical Turk. 2008.
[KK05] AR Karlin and D. Kempe. Beyond VCG: Frugality of Truthful Mechanisms. In Foundations of Computer Science, 2005. FOCS
2005. 46th Annual IEEE Symposium on, pages 615–624, 2005.
[KN04] D. Karger and E. Nikolova. VCG Overpayment in Random
Graphs. In DIMACS Workshop on Computational Issues in Auction
Design, 2004.
[KN05] D. Karger and E. Nikolova. Brief Announcement: On the Expected Overpayment of VCG Mechanisms in Large Networks.
In Proceedings of the twenty-fourth annual ACM symposium on
Principles of distributed computing, pages 126–126. ACM New
York, NY, USA, 2005.

232

REFERENCES
[Kra05] B. Kratz. Protocols For Long Running Business Transactions.
Technical Report 17, Infolab Technical Report Series, 2005.
[KS85] M.L. Katz and C. Shapiro. Network Externalities, Competition,
and Compatibility. The American Economic Review, pages 424–
440, 1985.
[KV98] S. Kochugovindan and N.J. Vriend. Is the Study of Complex
Adaptive Systems Going to Solve the Mystery of Adam Smith’s
Invisible Hand? Independent Review, 3:53–66, 1998.
[Lai05] K. Lai. Markets are Dead, Long Live Markets. ACM SIGecom
Exchanges, 5(4):1–10, 2005.
[Lam07] Steffen Lamparter. Policy-Based Contracting in Semantic Web Service Markets. PhD thesis, Universität Karlsruhe (TH), 2007.
[Lev81] T. Levitt. Marketing Intangible Products and Product Intangibles. Cornell Hotel and Restaurant Administration Quarterly,
22(2):37, 1981.
[Ley03] F. Leymann. Web Services: Distributed Applications without
Limits. Business, Technology and Web, 2003.
[LGS07] Jon Lathem, Karthik Gomadam, and Amit P. Sheth. SA-REST
and (S)mashups: Adding Semantics to RESTful Services. In
ICSC ’07: Proceedings of the International Conference on Semantic
Computing, pages 469–476, Washington, DC, USA, 2007. IEEE
Computer Society.
[LM94] SJ Liebowitz and S.E. Margolis. Network Externality: An Uncommon Tragedy. The Journal of Economic Perspectives, pages
133–150, 1994.
[LNZ04] Yutu Liu, Anne H. Ngu, and Liang Z. Zeng. QoS Computation
and Policing in Dynamic Web Service Selection. In Proceedings of
the 13th international World Wide Web conference on Alternate Track
Papers & Posters, pages 66–73, New York, NY, USA, 2004. ACM.
[LR00] D. Lucking-Reiley. Auctions on the Internet: What’s Being Auctioned, and How? Journal of Industrial Economics, 48(3):227–252,
2000.

REFERENCES

233

[LS06] S. Lamparter and B. Schnizler. Trading Services in OntologyDriven Markets. In Proceedings of the 2006 ACM symposium on
Applied computing, pages 1679–1683. ACM New York, NY, USA,
2006.
[LSW01] Z. Liu, M.S. Squillante, and J.L. Wolf. On Maximizing ServiceLevel-Agreement Profits. In Proceedings of the 3rd ACM conference on Electronic Commerce, pages 213–223. ACM New York, NY,
USA, 2001.
[LT64] R.D. Luce and J.W. Tukey. Simultaneous Conjoint Measurement:
A New Type of Fundamental Measurement. Journal of Mathematical Psychology, 1(1):1–27, 1964.
[LVO07] R.F. Lusch, S.L. Vargo, and M. OŠBrien. Competing Through
Service: Insights From Service-Dominant Logic. Journal of Retailing, 83(1):5–18, 2007.
[LW01] C.H. Lovelock and J. Wirtz. Services Marketing: People, Technology, Strategy. Prentice Hall, 2001.
[LW03] M. Little and J. Webber. Introducing WS-CAF – More Than Just
Transactions. Web Services Journal, 3(12):52–55, 2003.
[Mal85] T.W. Malone. Organizational Structure and Information Technology: Elements of a Formal Theory. 1985.
[Mal87] Thomas W. Malone. Modeling Coordination in Organizations
and Markets. Management Science, 33(10):1317–1332, 1987.
[MB09] T. Meinl and B. Blau. Web Service Derivatives. In Proceedings
of the 18th International World Wide Web Conference (WWW2009),
Madrid, Spain, 4 2009.
[MC94] Thomas W. Malone and Kevin Crowston. The Interdisciplinary
Study of Coordination. ACM Comput. Surv., 26(1):87–119, 1994.
[MCWG95] A. Mas-Colell, M.D. Whinston, and J.R. Green. Microeconomic
Theory. Oxford University Press New York, 1995.
[Men02] DA Menasce. QoS Issues in Web services. IEEE Internet Computing, 6(6):72–75, 2002.

234

REFERENCES
[Mer06] D. Merrill. Mashups: The New Breed of Web App – An
Introduction to Mashups. Technical report, IBM, 8 2006.
http://www.ibm.com/developerworks/xml/library/x-mashups.html.

[MLM+ 06] C. Matthew MacKenzie, Ken Laskey, Francis McCabe, Peter F
Brown, and Rebekah Metz. Reference Model for Service Oriented Architecture 1.0. Technical report, OASIS, 10 2006.
[MMV94] J.K. MacKie-Mason and H.R. Varian. Generalized Vickrey Auctions. Technology report. University of Michigan, July, 1994.
[MMW06] J.K. MacKie-Mason and M.P. Wellman. Automated Markets and
Trading Agents. Ann Arbor, 1001:48109–1092, 2006.
[MN02] A. Mani and A. Nagarajan. Understanding quality of service for
Web services. IBM developerWorks, 1 2002.
[MN08a] A. Mu’Alem and N. Nisan. Truthful Approximation Mechanisms for Restricted Combinatorial Auctions. Games and Economic Behavior, 64(2):612–631, 2008.
[MN08b] Ahuva Mu’alem and Noam Nisan. Truthful Approximation
Mechanisms for Restricted Combinatorial Auctions. Games and
Economic Behavior, 2008.
[MNM+ 07] M. Mohabey, Y. Narahari, S. Mallick, P. Suresh, and SV Subrahmanya. A Combinatorial Procurement Auction for QoS-Aware
Web Services Composition. In IEEE International Conference on
Automation Science and Engineering, 2007. CASE 2007, pages 716–
721, 2007.
[MPW08] R. Müller, A. Perea, and S. Wolf. Combinatorial Scoring Auctions. Technical report, 2008.
[MS83] R. Myerson and M. Satterthwaite. Efficient Mechanisms for Bilateral Exchange. Journal of Economic Theory, 28:265–281, 1983.
[MS84] T.W. Malone and S.A. Smith. Tradeoffs in Designing Organizations: Implications for New Forms of Human Organizations
and Computer Systems. 1984.
[MS86] R.E. Miles and C.C. Snow. Organizations: New Concepts for
New Forms. California Management Review, 28(3):62–74, 1986.

REFERENCES

235

[MSS+ 08] Justin Moore, Ratnesh Sharma, Rocky Shih, Jeff Chase, Chandrakant Patel, and Parthasarathy Ranganathan. Going beyond
CPUs: The Potential of Temperature-Aware Solutions for the
Data Center. Whitepaper, Hewlett Packard Labs, January 2008.
[MSZ01] S.A. McIlraith, T.C. Son, and H. Zeng. Semantic Web Services.
IEEE Intelligent Systems, pages 46–53, 2001.
[MT07] P. Maille and B. Tuffin. Why VVG Auctions Can Hardly be Applied to the Pricing of Inter-Domain and Ad Hoc Networks.
In 3rd EuroNGI Conference on Next Generation Internet Networks,
pages 36–39, 2007.
[Mul06] A. Mulholland. The End of Business as Usual: Service-Oriented
Business Transformation. Lecture Notes in Computer Science,
4294:540, 2006.
[MV98] P. Matthyssens and K. Vandenbempt. Creating Competitive Advantage in Industrial Services. Journal Of Business and Industrial
Marketing, 13:339–355, 1998.
[MvH04] Deborah L. McGuinness and Frank van Harmelen. Web Ontology Language (OWL). Technical report, W3C, 2 2004.
http://www.w3.org/2004/OWL/.
[MWL+ 06] T.W. Malone, P. Weill, R.K. Lai, V.T. D’Urso, G. Herman, T.G.
Apel, S. Woerner, and I. Author. Do Some Business Models Perform Better than Others? Technical report, 2006.
[MYB87] Thomas W. Malone, Joanne Yates, and Robert I. Benjamin. Electronic Markets and Electronic Hierarchies. Communications of the
ACM, 30(6):484–497, 1987.
[Mye81] R.B. Myerson. Optimal Auction Design. Mathematics of operations research, pages 58–73, 1981.
[Mye82] Roger B. Myerson. Optimal Coordination Mechanisms in Generalized Principal-Agent Problems. Journal of Mathematical Economics, 10(1):67–81, June 1982.
[Mye88] R.B. Myerson. Mechanism Design. 1988.

236

REFERENCES
[Neu04] Dirk Georg Neumann. Market Engineering – A Structured Design
Process for Electronic Markets. PhD thesis, Universität Karlsruhe
(TH), 2004.
[NKMHB06] Anthony Nadalin, Chris Kaler, Ronald Monzillo, and Phillip
Hallam-Baker. Web Services Security: SOAP Message Security 1.1 (WS-Security). Technical report, OASIS, 2 2006.
http://docs.oasis-open.org/wss/v1.1/.
[NR01] N. Nisan and A. Ronen. Algorithmic Mechanism Design. Games
and Economic Behavior, 35(1-2):166–196, 2001.
[NR07] N. Nisan and A. Ronen. Computationally Feasible VCG Mechanisms. Journal of Artificial Intelligence Research, 29:19–47, 2007.
[NRFJ07] Eric Newcomer,
Ram Jeyaraman.
Coordination).

Ian

Robinson, Max Feingold, and
Web Services Coordination (WSTechnical report, OASIS, 7 2007.
http://docs.oasis-open.org/ws-tx/wscoor/.

[NRFL07] Eric Newcomer, Ian Robinson, Tom Freund, and
Mark Little.
Web Services Business Activity (WSBusinessActivity).
Technical report, OASIS, 7 2007.
http://docs.oasis-open.org/ws-tx/wsba/.
[NRLW07] Eric Newcomer, Ian Robinson, Mark Little, and Andrew Wilkinson.
Web Services Atomic Transaction (WSAtomicTransaction).
Technical report, OASIS, 7 2007.
http://docs.oasis-open.org/ws-tx/wsat/.
[NRTV07] Noam Nisan, Tim Roughgarden, Eva Tardos, and Vijay V. Vazirani. Algorithmic Game Theory. Cambridge University Press,
2007.
[NS06] N. Nisan and A. Sen. Weak Monotonicity Characterizes Deterministic Dominant-Strategy Implementation. Econometrica,
pages 1109–1132, 2006.
[OEC05] OECD. Science, Technology and Industry Scoreboard 2005 – Towards a Knowledge-Based Economy. Technical report, OECD,
2005.

REFERENCES

237

[OMG07] OMG. The Unified Modeling Language (UML) 2.1.2. Technical report, Object Management Group (OMG), 4 2007.
http://www.omg.org/spec/UML/2.1.2/.
[Pap01] C. Papadimitriou. Algorithms, games, and the internet. In Proceedings of the thirty-third annual ACM symposium on Theory of
computing, pages 749–753. ACM New York, NY, USA, 2001.
[Pap08] P. Papazoglou. Web Services: Principles and Technologies. Prentice
Hall, 2008.
[Par01] D.C. Parkes. Iterative Combinatorial Auctions: Achieving Economic
and Computational Efficiency. PhD thesis, University of Pennsylvania, 2001.
[Pau08] C. Pautasso. BPEL for REST. In Proceedings of the 6th International
Conference on Business Process Management (BPM 2008), Milan,
Italy. Springer, September 2008.
[PBB+ 04] M. Pistore, F. Barbon, P. Bertoli, D. Shaparau, and P. Traverso.
Planning and Monitoring Web service Composition. Lecture
Notes in Computer Science, pages 106–115, 2004.
[PD04] M.P. Papazoglou and J. Dubray. A Survey of Web Service Technologies. Technical report, University of Tronto, Department of
Information and Communication Technology, 6 2004.
[PG03] M.P. Papazoglou and D. Georgakopoulos. Service-Oriented
Computing. Communications of the ACM, 46(10):25–28, 2003.
[Phe08] S.G. Phelps. Evolutionary Mechanism Design. PhD thesis, University of Liverpool, 2008.
[PK02] D. Parkes and J. Kalagnanam. Iterative Multiattribute Vickrey
Auctions. Technical report, Harvard University, 2002.
[PK05] D.C. Parkes and J. Kalagnanam. Models for Iterative Multiattribute Procurement Auctions. Management Science, 51(3):435–
451, 2005.
[PKE01] D.C. Parkes, J. Kalagnanam, and M. Eso. Achieving BudgetBalance with Vickrey-Based Payment Schemes in Combinatorial
Exchanges. Technical report, IBM Research, 2001.

238

REFERENCES
[PMS04] F.T. Piller, K. Moeslein, and C.M. Stotko. Does Mass Customization Pay? An Economic Approach to Evaluate Customer Integration. Production Planning & Control, 15(4):435–444, 2004.
[PS98] C.H. Papadimitriou and K. Steiglitz. Combinatorial Optimization:
Algorithms and Complexity. Dover Publications, 1998.
[PS00] W. Pesendorfer and J.M. Swinkels. Efficiency and Information
Aggregation in Auctions. American Economic Review, 90(3):499–
525, 2000.
[PZL08] C. Pautasso, O. Zimmermann, and F. Leymann. RESTful Web
Services vs. Big Web Services: Making the Right Architectural
Decision. ACM New York, NY, USA, 2008.
[Ram80] P.H. Ramsey. Choosing the Most Powerful Pairwise Multiple
Comparison Procedure in Multivariate Analysis of Variance.
Journal of Applied Psychology, 65(3,317-326), 1980.
[Rap04] M.A. Rappa. The Utility Business Model and the Future of Computing Services. IBM Systems Journal, 43(1):32–42, 2004.
[Rat66] J.M. Rathmell. What is meant by services? Journal of Marketing,
30(4):32–36, 1966.
[Rei77] Stanley Reiter. Information and Performance in the (New) Welfare Economics. The American Economic Review, 67(1):226–234,
1977.
[RH07] Stuart Rance and Ashley Hanna. Glossary of Terms and Definitions. Technical report, ITIL IT Service Management, 2007.
[RK02] R.T. Rust and PK Kannan. E-Service: New Directions in Theory
and Practice. ME Sharpe, 2002.
[RK03] R.T. Rust and PK Kannan. E-service: A New Paradigm for Business in the Electronic Environment. Communications of the ACM,
46(6):36–42, 2003.
[RL05] A. Ronen and D. Lehmann. Nearly Optimal Multi-Attribute
Auctions. In Proceedings of the 6th ACM conference on Electronic
commerce, pages 279–285. ACM Press New York, NY, USA, 2005.

REFERENCES

239

[Ron01] Amir Ronen. On Approximating Optimal Auctions. In Proceedings of the 3rd ACM Conference on Electronic Commerce, pages 11–
17. ACM, 2001.
[Rot02] A.E. Roth. The Economist as Engineer: Game Theory, Experimentation, and Computation as Tools for Design Economics.
Econometrica, pages 1341–1378, 2002.
[RP76] D.J. Roberts and A. Postlewaite. The Incentives for Price-Taking
Behavior in Large Exchange Economies. Econometrica: Journal of
the Econometric Society, pages 115–127, 1976.
[RPH98] M.H. Rothkopf, A. Pekeč, and R.M. Harstad. Computationally Manageable Combinational Auctions. Management Science,
pages 1131–1147, 1998.
[RR07] L. Richardson and S. Ruby. RESTful Web Services. O’Reilly, 2007.
[Saa80] T.L. Saaty. The Analytical Hierarchy Process. McGraw-Hill, New
York, 1980.
[Saa08] T.L. Saaty. Decision Making with the Analytic Hierarchy Process. International Journal of Services Sciences, 1(1):83–98, 2008.
[SB92] SS Sawilowsky and RC Blair. A More Realistic Look at the Robustness and Type II Error Properties of the T Test to Departures
from Population Normality. Psychological Bulletin, 111(2):352–
360, 1992.
[SB99] RS Sutton and AG Barto. Reinforcement Learning. Journal of
Cognitive Neuroscience, 11(1):126–134, 1999.
[SB04] M. Salle and C. Bartolini. Management by Contract. Network Operations and Management Symposium, 2004. NOMS 2004. IEEE/IFIP, 1, 2004.
[SBF98] R. Studer, V.R. Benjamins, and D. Fensel. Knowledge Engineering: Principles and Methods. Data & Knowledge Engineering,
25(1-2):161–197, 1998.
[Sch07] B. Schnizler. Resource allocation in the Grid. A Market Engineering
Approach. PhD thesis, Universität Karlsruhe (TH), 2007.

240

REFERENCES
[SGL07] Amit P. Sheth, Karthik Gomadam, and Jon Lathem. SAREST: Semantically Interoperable and Easier-to-Use Services
and Mashups. IEEE Internet Computing, 11(6):91–94, 2007.
[Sho85] G.L. Shostack. Planning the Service Encounter. The Service Encounter, Lexington Books, Lexington, MA, pages 243–54, 1985.
[Smi82] V.L. Smith. Microeconomic Systems as an Experimental Science.
The American Economic Review, pages 923–955, 1982.
[Smi89] C.W. Smith. Auctions: The Social Construction of Value. University
of California Press, 1989.
[SMS+ 02] A. Sahai, V. Machiraju, M. Sayal, A. Van Moorsel, F. Casati, and
L.J. Jin. Automated SLA Monitoring for Web services. Lecture
Notes in Computer Science, pages 28–41, 2002.
[SNP+ 05] J. Shneidman, C. Ng, D.C. Parkes, A. AuYoung, A.C. Snoeren,
A. Vahdat, and B. Chun. Why Markets Could (But DonŠt Currently) Solve Resource Allocation Problems in Systems. In Proceedings of the 10th Conference on Hot Topics in Operating Systems,
pages 7–7, 2005.
[SSGL05] T. Sandholm, S. Suri, A. Gilpin, and D. Levine. CABOB: A Fast
Optimal Algorithm for Winner Determination in Combinatorial
Auctions. Management Science, 51(3):374–390, 2005.
[Sta79] T.M. Stanback. Understanding the Service Economy: Employment,
Productivity, Location. Johns Hopkins Univserity Press, 1979.
[Ste04] F. Steiner. Formation and Early Growth of Business Webs: Modular
Product Systems in Network Markets. Physica-Verlag Heidelberg,
2004.
[Sto09] Jochen Stoesser. Market-Based Scheduling in Distributed Computing Systems. PhD thesis, Universität Karlsruhe (TH), 2009.
[SV99] C. Shapiro and H.R. Varian. Information Rules. Harvard Business
School Press Boston, Mass, 1999.
[Tal03] K. Talwar. The Price of Truth: Frugality in Truthful Mechanisms.
Lecture Notes in Computer Science, pages 608–619, 2003.

REFERENCES

241

[Tes01] L. Tesfatsion. Introduction to The Special Issue on Agent-Based
Computational Economics. Journal of Economic Dynamics and
Control, 25(3-4):281–293, 2001.
[Tho91] G. Thompson. Markets, Hierarchies and Networks: The Coordination of Social Life. Sage, 1991.
[TLT00] D. Tapscott, A. Lowy, and D. Ticoll. Digital Capital: Harnessing
the Power of Business Webs. Harvard Business School Press, 2000.
[TW06] D. Tapscott and A.D. Williams. Wikinomics: How Mass Collaboration Changes Everything. Portfolio, 2006.
[Var09] H.R. Varian. Online Ad Auctions. American Economic Review,
2009.
[vHV07] E. van Heck and P. Vervest. Smart Business Networks: How the
Network Wins. Communications of the ACM, 50(6):29–37, 2007.
[Vic61] William Vickrey. Counterspeculation, Auctions, and Competitive Sealed Tenders. The Journal of Finance, 16(1):8–37, 1961.
[VL04] S.L. Vargo and R.F. Lusch. Evolving to a New Dominant Logic
for Marketing. Journal of Marketing, 68(1):1–17, 2004.
[VvHPP05] P. Vervest, E. van Heck, K. Preiss, and L.F. Pau. Smart Business
Networks. Springer, 2005.
[Wal80] M. Walker. On the Nonexistence of a Dominant Strategy Mechanism for Making Optimal Public Decisions. Econometrica: Journal of the Econometric Society, pages 1521–1540, 1980.
[WCL+ 05] S. Weerawarana, F. Curbera, F. Leymann, T. Storey, and D.F.
Ferguson. Web Services Platform Architecture: SOAP, WSDL,
WS-Policy, WS-Addressing, WS-BPEL, WS-Reliable Messaging and
More. Prentice Hall PTR Upper Saddle River, 2005.
[WD92] C.J.C.H. Watkins and P. Dayan. Q-Learning. Machine learning,
8(3):279–292, 1992.
[WHN03] C. Weinhardt, C. Holtmann, and D. Neumann. Market Engineering. Wirtschaftsinformatik, 45(6):635–640, 2003.

242

REFERENCES
[Wil79] O.E. Williamson. Transaction-Cost Economics: The Governance
of Contractual Relations. The journal of Law and Economics,
22(2):233, 1979.
[Win99] Dave Winer.
Extensible Markup Language Remote
Procedure Call (XML-RPC).
Technical report, 7 1999.
http://www.xmlrpc.com/spec/.
[Win02] A. Winter. Exchanging Graphs with GXL. Lecture Notes in Computer Science, pages 485–500, 2002.
[WNH06] C. Weinhardt, D. Neumann, and C. Holtmann. ComputerAided Market Engineering. Communications of the ACM, 2006.
[WV03] Y. Wang and J. Vassileva. Trust and Reputation Model in Peerto-Peer Networks. In Proceedings of the 3rd International Conference on Peer-to-Peer Computing, pages 150–157, 2003.
[ZBD+ 03] Liangzhao Zeng, Boualem Benatallah, Marlon Dumas, Jayant
Kalagnanam, and Quan Z. Sheng. Quality Driven Web Services
Composition. In Proceedings of the 12th international conference
on World Wide Web, pages 411–421, New York, NY, USA, 2003.
ACM.
[ZVB96] A. Zeithaml Valarie and M.J. Bitner. Services Marketing. 1996.

The fundamental paradigm shift from traditional value chains to agile service value networks (SVN) implies new economic and organizational challenges. In service
value networks, a multitude of participants co-create complex services that create
added value for customers by providing highly specialized service components and
by leveraging lightweight paradigms such as RESTful architectures and mashup technologies. Addressing the challenge of coordinating distributed activities in order to
achieve a desired outcome, auctions have proven to perform quite well in situations
where intangible and heterogeneous economic entities are traded.
Nevertheless, traditional approaches in the area of multidimensional combinatorial
auctions are not quite suitable to enable the trade of composite services. A flawless
service execution and therefore the requester’s valuation highly depends on the accurate sequence of the functional parts of the composition, meaning that in contrary to
service bundles, composite services only generate value through a valid order of their
components. From a technical perspective, service composition research traditionally
assumes complete information about QoS characteristics and prices and does not
account for self-interested service owners that intent to maximize their utility and
therefore behave strategically.

ISBN 978-3-86644-724-0

ISSN 1862-8893
ISBN 978-3-86644-724-0

9 783866 447240