Corporate Decision-Making For The Adoption of Cloud Computing: Economic and Organizational Factors

Corporate Decision-Making for the Adoption of Cloud
Computing : Economic and Organizational Factors

Jacques Bou Abdo
To cite this version:

Jacques Bou Abdo. Corporate Decision-Making for the Adoption of Cloud Computing : Economic
and Organizational Factors. Business administration. Université Paris-Saclay, 2021. English. �NNT :
2021UPASI017�. �tel-03631249�
HAL Id: tel-03631249

https://theses.hal.science/tel-03631249
Submitted on 5 Apr 2022
HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Prise de décision commerciale pour
l’adoption du cloud computing:
facteurs économiques et
organisationnels
Corporate Decision-Making for the Adoption
of Cloud Computing: Economic and
Organizational Factors
Thèse de doctorat de l’Université Paris-Saclay
École doctorale n◦ 630, Droit Economie Management (DEM)

Spécialité de doctorat: Sciences de gestion et du management
Unité de recherche: Université Paris-Saclay, Réseaux Innovation
Territoires et Mondialisation, 92330, Sceaux, France
Référent : Faculté de droit, économie et gestion
Thèse présentée et soutenue à Sceaux, le 07/12/2021, par
Jacques BOU ABDO

Thèse de doctorat
Composition du jury:
Grazia Cecere Présidente

NNT: 2021UPASI017
Professor, Institut Mines-Télécom Business School

Thomas J Housel Rapporteur
Professor, Naval postgraduate School
Euripidis N. Loukis Rapporteur
Professor, University of the Aegean
Emmanuel Waller Examinateur
Maître de conférences, HDR, Université Paris-Saclay
Ahmed Bounfour Directeur de thèse
Professor, Université Paris-Saclay
Remerciements
At the end of this journey, and at the start of a new epoch in my life, I am now
sure more than any time before that the path to success cannot be walked alone.
I consider myself to be blessed and extremely lucky to have been surrounded by
compassionate and supportive mentors, colleagues and family who walked with me
and sometimes carried me through this journey. I count on you for the ways to come.
I would like to thank Prof. Ahmed Bounfour for his support, guidance,
and supervision over the past six years. I learnt a lot from the breadth of knowledge
he offered as a world-class scholar. Perhaps more importantly, I was humbled by his
faith in my ability which allowed me to mature as a scholar under his guidance.
According to academic genealogy, mentors or thesis advisors are considered parents.
I am extremely proud to have my academic lineage traced back to Prof. Bounfour.
I wish also to thank Prof. Thomas Housel and Prof. Euripidis Loukis
for their extremely helpful comments and advice. I am very proud of having you
as assessors of my work. I am very thankful for Prof. Grazia Cecere and Prof.
Emmanuel Waller for agreeing to assess my work and providing me with their
invaluable advice.
I can’t but thank all my colleagues at “Université Paris-Saclay, Réseaux

Innovation Territoires et Mondialisation” who have been very supportive. We are
siblings of the same institution which I am very proud to belong to. I wish to namely
thank Mrs. Maryse Chomette, Mrs. Marielle Rosine and Mrs. Noémie Moutty and I
apologize for forgetting anyone of the many who showed dedication and commitment
in helping me during the past six years.
My family has unequivocal impact on this journey by growing the love of
knowledge and science in me. They cultivated me form the first minute of my life
and still support me with love and prayers. To my dad (Elias), Mom (Karmen) and
Brother (Charbel), you made me the man I am. To my grandfather (Sleiman), I hope
your desire has been achieved.
Love of science and knowledge is not comparable to the love of my angel

which is the reason for my dedication and hard work. Thank you, Esther, for always
being there for me. You are a fruitful vine by the sides of my house. To my children,
Angela, Charbel and Jacintha, you are the reason for it all. You are vigorous young
olive trees around my table. I hope I made you proud.
Above all, I am thankful for God’s grace.
Ad maiorem Dei gloriam

Titre: Prise de décision commerciale pour l’adoption du cloud computing: facteurs
économiques et organisationnels
Mots clés: Modèles économiques, Cloud Computing, Filtrage collaboratif, Solutions d’aide à
la décision, Technologies de rupture, Réseaux mobiles, Économie de réseau, Systèmes de recom-
mandation, Adoption de la technologie, 5G et au-delà
Résumé: L’admission des technologies dis- des organisations, et nous étudions soigneuse-
ruptives est une décision assez cruciale et ment les besoins et la préparation des organi-
délicate qui affecte la survie et la continu- sations pour une technologie de rupture. Ce tra-
ité des entreprises, startups et multinationales. vail considère le cloud computing comme la tech-
Cette décision est considérée par de nombreux nologie disruptive étudiée et concentre sa con-
chercheurs et décideurs comme étant un juge- tribution sur l’admission du cloud computing
ment que chaque organisation devrait juger et par les organisations. Au niveau organisation-
décider. La question qui se pose est la suiv- nel, une solution d’aide à la décision est conçue
ante : les organisations, sont-elles vraiment li- à l’aide des systèmes de recommandation de fil-
bres de décider d’adopter ou non une technolo- trage collaboratif. Cette solution vise à aider les
gie spécifique ? En fait, de nombreuses or- décideurs à évaluer leurs besoins organisation-
ganisations n’ont pas la concession d’imposer nels et à choisir de façon optimale l’ensemble
leurs besoins aux fournisseurs, leurs services des technologies adaptées aux besoins de leurs
aux clients, ou leurs modèles commerciaux aux organisations. Au niveau économique, les fac-
concurrents. Dans ce travail, nous étudions, teurs économiques sont modélisés à l’aide de
à deux niveaux, la décision des organisations la modélisation à base d’agents. Nous consid-
en ce qui concerne l’adoption des technologies érons l’adoption du cloud computing dans les
de rupture. En premier lieu, nous considérons réseaux mobiles 5G comme un cas spécifique
les forces économiques (forces externes) qui in- d’admission du cloud. Cette spécification est
fluencent la décision des organisations. Ainsi, nécessaire pour permettre une modélisation pré-
nous examinons minutieusement la réponse du cise des facteurs économiques. Deux scénarios
marché aux technologies de rupture. En deux- sont envisagés, le premier suppose une concur-
ième lieu, nous considérons les forces organisa- rence parfaite sur le marché de la 5G, alors que
tionnelles (forces internes) affectant la décision le second assume un marché oligopolistique.
Title: Corporate Decision-Making for the Adoption of Cloud Computing: Economic and
Organizational Factors
Keywords: Business models, cloud computing, collaborative filtering, decision support solu-
tions, disruptive technologies, mobile networks, network economics, recommender systems, tech-
nology adoption, 5G & Beyond
Abstract: The adoption of disruptive tech- nization’s decision. More thoroughly, we look at
nologies is a very critical decision affecting the the organization’s needs and readiness for a dis-
survival and continuity of businesses, startups ruptive technology. This work considers cloud
to multinationals. Many researchers and deci- computing as the studied disruptive technology
sion makers consider technology adoption as a and focuses its contribution on cloud computing
decision each organization should evaluate and adoption. In the organizational level, a Decision
decide upon. But are organizations really free to Support Solution is designed using collaborative
decide whether to adopt a specific technology or filtering recommender systems to help decision
not? Many businesses do not have the privilege makers evaluate their organizational needs and
to dictate its needs to vendors, its services to decide on the optimal technology mix suitable
customers or its business models to competitors. for their organization. In the economic level,
In this work we study the organization’s deci- the economic factors are modeled using Agent-
sion on disruptive technology adoption at two Based Modeling. We consider cloud computing
levels. In the first level, we consider the eco- adoption in the 5G mobile networks as a spe-
nomic forces (external forces) affecting the orga- cific case of cloud adoption. This specification
nization’s decision. More thoroughly, we look at is needed to allow accurate modeling of the eco-
the market’s response to disruptive technologies. nomic factors. Two scenarios are considered, the
In the second level, we consider the organiza- first assumes perfect competition in the 5G mar-
tional forces (internal forces) affecting the orga- ket and the second assumes an oligopolistic mar-
ket.
Université Paris-Saclay
Réseaux Innovation Territoires et Mondialisation,
92330, Sceaux, France
Contents
Introduction 1
Background on Cloud Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Chapter Abstracts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1 Methodology 13
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Research Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.2.1 Project-related previous contributions . . . . . . . . . . . . . . . . . 23
1.2.2 CBOD Objectives and Tasks . . . . . . . . . . . . . . . . . . . . . . . 23
1.2.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.3 General discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4 Epistemology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2 Decision Support Solution for Cloud Computing Adoption 39

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.1 Decision support solution . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.2 Legacy technology adoption models . . . . . . . . . . . . . . . . . . . 47
2.3 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.4 Design Constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.5 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.5.1 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.5.2 DSS objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.5.3 Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.5.4 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
v
2.5.5 Auto-learning and Self-correcting . . . . . . . . . . . . . . . . . . . . 61
2.5.6 Fine-tuning algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.5.7 Data quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.6 Collected Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.7 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Appendix - A: Active Learning Questions . . . . . . . . . . . . . . . . . . . . . . 69
Appendix - B: Accuracy metrics for recommender systems . . . . . . . . . . . . 72
3 The Network Economics of Cloud Computing 77

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.2.1 Traditional EPC vs. vEPC . . . . . . . . . . . . . . . . . . . . . . . . 80
3.2.2 EPCaaS and Multi-tenancy . . . . . . . . . . . . . . . . . . . . . . . 82
3.2.3 Pricing scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.3 Proposed Approach and Evaluation . . . . . . . . . . . . . . . . . . . . . . . 93
3.3.1 Expected scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.3.2 Proposed approach and metrics . . . . . . . . . . . . . . . . . . . . . 94
3.3.3 Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.3.4 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.4 Conclusion and Future Challenges . . . . . . . . . . . . . . . . . . . . . . . 103
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4 The Oligopoly of Cloud Computing 131

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2 Research Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.2.1 Dynamics of oligopoly . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.2.2 Game theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.3 Research Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.4 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.5 Oligopoly of Cloud Computing . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.5.1 Oligopoly of Cloud Computing: elastic demand use case . . . . . . . 141
4.5.2 Oligopoly of Cloud Computing: inelastic demand use case . . . . . . 145
4.6 Conclusion and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5 General Discussion 167

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.2 Impact on Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.3 Impact on Managerial Practices . . . . . . . . . . . . . . . . . . . . . . . . . 171
5.4 Future Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.4.1 Simulation using system dynamics . . . . . . . . . . . . . . . . . . . 173
5.4.2 Optimizing the recommender system . . . . . . . . . . . . . . . . . . 173
5.4.3 Investment and time series . . . . . . . . . . . . . . . . . . . . . . . . 174
5.4.4 Oligopoly models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
6 Conclusion 175
6.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.1.1 Generic and vendor-neutral decision support solution for cloud com-
puting adoption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.1.2 Modeling 5G networks under perfect competition conditions . . . . 178
6.1.3 Modeling 5G networks under oligopolistic conditions . . . . . . . . . 179
6.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.2.1 Cold-Start problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.2.2 Recommender system . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.2.3 Parameter fine-tuning . . . . . . . . . . . . . . . . . . . . . . . . . . 180
References 181
List of Figures
0.1 Cloud stack (Well-known layers). . . . . . . . . . . . . . . . . . . . . . . . . . . 8

0.2 Mobile Cloud computing (MCC). . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1 Cloud adoption literature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.2 Focus Areas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3 Objectives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4 Systematic analysis of cloud adoption approaches. . . . . . . . . . . . . . . . . 27
1.5 Systematic analysis of blockchain adoption approaches. . . . . . . . . . . . . . 29
1.6 Cloud adoption approaches vs. Cloud adoption risk. . . . . . . . . . . . . . . . 30
1.7 Relationships of science and technology with truth and utility. . . . . . . . . . 33
1.8 The epistemological orientations used in management research (inspired from
[1]). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.1 Decision-Making Process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.2 Conceptual Overview of the Architecture of DSS. . . . . . . . . . . . . . . . . . 44
2.3 Technology Adoption Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.4 Technology – Organization – Environment Framework. . . . . . . . . . . . . . 49
2.5 DeLone and McLean IS Success Model. . . . . . . . . . . . . . . . . . . . . . . 50
2.6 The proposed DSS’ message flow diagram. . . . . . . . . . . . . . . . . . . . . . 59
2.7 Flowchart of the processing operation i.e. to select the closest 3 technologies. 60
2.8 Fine-tuning algorithm flowchart. . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.9 Distribution of respondents by position. . . . . . . . . . . . . . . . . . . . . . . 65
2.10 Distribution of respondents by market sector. . . . . . . . . . . . . . . . . . . . 66
2.11 Distribution of respondents by company size. . . . . . . . . . . . . . . . . . . . 66
2.12 Distribution of respondents by company age. . . . . . . . . . . . . . . . . . . . 67
2.13 Technology mixes selected by respondents. . . . . . . . . . . . . . . . . . . . . 67
viii
2.14 Evaluation of the recommendation’s accuracy Using Predictive Accuracy Met-
rics - MAE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.1 Scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.2 Outcome 1: All legacy mobile operators driven out of business with the final
user cost exceeding the initial cost. . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.3 Outcome 2: All legacy mobile operators driven out of business with the final
user cost less than the initial cost. . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.4 Outcome 3: Co-existence between legacy and cloud-based mobile operators.
Some of the legacy and cloud-based operators left the market. . . . . . . . . . 126
3.5 Outcome 4: Cloud-based mobile operators join the market and leave later
because of competition from the legacy mobile operators. . . . . . . . . . . . . 127
3.6 Simulator’s Main View. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.7 First Sample Round. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.8 Fifth Sample Round. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.9 Twentieth Sample Round. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.1 Oligopolistic equilibrium with legacy operators out of business (cloud-based

operators have the same operational cost which is lower than that of legacy
operators for the considered number of clients). . . . . . . . . . . . . . . . . . . 146
4.2 Oligopolistic equilibrium with cloud-based operators out of business (cloud-
based operators have the same operational cost which is higher than that of
legacy operators for the considered number of clients). . . . . . . . . . . . . . . 147
4.3 Oligopolistic equilibrium ending up in monopoly. . . . . . . . . . . . . . . . . . 148
List of Tables
0.1 Advantages and disadvantages of cloud computing. . . . . . . . . . . . . . . . 5
1.1 Previous contributions to CBOD objectives and tasks . . . . . . . . . . . . . . 25

1.2 Blockchain selection schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.1 Chapter 2 contributions to CBOD objectives and tasks . . . . . . . . . . . . . 69

2.3 Confusion Matrix Used in Recommender Systems . . . . . . . . . . . . . . . . 74
2.2 Details of the Predictive Accuracy Metrics . . . . . . . . . . . . . . . . . . . . . 75
2.4 Details of the Classification Accuracy Metrics Used in this Study . . . . . . . 76
3.1 Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

x
Introduction
echnology is a driving force for current businesses and is among the key decisions
T and investments directly supervised by the senior executives. Correct decisions
help businesses thrive while wrong ones can dramatically limit an organization’s competi-
tiveness and possibly lead it to bankruptcy. The history is full of companies that gambled
on technologies that were proven to be a big success and others that took decisions we
currently consider as very short-sighted.
Large organizations are searching the industry for technologies developed by star-
tups and small companies that can complement their arsenal of services and technologies.
Those startups and small companies receive generous offers for acquisition in the hope
that their technologies offer market superiority for the buyer. Venture capitals, incubators,
and angel investors are also on the hunt for promising technologies, ideas, or services
that will become disruptive.
Selecting winner technologies is a tedious task, contains many ambiguities and

variables, and is relatively not very accurate. The literature does not propose any suc-
cessful formula and the decision relies on organizational learning. Epistemologically,
the organization cannot know a winner technology ahead of time but can use existing
evidence, predictions, statistics, surveys, and other techniques to justify its belief in the
correctness of its decision. Dodgeball is an example of a service acquired by Google and
then completely discontinued after 4 years.
The majority of emerging technologies will have slight impact and marginal market
penetration, and will not mature into becoming a mainstream technology. Business
1
INTRODUCTION
executives, in contrary to legislators, have to assess all emerging technologies, thus

spending their limited resources on failing technologies, and this sometimes leads them to
overlook real disruptive technologies and thus risk becoming late adopters. Although late
adoption eliminates many of emerging technologies’ intrinsic risks but also eliminates a
considerable portion of the profit margin.
Cloud computing is a typical disruptive technology that divides the market as

adopter vs. non-adopter and the adopters themselves are divided by the type of cloud
selected. The meaningful division, due to cloud computing, is success vs. failure. Selecting
whether or not to adopt cloud computing is just a meaningless question if not associated
with its impact on the organization’s success. An organization’s success depends on
internal and external factors, thus the decision to adopt a certain technology is not trivial
and does not rely on the organization’s internal state, but rather on the market’s future
opportunities and dynamics.
The literature addressed organizational cloud computing adoption using two differ-
ent and distinct approaches. The first approach considered the internal factors preparing
an organization to adopt cloud computing and determining which type is most convenient.
The second approach considered the external factors that set the market price and service
standard. In this thesis, we covered both approaches to provide a more complete answer.
This work has been applied to cloud computing but can be extended to other emerging
and disruptive technologies such as blockchain.
Blockchain is the technology supporting many applications in financial, healthcare,

and IoT sectors, but the most famous among all is the bitcoin. Unknown participants
can agree on the state of the blockchain using a consensus algorithm. The agreed upon
state is the generated block which contains the submitted transactions. "Blockchain
companies’ funding grew by 10 times from 2013 till 2016" [2]. This growth in funding
will dramatically accelerate the speed of blockchain application development. Blockchain
is an appealing technology that attracted a lot of interest from fans, investors, activists,
businesses, and academics. It enables mutually distrusted entities to conduct a variety
of transactions without relying on a central authority, while also providing integrity-
protected data storage. With all this hype surrounding blockchain, it is not always
convenient to adopt it, technically and economically [2].
2
INTRODUCTION
We will start this work by introducing cloud computing as the case study technology
used throughout the manuscript.
Background on Cloud Computing
Nowadays, business owners are faced with building and managing complex Informa-
tion Technology (IT) infrastructures. They are supposed to be able to install a variety
of software and applications, and configure, and upgrade them. However, IT resources
may become obsolete over time. Therefore, cloud computing (CC) becomes an attractive
paradigm for business owners since it allows them to deal with complex IT infrastructures
by using outsourced cloud computing platforms. In cloud computing, Information Tech-
nologies (ITs) capabilities are moved away from personal computers and mobile devices
towards a remote cloud. Its main concept is to allow users and businesses to increase and
release rapidly their computing and storage resources, in real-time, without the need to
investigate new infrastructure, or train new personnel, or buy a new software.
Many definitions for cloud computing can be found in literature, each trying to
give a complete and specific explanation of what this technology is all about. We start by
surveying these definitions and then give our own based on the lessons learned. Cloud
computing is:
• "a model for enabling ubiquitous, convenient, on-demand network access to a shared
pool of configurable computing resources that can be rapidly provisioned and released
with minimal management effort or service provider interaction." [3].
• "a parallel and distributed computing system consisting of a collection of inter-connected

and virtualized computers that are dynamically provisioned and presented as one
or more unified computing resources based on Service-Level Agreements (SLA)
established through negotiation between the service provider and consumers." [4],
[5].
• "a large pool of easily usable and accessible virtualized resources (such as hardware,
3
INTRODUCTION
development platforms and/or services). These resources can be dynamically recon-

figured to adjust to a variable load (scale), allowing also for an optimum resource
utilization. This pool of resources is typically exploited by a pay-per-use model in
which guarantees are offered by the Infrastructure Provider by means of customized
Service Level Agreements." [5], [6].
• "hardware-based service offering compute, network, and storage capacity where: Hard-
ware management is highly abstracted from the buyer, buyers incur infrastructure
costs as variable OPEX, and infrastructure capacity is highly elastic." [5], [7].
Cloud computing can be better described by two complementary definitions, one

conveying the user’s perspective, and the other conveying the provider’s perspective. So,
we define Cloud computing to be:
• “Online, infinite-like, easily provisioned and pay-as-you-go resource pool that are bound
by SLAs.” [8]
• “Pool of resource under the management of a centralized entity that is capable of

dynamically provisioning it.” [8]
The last component to define is the "Cloud Manager". It provides higher-level

functionalities such as authentication [9]. It also includes the federation manager [10],
which performs the following abstracted operations:
• Selecting a foreign cloud provider and performing the abstraction procedures needed
for having the foreign IdP (Identity Provider) trusted by the client’s IdP which is
necessary for establishing a secure connection.
• Transferring the Virtual Machines and extending the hypervisor.
The adoption of CC has a lot of benefits. The user can offload her job (computation,
storage, etc.) to the Cloud to decrease the utilization at his local devices and to get
additional functionalities he doesn’t have locally. CC allows businesses to avoid up-front
infrastructure investment. It provides customers with limitless computing resources that
4
INTRODUCTION
are available on demand, removing the need for resource provisioning planning. It also
enables businesses to start small and scale up their hardware resources just as their
demands grow. Users can pay for computing resources on a short-term basis (hourly
processing, daily storage) and release them as needed with CC. There are no upfront
costs, and software upgrades are automated. Because Cloud customers do not have to
acquire hardware or license software, the expense and complexity of owning and operating
computers and networks is minimized. All the information and applications on the Cloud
are mobile accessible and they can be accessed remotely.
Table 0.1 – Advantages and disadvantages of cloud computing.
Advantages Disadvantages
Thin clients Always online
Compatibility High-speed connection
Lower IT costs Limited features
Fewer maintenance Data confidentiality
Software patching Data availability
Computation/Storage on-demand
Easier Collaboration
Access to Documents
The advantages and disadvantages of CC as described in [11] are summarized in

Table 0.1.
Deployment models
The Cloud infrastructure supports four deployment models: public, private, hybrid,
and Community [12], [13].
• Public: It is the dominant Cloud model. It is provided for the use by the general public.
The Cloud service provider "is the owner of the Cloud and anyone can access
its services through web interfaces" [14]. Thus, Cloud customers have no control
over the location of the Cloud. The most common uses of Public Clouds are for
application development and testing, file-sharing, web applications, email service,
etc. The Public Cloud is adopted by many popular Cloud providers. One can cite
IBM, Amazon EC2, Microsoft Azure, and Google App Engine. The Public Cloud has
the advantage to be flexible, reliable, highly scalable, allows easy access to data, and
5
INTRODUCTION
cost-effective. "However, it is less secured: among all the Cloud deployment

models, the Public Cloud poses the major security issues [15], [16]" [14].
• Private: The private Cloud, also called “Internal Cloud “, is a Cloud model provided for
the use within a single organization. It is owned and managed either by the organi-
zation itself or by the Cloud Service Provider, or by some combination of both of them.
Within this Cloud infrastructure, only authorized users within the organization have
access to the Cloud resources and applications. This provides the organization with
a high control level over its data and resources. The Private Cloud infrastructure
is suitable for the business societies that have high-security requirements, and
dynamic and critical management demands. The Private Cloud is the best Cloud
model to deal with the security and privacy issues of the organizations adopting
Cloud technology. "Unlike the Public Cloud, the Private Cloud is highly se-
cured and it is control-oriented. It has more control over its resources as
these later can only be accessed within the organization boundaries." [14].
Despite all its benefits, the Private Cloud model is less scalable than the Public
Cloud, and it is very expensive to be maintained and managed [17], [18].
• Hybrid: It is a mix of several distinct Cloud infrastructures. It is a single architecture

that combines different interconnected public, private, and community Clouds. "In
fact, Clouds must be used in conjunction with each other. For instance, an
organization may use a Public Cloud to perform non-critical tasks, whereas
crucial tasks such as data processing and storage are carried out in a Pri-
vate Cloud. A Hybrid Cloud has the advantage of being flexible, scalable,
cost-effective, and more secure than the Public Cloud [17], [18]." [14]
• Community: "A Cloud of type community is a collaborative and shared Cloud

computing system. It is provided for the use by a particular group of Cloud
consumers belonging to several organizations that have common concerns,
such as banks, commercial enterprises, universities, police departments,
etc. It is owned and managed by one or more organizations belonging to
the community, or by the Cloud Service Provider, or by some combination of
both of them. The access to the data and resources within the Community
Cloud is restricted to the users belonging to the community [17],[18]." [14]
6
INTRODUCTION
Service models
"The Cloud architecture stack is composed of main layers such as SaaS

(Software as a Service), PaaS (Platform as a Service), and IaaS (Infrastructure
as a Service) [19], as shown in Figure 0.1, and additional supporting layers, such
as PasS [20]" [21] and HuaaS [19].
• Software as a Service (SaaS): is a web-based application interface where the end-

users have the capability to access applications and software hosted on the Cloud
premises. These applications are provided to end-users via the Internet, and are
accessible via a web interface. The provided software and applications are installed,
maintained, and upgraded remotely by the Cloud Service Provider on its own data
center. This allows the end-users to get rid of buying software licenses, and installing,
controlling, and upgrading the software on their computers. Google Docs, Microsoft
Office 365 and Salesforce.com are some of the well-known SaaS providers.
• Platform as a Service (PaaS): is a platform on which software and applications can

be directly developed and hosted. PaaS provides libraries allowing its users to
develop better applications faster. Unlike SaaS where the Cloud users have at
their disposal “ready to use” software, PaaS provides the users with opportunities
to design, develop, test, and deploy their own software, thus, providing them the
ability to control their software lifecycle. Examples of PaaS providers include Google
AppEngine and Windows Azure.
• Infrastructure as a Service (IaaS): IaaS is the Cloud stack’s lowest layer responsi-
ble for on-demand provisioning of servers [5]. "It consists of providing the Cloud
users several computing, storage, and network resources using which the
users can deploy their code and run their own applications" [14]. IBM, Ama-
zon Web Services, S3 (Secure Storage Service), and HP Public Cloud are examples
of IaaS providers. VRS (Virtual Resource Set) and PRS (Physical Resource Set) [5]
are two services offered at this layer. The first is hardware dependent and creates,
for the second, an abstraction layer. The second is hardware-independent and its
role is to track the running applications. Hypervisor, one of the VRS technologies,
is a widely used technology. both layers are shown in Figure 0.1.
7
INTRODUCTION
Figure 0.1 – Cloud stack (Well-known layers).
Issues
Cloud computing adopters experience the following issues in variable extents.
• Data availability: One of the main issues in Cloud computing environments is data
availability. It is a critical Cloud requirement that needs to be achieved to meet the
users’ needs. Thus, different techniques have to be applied by the Cloud Service
Provider in order to ensure that data is available and easy and quickly accessible
by the end-users, even with the increasing number of Cloud users, or in the case of
network failure. Data availability can be guaranteed by implementing data storage
redundancy, data security policies, and network optimization.
• Data security: Cloud computing saves time and money, but trusting the system is a top
priority. Data security is one of the most significant roadblocks in Cloud computing.
As a result, data security in Cloud computing is a major concern. It is vital to preserve
and secure the security of the data kept in the Cloud in order to assure the Cloud’s
dependability and the users’ confidence in this environment. The deployment model,
as well as the data protection and prevention mechanisms deployed, determine the
level of trust in a Cloud context. In a public cloud deployment, data access control
is delegated to the infrastructure owner, who is responsible for defining a security
8
INTRODUCTION
policy. The integrity, anonymity, and confidentiality of data on the Cloud are all
aspects of data security. The goal of data integrity preservation is to prevent illegal
data deletion, alteration, or fabrication. Database restrictions and transactions
are used in conjunction with a database management system to accomplish this.
When users’ private data is kept in the Cloud, the security of such data becomes
critical in order to boost the Cloud’s dependability. Authentication and access control
measures, data encryption, and data storage distribution can all help to ensure data
confidentiality.
• Costing model: The cost dimension is based on several factors. It is defined according
to the service provided, the rental period of the service, the quality of the service,
the cost of maintenance, the age of the resources, and the investment cost of the
service provider [22].
• Lock-in: It is one of the principal constraints of the Cloud environment. It concerns

the cost of services mobility between different Clouds. Some businesses are hesitant
to use Cloud computing because it is difficult to extract and move data and services
from one Cloud to another. When using Cloud services, this risk should be taken
into account.
• Platform control: Businesses are generally wary of the Cloud environment. In fact,
unlike the Cloud providers who can concept and change their platforms when and
how they want and without the consent of the customers, companies are unable to
change the technology of their platforms when they need it. Besides, the conception
of Cloud platforms does not depend on business-specific IT and business practices,
which will limit the appeal of Cloud computing.
• Business domain: This criterion refers to companies adopting the Cloud computing
solution. It is the strategy used by the company to manage its resources and to offer
the market the best services and products that its competitors. A good business model
of an enterprise is supposed to define the customers, their needs and appreciated
services, as well as the strategy to be adopted to earn money while offering the
customers the services and products at appropriate costs [23].
9
INTRODUCTION
• Disaster recovery: Cloud disaster recovery is an important feature of the Cloud com-
puting systems. It allows to backup and to restore the data of the users in disaster
cases. It consists of storing several electronic copies of the users’ data stored in the
Cloud.
Latest Applications
Cloud computing refers to a set of services delivered via the internet. Cloud
computing’s primary technology is the centralization of computer processing, services,
and specialized applications. In Mobile Cloud computing (MCC), thin mobile devices can
request cloud services through the mobile network. Virtually any service that can be
developed on the mobile device can be exteneded to the cloud. As shown in Figure 0.2,
MCC is the combination of Cloud computing and Mobile Computing.
Huerta-Canepa and Lee [24] define Mobile Cloud as a group of mobile devices in
close proximity that are all interested in processing the same data. In this instance, the
processing costs (battery power and CPU cycles) will be shared among the participating
devices, allowing the mobile Cloud to achieve its objectives. Mobile Cloud is divisive
because we want a single architecture to be sufficient for all mobile apps while also
reducing power consumption in mobile devices and, most crucially, being inexpensive.
There was no architecture in the literature that could meet the aforementioned criteria,
and none of them were standardized.
Mobile Cloud computing has been understood differently by the research com-
munity and this explains the deep difference in defining this technology and designing
its architectures. To give a generalized definition that includes all points of view and
architectures, it should be abstract and doesn’t specify detailed features. For this reason,
we define mobile Cloud computing as: “A technology which allows the user to access Cloud
services through mobile devices”.
MCC provides a lot of benefits. It allows compute and data storage to be moved
from mobile devices to the cloud. It overcomes mobile devices’ constraints, particularly
their processing power and data storage capacity. MCC also allows you to increase the
battery life of your mobile device by transferring the computation-intensive application
10
INTRODUCTION
to the cloud. Finally, it enables the delivery of e-payment services.
Figure 0.2 – Mobile Cloud computing (MCC).
Chapter Abstracts
• Chapter 1 - Methodology
This chapter identifies the research gap addressed in this work. A systematic survey
is performed, and the literature is critically analyzed. The research objectives
are clearly stated and articulated in alignment with the funding research project.
Finally, the epistemological orientation for this work is specified and argued in
comparison with the available epistemological orientations.
• Chapter 2 - Decision Support Solution for Cloud Computing Adoption

"Selecting the technologies suitable for an organization is a very critical
and difficult task especially in the absence of historical data. Organiza-
tions have to outsource expensive cloud experts to design their technology
mix and help them set their strategy. Vendors’ experts can help organiza-
tions but with the expense of less network optimization and more vendor
11
INTRODUCTION
lock-in. This chapter presents a generic and vendor-neutral decision sup-

port solution that aids decision-makers in setting their strategy towards
cloud computing. The novelty of this proposed solution is based on its deci-
sion algorithm that uses the Collaborative Filtering Recommender System
to enhance its accuracy in selecting the suitable/recommended technologies
for each user." [25]
• Chapter 3 - The Network Economics of Cloud Computing

The literature around cloud computing adoption hasn’t considered the variable
effects of the market factors and this made researchers propose a relationship
between organizational factors and suitable cloud technologies. In this chapter,
we showed the effect of market factors on the suitability of cloud computing and
proved that direct relationships between organizational factors and suitable cloud
technologies are not very accurate. This reinforced our contribution in Chapter 2
by using recommender systems to match between organizational needs and cloud
technologies. We used agent-based modeling to simulate the market and predict the
interaction between players and finally predicting the emergent market behavior.
• Chapter 4 - The Oligopoly of Cloud Computing

In this chapter, we evaluated the market behavior under oligopolistic conditions.
Two cases were considered the elastic and inelastic demands for cloud services.
For the first case, the market stability was modeled mathematically where legacy
mobile operators have a limited capacity inherited from the nature of their infras-
tructures, while cloud-based mobile operators are free from this constraint. The
market behavior of the second case was modeled using Agent-Based-Modeling.
• Chapter 5 - General Discussion

In this chapter, our contributions are transversally discussed and put into a broader
context of scientific literature and managerial practice development. In this chapter,
we also positioned our orientation for future work.
• Chapter 6 - Conclusions
In this chapter, we concluded this work and specified its contributions and limita-
tions.
12
Chapter 1
Methodology
Contents
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Research Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.2.1 Project-related previous contributions . . . . . . . . . . . . . . . . 23
1.2.2 CBOD Objectives and Tasks . . . . . . . . . . . . . . . . . . . . . . 23
1.2.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.3 General discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4 Epistemology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.1 Introduction
Cloud computing can be represented by a three-layer business model framework [26]

where each layer is related to one of its core services which are Software as a Service (SaaS),
Platform as a Service (PaaS) and Infrastructure as a Service (IaaS). More advanced and
complex business models evolved to respond to the new features of cloud computing such as
cloud federation [27]. Such business models could be evaluated as a chain of transactions
each based on one of the layers of cloud’s business model framework mentioned above. In
13
1.1. INTRODUCTION
the case of broker-based [21] or peer-based federation, new and more complex business
models are needed [28] [29], [30].
Cloud computing, like all disruptive and legacy technologies, is susceptible to

failures and downtime thus considering adopting it involves risk [31], sophisticated
contracts [32] and complex decision-making process [33]. Naldi [34] evaluated customer
losses due to cloud outages and proposed a loss model for customers using the cloud.
Mastroeni and Naldi [31] then calculated the price of insurance premiums covering cloud
outages. They also calculated the refund to equate customer’s value-at-risk, insurance
premium and Naldi’s loss model. The insurance premium is recalculated in [35] to
include new outage models and QoS metrics. An important requirement for enforcing
the insurance policy is a trusted and agreed upon outage measuring mechanism. It has
been claimed by Naldi [36] that the statistical accuracy of the measurement mechanisms
included in cloud monitoring systems, "is often neglected. Such measurements may
therefore be useless in SLA-related disputes between cloud providers and customers"
[36]. Adiani and Naldi [37] and Naldi [38] proposed a procedure to identify the failures
actually due to the cloud itself and provide a correct cloud availability measure.
Adopting cloud computing, and other disruptive technologies, is a core decision

that should be taken at the highest corporate levels because it can lead to unexpected
losses if not bankruptcy. The severity of the decision’s consequences and the value-at-
risk motivated researchers to help decision-makers select the suitable technology. The
literature is divided into two distinct methodologies for deciding on the adoption of cloud
computing. The first methodology relies on questionnaires sent to experts and decision-
makers to extract a set of determinants for cloud computing adoption. The literature
has lately shown weaknesses in this methodology such as in [39] [40] [41]. The second
methodology relies on modeling the market dynamics and predicting the success of the
adoption. The second methodology is becoming the dominant in recent years as shown in
Figure 1.1.
In the first methodology, researchers tried to identify all the determinants that
affect cloud’s business models in order to design a realistic model for a viable process.
Including all the determinants in a business model is not a guarantee for success. Youssef
et al. [42] relied on questionnaire to measure organizations’ willingness to adopt cloud
14
1.1. INTRODUCTION
Figure 1.1 – Cloud adoption literature.
computing. They based their research on two strands, the first exploring the determinants
for cloud computing adoption. The second strand focuses on the special role of technological
15
1.1. INTRODUCTION
absorptive capacity in IT adoption, where cloud computing is found to increase the

propensity in emerging countries to innovate and to adopt newer innovations at lower cost.
Tran and Bertin [43] proposed an original theoretical typology of the relationships between
Information System (IS) and their stakeholders. They focused on the relational modalities
between the IT department, the business departments, software editors and online service
providers. Four models were proposed which are: IS as product, service, access, and
platform. Lang et al. [44] determined the criteria for selecting cloud service providers
using a Delphi study. This study relied on 16 professionals to rank QoS importance to
companies with different characteristics. Zhang and Ravishankar [45] studied the vendor
capabilities in the cloud environment. They found that the five important capabilities
provided by Alibaba, china’s major IaaS provider, are cloud platform development, cloud
platform deployment, IaaS imitation, IaaS commercialization, and IaaS improvement.
They also expected that these capabilities are central to the adoption of cloud computing.
Leroux and Pupion [46] also relied on 61 CIOs from the French market to design synthetic
model for the adoption of cloud computing by local authorities. Fahmideh et al. [47]
proposed a generic cloud migration process model which uses a metamodeling approach.
They relied on domain experts to evaluate and refine their model. Khalil et al. [48]
found that even if companies want to adopt cloud computing, they cannot have successful
adoption unless they achieve improved communication between their IT and business
departments. Improved communication can be considered as a requirement for successful
adoption as confirmed by Fautrero et al. [49].
Biases influence the experts’ recommendations (especially if they are affiliated

with vendors), decision-makers’ answers (it has been shown by [39] that decision-makers
change their perception to cloud computing adoption in phases where it can fluctuate
between adoption and not over a period of time) and corporate officials’ perception to the
used adoption model or determinants [41]. Bayerl at al. [39] established the notion of
Technology Adoption States (TAS) to characterize the rationale for collective adoption
choices and to give a framework for explaining their dynamics through time. They also
shown that the collective adoption is affected by team triggered events.
Chan and Ma [41] highlighted the biased decisions of CEOs concerning IT-based
strategies. They showed that CEOs with fixed salaries have less tendency to migrate
16
1.1. INTRODUCTION
to cloud and use green IT services while CEOs with shares have higher tendency to
adopt cloud computing. Chang et al. [32] stated that cloud adoption increases bilateral
dependence and will elevate the risk of post-contractual opportunistic behavior. The
decision to migrate or not is not always meaningful. Sabherwal et al. [40] showed that
effect of the interaction between Strategic IT Alignment (SITA) and IT Investment (ITI)
can be negative under some environments. They also showed that in stable, simple,
and munificent environments, SITA reflects a rigidity that reduces the positive effect
of ITI on firm performance. So even if the right decision has been chosen, internal and
external factors participate in the success of the adoption. Saunders and Brynjolfsson [50]
found that in average 90% of a company’s market value is related to the IT’s intangible
assets such as purchased and internally developed software, other internal IT services, IT
consulting, and IT-related training. Retana et al. [51] studied the impact of basic support
vs. full support on service usage by cloud users. It was found out that users who have or
had full support tend to rely more on cloud services than customers who didn’t experience
the features of the full service. Wunderlich and Veit [52] studied the end users’ imperfect
decision making when deciding on the adoption of sustainable technologies. Trenz et al.
[53] studied the social factors end users rely on when considering the adoption of cloud
computing especially when the relation with the cloud providers is uncertain. Relying on
all those concerns, the second methodology is becoming dominant since 2017.
The second methodology considers that strategies and clients’ business models,
suppliers and competitors’ processes are very important in deciding the success of the
user’s strategy. For this reason, simulating the interaction between business models
is important to fine tune the designed model and decrease failure risk by preventing
any miscalculated behavior. Various modeling techniques are available to study the
interaction between processes and to predict the success rate under each environment.
Naldi and Mastoeni [33] calculated the fees to migrate to the cloud. They then
utilized cash flow metrics to create decision criteria. It is important to highlight the fact
that their proposed decisions are independent from competitors’ decisions and only relied
on price, failure and utilization predictions. Such decision method is typical to the second
methodology. Naldi and Mastoeni [33] also calculated the risk resulting from wrong
decisions. Sen et al. [54] applied a similar logic for providing a model to help decision
17
1.1. INTRODUCTION
makers decide whether to utilize shared or separate networks. Like [33], they relied on
organization’s internal costs and predicted prices to decide on the network type to be
used. Guerin et al. [55] developed a framework that helps decision makers evaluate the
benefits of convergence and deciding between shared and dedicated infrastructures. Their
proposal uses two operational metrics, gross profit margin and return on capacity when
considering reprovisioning resulting from excess demand. Naldi et al. [56] showed that
as the number of repositories in the cloud rises, the net advantage of investing in security
diminishes until it becomes unprofitable. As a result, unless the cloud provider promises
a better return on security investment, the cloud solution has a smaller net benefit than a
centralized solution. Gupta et al. [57] motivated researchers to use economic experiments
in information systems and presented its advantages. Aligned with this proposal Guo and
Ma [58] developed "an analytical model to study the competitive pricing strategies of an
incumbent perpetual software vendor in the presence of SaaS competitor" [58]. Multiple
strategies are being considered for the software vendor. They discovered that vendor
competition does not always result in better consumer surplus, and in certain cases, it
might even result in a socially unproductive outcome. Cocco et al. [59] simulated the
competition between traditional and on-demand software vendors. Their model can be
used as a tool to "forecast future market trends, or to plan business policies concerning
investment and pricing. Firms should calibrate this model depending on their past
business trends" [59].
Nan et al. [60] mathematically modelled the optimal pricing strategies of a cloud
service provider in an incumbent entrant setting under user upgrade cost and switching
cost. They found that in equilibrium the market structure is not unique, and this is very
important since it proves that the organizational factors are not enough for deciding on the
adoption of cloud computing. Feng et al. [61] investigated the competition between a new
entrant and an incumbent in a SaaS market and derived the optimal market entry strategy
for the new entrant. Naldi et al. [62] evaluated the effects of new entrants on existing
players using the Hirschman-Herndahl Index (HHI). Song et al. [63] studied the market
dynamics between two competing pricing strategies in contrast to the traditional fixed-fee
perpetual licensing strategy. They used Agent-Based Modelling (ABM) to study the
interaction between pricing strategies. Postmus et al. [64] used mathematical modeling
to compare pay-per-use and fixed-fee licensing. Mastroeni et al. [65] applied ABM on scale-
18
1.1. INTRODUCTION
free networks to study user side decisions. Ketter et al. [66] used multiagent competitive
gaming platforms to study dynamic electricity trading. They gave preliminary empirical
evidence for the efficacy of competitive gaming platforms. Guijarro et al. [67] studied the
mathematical equilibrium in a market for data-based services. They were able to calculate
the Nash equilibrium when certain market conditions are available. Guijarro et al. [68]
[69] studied the mathematical equilibrium for a Multi-Sided Platform Sensor-Based
Services in the Internet of Things. They deduced that an increase in any market side will
benefit all the users across the other sides. Bhattacharya et al. [70] used game theory to
compare different selling strategies under monopolistic and competitive environments.
Feng et al. [71] used system dynamics (SD) to model the adoption of the electric vehicles
compared to internal combustion engine vehicles.
Understanding the adoption of cloud computing requires exposure to multiple focus

areas and theories. I was able to identify two supporting focus areas in addition to the
methodologies discussed above. The focus areas are shown in figure 1.2.
Figure 1.2 – Focus Areas.
Focus areas 1 and 2 are discussed above and will be explored thoroughly in the re-
maining of this thesis. In supporting area 1, the literature argues that the organizational
approach is not enough to justify IT investment, but rather financial analytics should
be supportive of sometimes speculative investments. Housel and Nelson [72] proposed
using complexity and information theory with knowledge valuation analysis to measure
19
1.1. INTRODUCTION
corporate knowledge and thus assign revenue streams to IT which provides an accounting
and financial justification for potential investment in cloud computing. Cintron et al. [73]
simulated and measured the knowledge value added by core IT processes. Such simulation
is critical in predicting the financial feasibility of the potential adoption. Additionally, in
uncertain conditions, real option analysis [74] provides a rigorous framework that helps in
valuating investments. This framework can be extended to consider investments related
to cloud adoption. In [75] the authors showed that the adoption of single innovation does
not concur impact or economic performance. Even multiple adoptions may not guarantee
impact. Economic performance is achieved through "innovative combinations" and is
very specific and case dependent. Extending this discussion from environmental inno-
vation into disruptive technologies has profound impact as is shows that organizational
approach’s recipe model falls short to deal with "innovative combinations" and case depen-
dency. In the same context, Cecere et al. [76] discussed the lock-in and lock-out factors
that prevent an organization from adopting eco-innovations. This work, if extended or
generalized, can explain the possibility of being locked out from adopting a technology
even if the organization wants. This literature body, called supporting area 1, does not
focus on technology adoption, but is critical in understanding the theoretical basis for the
first methodology.
In supporting area 2, we can rely on complexity/complex systems theory to under-

stand the emergent rules that govern the market dynamics and consequently technology
adoption. I will quote Brian Arthur’s groundbreaking work that was summarized in [77].
Economics, the understanding of market dynamics, is a complex system since the
"recursive loop that makes the economy a complex system [77]"
and thus, the theoretical foundations of complexity should be applied to understand

these dynamics. This sub-field has been named complexity economics [78]. The strategic
decision of technology adoption is neither solely dependent of internal factors, nor a game
theory decision between multiple actors, but a loop of tactics and dynamic game theory
decisions. Arthur defined this concept clearly
"Companies in a novel market may have different technologies, different motivations

and different resources, and they may not know who their competitors will be or, indeed,
20
1.1. INTRODUCTION
how they will think. They are subject to what economists call fundamental uncertainty...
About these matters there is no scientific basis on which to form any calculable probability
whatever. We simply do not know. As a result, the decision problem faced by agents is
not logically defined and, so, it cannot have a logical solution. It follows that rational
behaviour is not well- defined. Therefore, there is no ‘optimal’ set of moves, no optimal
behaviour [77]."
Agent-Based Modelling and mathematical modelling are the natural methods for
understanding the complex nature of markets and its emerging nature. Arthur defined
this concept clearly
"it was computation more than anything else that allowed economic theorists to
venture beyond the standard neoclassical assumptions — for instance, to allow complicated
inductive reasoning and compute its consequences. If we turn these new possibilities into a
theoretical framework, we get complexity economics, or something like it. If we turn them
into a solution method, we get agent- based computational economics.So there is no well-
marked boundary between the two approaches. One could, therefore, regard agent- based
computational economics as a key method within the framework of complexity economics;
or one could regard complexity economics as a conceptual foundation behind agent- based
economic modelling [77]."
Adoption can be looked at as a propagation of change that is guided and dictated

by network dynamics. Arthur defined this concept clearly
"The topology of a network matters as to whether connectedness enhances its

stability or not. Its density of connections matters, too. When a transmissible event
happens somewhere in a sparsely connected network, the change will fairly soon die out for
lack of onward transmission; if it happens in a densely connected network, the event will
spread and continue to spread for long periods. So, if a network were to slowly increase
in its degree of connection, the system will go from few, if any, consequences to many,
even to consequences that do not die out. It will undergo a phase change. This property
is a familiar hallmark of complexity. Notice that the propagation of events brings time
inexorably into systems; without such propagation, time disappears."
21
1.1. INTRODUCTION
Since the market of non-adopters can be represented as a network of nodes, where

the links between these nodes represent comparability and decision trends, the stability
of this network can be predicted. In other words, the ability of non-adopters to withstand
a new trend, cloud computing in our case, and thus maintain their choice of non-adoption
has been elegantly represented by Gao et al. [79]. The relationship between nodes has to
be mutualistic for [79] to stand, otherwise the variation proposed by Arnoldi et al. [80]
has to be applied. Additionally, the competition between adopters and non-adopters has
been theoretically modelled by Aguirre et al. [81]. This theoretical work is the building
block for any real understanding of the adoption of cloud computing and generally any
disruptive technology.
In this work we revisited both methodologies. We found that first methodology’s

weaknesses can be categorized into:
1. Implementation weaknesses: these weaknesses are due to limitations in the deter-

minant’s selection, questionnaire distribution and rigidity of the outcomes. For the
outcomes to be meaningful, the questionnaire must identify specific environment
such as: country, sector, size and others. This makes the number of respondents
small enough for any bias to affect the accuracy of the outcome. Inversely, accumu-
lating responses from different countries, sectors or sizes also decreases the accuracy
of the outcome.
2. Conceptual weaknesses: these weaknesses are due to the nature of the methodology
such as: experts’ biased recommendations, decision-makers’ biased answers and
corporate officials’ biased perception to the used adoption model or determinants.
The implementation weaknesses will be tackled in Chapter 1. The second method-

ology is revisited in Chapters 2 and 3.
22
1.2. RESEARCH PROJECT
1.2 Research Project
This work is under the umbrella of the CBOD Project (Cloud-Based Organizational Design
– ANR). CBOD is “inscribed in fundamental research, but with applicability in real world
settings, aims at developing the management and IS community knowledge on cloud
computing, from an organizational and business point of view” [82].
1.2.1 Project-related previous contributions
Previous contribution on the CBOD project includes [83] which found that "perceived
usefulness, perceived ease of use, complexity and compatibility are key factors for cloud
adoption. It also found that most small cloud providers have more stable and better
computing performance than large cloud providers and that the performance of CPU
impact price significantly" [82].
1.2.2 CBOD Objectives and Tasks
The CBOD project identified 6 objectives which are [82]:
1. "Make an inventory of cloud-based research and approaches in the domain of man-

aging information systems." [82]
2. "Elaborate a conceptual and analytical holistic framework of cloud computing for

organizations. A particular attention will be given at articulating the different
dimensions, rather than considering each of them separately." [82]
3. "Elaborate different means to model and simulate cloud-based design in organiza-

tions enabling a more active and intelligent exploration of Cloud-based solutions
from an organizational and business point of view. The means considered include ap-
proaches such agent-based modelling, system dynamics, equation-based modelling,
23
ontology modelling, and associated tools, with a particular interest to solutions that
are very comprehensible (both in exploring and designing the models) (action led by
Paris Sud)" [82]
4. "Provide a techno-economic model of cloud technology in relation to their dimension-

ing, and cost models, so as to inform the selection of the way services are offered and
charged to the customers, and for instance to guide the choice of the most effective
business models (action led by INRIA)" [82]
5. "Investigate issues related to Cloud Computing solutions adoption by organizations

(action led by Telecom ParisTech)" [82]
6. "Build a set of cases and scenario able to reflect the diversity of issues, situation and
perspective of Cloud-based information systems (e.g. business models, resistance to
change, national specificities and international comparison, etc.), that will be also
used to validate the approach." [82]
To do this, the CBOD project specified 6 tasks which are [82]:
1. "TASK 1: Cloud Computing: the Conceptual and Analytical Foundation

The goal is to set up the foundation of the project, notably by making an overview
of the state of the art of cloud computing from a non-technical perspective, and
contribute to the establishment of a common ground amongst the partners." [82]
2. "TASK 2. Preparing the Initial Model (Inria/Paris-Sud) It will be aimed at

elaborating an economic modeling of cloud computing systems and technologies."
[82]
3. "TASK 3: Business Design (Paris-Sud/TPT) The objective is the analysis of the

different modeling advanced approaches (agent-based, system dynamics, etc.), and
in the elaboration of the CBOD modeling toolbox (conceptual and technical)." [82]
4. "TASK 4: Exploration of organizational appropriation or rejection Cloud

Computing solutions (TPT/Paris-Sud) Addresses the modeling of the adoption
of the cloud in the organisations." [82]
24
5. "TASK 5: Modeling and simulation of scenarii (all) It is first aimed at modeling

the different non-technical aspects of cloud computing systems using the CBOD
toolbox, and notably the economic models and the adoption models developed in
Task 2 and Task 4. This task will also assess the different modeling approaches
(and more specifically the CBOD toolbox). Finally, we will summarize the findings
of the project in relation to the non-technical aspects of cloud computing." [82]
6. "TASK 6: Management, coordination and dissemination" [82]
The project-related previous contributions, discussed in section 1.2.1, are reflected

in table 1.1.
Table 1.1 – Previous contributions to CBOD objectives and tasks
Status Status
Objective 1 Done Task 1 Done
Objective 3 Pending Task 3 Pending
1.2.3 Objectives
In this work, our general aim is completing the remaining tasks and objectives set by
CBOD. Studying the organizational impact of cloud computing adoption is a declaration
of the new technology’s disruptive nature. Our aim is to help decision makers assess the
value of disruptive technology (and cloud computing in particular) on their organizations
to be able to draw better conclusions. Our objectives can be categorized into economic
and organizational which are:
• Objective 1 (The economic objective): is understanding the market’s response to

disruptive technologies such as cloud computing. This understanding is critical
for predicting future market trends that can be translated into decisions at the
organizational level.
25
1.3. GENERAL DISCUSSION
• Objective 2 (The organizational objective): is providing practitioners and decision

makers with tools to help them analyze internal and external factors that influence
their technology adoption decisions.
The objectives are presented graphically in Figure 1.3.
Figure 1.3 – Objectives.
1.3 General discussion
Although it may seem from the above discussion that the literature is offering two distinct
alternative approaches and has favorized one over the other, the literature doesn’t contain
any comparison between the two approaches nor a systematic analysis specifying the
relationship between them. The relationship can span from complimentary to alterna-
tive, but without systematic analysis it can be ambiguous and possibly subjective. My
26
observations pushed me to believe that the organizational and economic approaches have
the following relationship:
1. The organizational approach precedes the economic approach as the first attempt to
understand an emerging technology.
2. The organizational approach provides, on the long run, an inexpensive, simple and
readily available hand rule for decision makers when the economic approach is too
complicated, too expensive or too slow to be evaluated.
In the Figure 1.4, we plot the number of publications per year for every approach.
The x-axis of Figure 1.4 is the publication year, while the y-axis is the number of publica-
tions (cumulative).
Figure 1.4 – Systematic analysis of cloud adoption approaches.
As seen in Figure 1.4, the organizational approach precedes the economic approach
by 6 years. This observation is consistent across other emerging technologies. Table 1.2
shows the blockchain selection papers and their relative approach.
27
Table 1.2 – Blockchain selection schemes
Ref. Author Decision scheme Publication year Approach

[84] Greenspan Questionnaire 2015
[85] Suichies Flow-chart 2015
[86] Nandwani Flow-chart 2016
[87] PWC Questionnaire 2016
[88] Birch et al. Flow-chart 2016
[89] Saiko Flow-chart 2016
[90] IBM Flow-chart 2016
[91] Lewis Flow-chart 2016
[92] Meunier Questionnaire 2016
[93] Quindazzi Flow-chart 2016
[94] Deloitte Questionnaire 2016
[95] Peck Flow-chart 2017
[96] Lixar Flow-chart 2017
[97] Verslype Flow-chart 2017
Organizational
[98] Lo et al. Flow-chart 2017
[99] Verified ICOs Flow-chart 2017
approach
[100] Lin et al. Flow-chart 2017
[101] Meuller Flow-chart 2017
[102] Henkel Flow-chart 2017
[103] Maull et al. Flow-chart 2017
[104] Xu et al. Flow-chart 2017
[105] CapGemini Questionnaire 2017
[106] Klein et al. Questionnaire 2018
[107] Pahl et al. Flow-chart 2018
[108] Wessling et al. N/A 2018
[109] World Economic Forum Flow-chart 2018
[110] Wust Flow-chart 2018
[111] Gardner Flow-chart 2018
[112] Gatteschi et al. Questionnaire 2018
[113] Koens et al. Flow-chart 2018
[114] Scriber Questionnaire 2018
[115] Chand Flow-chart 2020
[2] Bou abdo et al. Neural network 2020
28
In the Figure 1.5, we plot the number of publications per year for every approach.
The x-axis of Figure 1.5 is the publication year, while the y-axis is the number of publica-
tions (cumulative).
Figure 1.5 – Systematic analysis of blockchain adoption approaches.
As can be seen in Figure 1.5, blockchain adoption decision schemes started to

be discussed in 2015 following the organizational approach and none has followed the
economic approach yet. If all emerging technologies follow the same trend in adoption
approaches, we should expect a new blockchain adoption schemes following the economic
approach soon.
Plotting the cloud adoption risk publications on the same graph with the two cloud
adoption approaches, as shown in Figure 1.6, instantly visualizes the direct correlation
between the initiatives to understand cloud adoption risk and initiatives to rely on the
economic approach for deciding cloud adoption. The “economic approach” and “cloud
adoption risk” have clear positive correlation (r = 0.88).
29
1.4. EPISTEMOLOGY
Figure 1.6 – Cloud adoption approaches vs. Cloud adoption risk.
1.4 Epistemology
Epistemology is a philosophical branch which aims to study the sources of knowledge. It

has been diversely defined in literature with intersections and distinctions in respective
definitions. Some of the definitions, found in literature, include:
• “Epistemology is the branch of philosophy that evaluates competing views of the

morality, nature, standards, sources, and functions of knowledge” [116].
• “Epistemology is the study of the criteria by which we can know what does and does
not constitute warranted, or scientific, knowledge” [1].
• The functional/systematic approach for defining epistemology in contemporary man-

agement is “oriented on the creation of integrated systems, verification of truth
with the help of objective quantitative methods” [117]. The symbolic/interpretative
approach, on the other side, considers epistemology to be built on the assumption of
“a constructional and conventional character of the social and organizational reality
30
1.4. EPISTEMOLOGY
[118]. Organizational order does not exist objectively but is being continuously
maintained, reconstructed and modified by individuals and groups acting in and
around an organization” [117].
The distinctions in epistemology definitions create a diversity in research methods

and goals. As specified by Levant and Zimnovitch [119], the goal in management research
should be the utility of the theory rather than sheer correctness. We will demonstrate
throughout the following how our objective was to search for the best theory regarding
our work.
The principle of epistemology in research is the status of knowledge, its formation

and elaboration, and the obtained results. The sources of knowledge can be divided
into four: The intuitive knowledge based on intuition, the rational knowledge based on
trusted information, the empirical reasoning based on demonstrated facts and the Logical
knowledge based on logical reasoning. One can use the 4 sources or can refer to some of
them. In this thesis, we referred to the rational knowledge when studying our literature
review and when discussing the results obtained to build up new ideas and theories.
While using questionnaires to extract a set of determinants, we were going throughout
empirical reasoning. At the end, our results are the fruits of a logical reasoning ending
up in a logical knowledge.
Two of the well-known epistemological orientations are positivism and pragma-

tism. These two approaches have a different perception of knowledge. In the positivist
approach, a theory cannot be considered as true unless it is measured and verified, while
in the pragmatist approach, the correctness of a theory is related to its practical interest.
The pragmatists define the truth of a theory in terms of its success and utility. Thus,
“truth serves practical interests” as specified in [120] and can be measured according to
satisfaction. However, Popper separates the notion of utility and truth, thus opposing
the pragmatist position which combines value and truth [121] (value and utility are used
interchangeably in this section). According to Popper, "falsification is what distinguishes
scientific logic from other types of discourse. He considers that scientific knowledge
develops according to a ‘verisimilitude’. Thus, there are no statements that can be said to
be unquestionably true" [121] . A theory should make postulations that can be approved
31
1.4. EPISTEMOLOGY
or declined experimentally [121]. According to [122], the epistemological issue is not truth
but action. Pragmatism is concerned with action and the interplay between knowledge
and action. The authors of [123] proposed the experience-based research as a general
framework for management research.
Management sciences are part of human sciences [119] hence, a “criteria of utility
rather than ‘truth’ should be employed in making judgements on the adequacy of theory”
[124]. The stands, from utility and value, used in management research, as specified by
Levant and Zimnovitch [119], are as follows: Although an important part of management
research falls into dogmatism and sophism, action research provides real perspectives
and the hermeneutical-historical approach valuable information. It is what is called
Dead ends and prospects. On one hand, the management sciences theory degenerates
into dogmatism when it pretends to express the truth and tries to produce universal
regulations in order to succeed. If science pretends to apply to utilitarian things, it will
deteriorate into dogmatism. On another hand, the application of rhetoric as a technique
for management and its teaching is quite common. However, the use of rhetoric as a goal
or science will at last lead to a sophism that shifts utility over truth. The connection
between sophism and marketing is quite clear. The last situation, raising a technology,
which is relative to a context, a point of view, or a preference, to the rank of truth, would be
to deny it and destroy its claims to universality. This is sophism. According to [125], they
both, ultimately, aim to seduce, convince, and persuade. Our research aims to provide real
perspectives and not speculation regarding modeling the factors influencing the adoption
of cloud computing and the success of the adoption decision.
The stands from utility and truth can be reformulated into the stands between
science and technology. As shown in Figure 1.7, science and technology have four possible
relationships with truth and utility. These relations can be described as follows:
• Science is based on empirical content as well as a set of assumptions that can be

rejected. Thus, it is related to the concept of truth and not utility, and without any
assertion to accomplish it definitively.
• If science pretends to apply to utilitarian things, it will deteriorate into dogmatism.
• Technology is know-how and is linked to what is beneficial. Technology is designed
32
1.4. EPISTEMOLOGY
pragmatically: “the great advantage of a pragmatist’s position residing in the inter-

dependence it establishes between what is true and what is useful” [126]. It is an
operative knowledge in a situation.
Figure 1.7 – Relationships of science and technology with truth and utility.
The epistemological orientations used in management research are:
• Positivist and Neopositivist Epistemology: Positivism remains the dominant orien-

tation in management sciences [1], although this dominance is camouflaged “since
hardly anyone openly applies a positivist label to their own work”. Logical positivist
orientation has the following 4 claims [1]:
– Observations can be neutral, value-free and objective.
– All theoretical statements should be subject to empirical testing.
– All sciences, including social sciences, should follow the natural sciences model.
– The task of science is to control social and natural events.
• Conventionalist Epistemology: Conventionalism is centered around the absence of a

theory-neutral observational language. The consequences of this assumption have
been clearly stated in [1], [127] as:
– Scientific statements are the creations of the scientist, thus cannot be falsified
using external observations.
33
1.4. EPISTEMOLOGY
– Acceptability of a scientific statement is not based on universally valid criteria

or objective standards of evaluation, but rather by the scientist’s subjective
understanding of reality which is usually derived from socially sanctioned
conventions.
– The absence of a theory-neutral observational language makes scientific state-

ments objectively uncontrollable.
• Postmodernist Epistemology: It provides a relativist stand and dismisses the rational

certainty of positivism [1]. This stand is due to the limitation in our observational
methods. It has been clearly stated by De Cock as “The consequence of a postmodern
stance is that we are advised to stop attempting to systematically define or impose a
logic on events and instead recognize the limitations of all our projects” [128].
• Pragmatism and Critical Realism: Critical Realism considers that people’s observ-
able behavior is not explicable using empirical methods and requires the casual
context of non-empirical structures and their interactions.
• Critical Theory: Critical Theory is oriented towards power structures and related
issues such as "exploitation, asymmetrical power relations, distorted communication,
and false consciousness" [1].
• Constructive Epistemology: Constructive empiricism considers that science aims to

provide theories that are empirically adequate [129]. It also considers that scientific
knowledge is constructed by scientists, compared to the observed environment, but
not discovered from the environment. Constructive empiricism is not limited to one
single valid methodology for social science. Dicken reported that Van Fraassen [130]
defined constructive empiricism to be the view about:
– What is to make an epistemic judgement?
– The obligations of being a rational scientific agent.
– How best to understand the very philosophical debate concerning the aim of
science?
34
1.4. EPISTEMOLOGY
The epistemological orientations used in management research are represented in

Figure 1.8. The Y-axis represents the ontological stand of each orientation. The X-axis
represents the epistemological stand of each orientation.
Figure 1.8 – The epistemological orientations used in management research (inspired from [1]).
Our research is oriented towards epistemological objectivism since reasoning is the

foundation for our work without relying on personal presumptions and beliefs. The first
part of this work is oriented towards ontological subjectivism taking the organization as
reference. Constructing the decision based on the organizational situation or perspective
without relying on our presumptions is critical for escaping biases. The second part of
this work is oriented towards ontological objectivism since the foundational assumptions
of economics truly believe that the market is the interactions between supplies and clients
irrespective of any single individual or organization.
35
1.4. EPISTEMOLOGY
The first part of our research is aligned with the constructivist approach since
we have been immersed in observing the environment of our study to know it, model
it and to build our results accordingly. Our constructivist approach is based on three
principles: the phenomenological hypothesis, i.e. the relation between economic factors
and corporate’s success, the relation between subject and object, i.e. how the object of
knowledge is constructed through comings and goings of analysis and interpretation, and
the construction of new truths. Other proposals found in literature using decision trees or
rigid recommendations follow ontological subjectivism and epistemological subjectivism
orientations and thus fall prone to biases.
Reality is constructed from the work that is done on it, from the observations that
emerge from it. Knowledge is the fruit of constructions that are realized progressively,
that are not given by themselves, but that are built continuously. This reality is not
directly accessible, but it results from interpretations. It cannot be directly deduced
according to pre-defined criteria, but it is deduced according to standards of feasibility
that must be reviewed on a case-by-case basis.
In the first part of our work, we considered the factors involved in generating
accurate decisions without oversimplification or overgeneralization. The factors used are
based on a previous contribution of the CBOD project [131]. Similar factors have been
grouped by Loukis et al. [132]. The constructed models are not necessarily final, and the
included factors are not necessarily complete. Increments can be done to the constructed
models to included missed factors that can enhance the model’s accuracy. The use of a
recommender system and the designed auto-tuning component help enhance the accuracy
of the model without manual intervention from our side.
This construction is carried out with the actors concerned, it is not a theoretical
construction, but a construction that combines theory and practice in the field of action.
It is an articulated approach, based on a learning process between me as a researcher
and the experts in the field.
The methods used define the project that allowed the construction of scientific
truths or scientific knowledge. According to this constructivist approach, such research is
a project in continuous progress. This evolution implies that it is not an elaborated and
36
1.4. EPISTEMOLOGY
definitive reality, but several realities likely to evolve according to the individuals who
apprehend it.
Self-Adaptive decision-making systems usually follow the constructivist approach

and this can be seen similar research such as in [133] and [134]. In [133] the authors
stated that "particularly challenging in unknown and changing complex environments,
where providing a complete a priori representation to the system is not possible" [133].
This is also applicable to our research and consequently following a onstructivist approach
serves our aim the most.
The second part is based on a positivist approach which can be defined by the
consideration of knowledge obtained by observation and measurement. It is the study of
the social world. This approach is based on an hypothetico-deductive model that verifies
and experiments hypotheses set a priori by operations and measures in order to find
causal relationships between the elements of the hypothesis. It consists of passing from
observable social reality to a generalization considered as a law. These relationships are
important since they will be considered as the main base of Management.
In our studies, we started by defining the terms and concepts used in order to
establish a clear and valid communication. Then, we established hypotheses defined
following the research needs. The second part of the research is based on a quantitative
method which was a market simulation in our case. This part has the role to test the
hypothesis in order to confirm it.
These results were based on statistical analysis which led to a concentration on

facts. The results were validated by rational and logical ways of reasoning in order to be
generalized. The results can be formulated in abstract and universal laws which can be
stated as “Internal and external factors affect the success of an organization’s decision”.
37
Chapter 2
Decision Support Solution for Cloud

Computing Adoption
Jacques Bou Abdo, Najma Saidani, Ahmed Bounfour 1
Contents
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.1 Decision support solution . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.2 Legacy technology adoption models . . . . . . . . . . . . . . . . . . 47
2.3 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.4 Design Constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.5 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.5.1 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.5.2 DSS objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.5.3 Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.5.4 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.5.5 Auto-learning and Self-correcting . . . . . . . . . . . . . . . . . . . 61
2.5.6 Fine-tuning algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 62

1
Jacques Bou Abdo, Najma Saidani, Ahmed Bounfour, “Organizations’ Investment in Cloud Computing:
Designing a Decision Support Platform”, 14th International Conference on Data Science (ICDATA’18), Las
Vegas, July 31, 2018.
39
2.1. INTRODUCTION
2.5.7 Data quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.6 Collected Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.7 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Appendix - A: Active Learning Questions . . . . . . . . . . . . . . . . . . . 69
Appendix - B: Accuracy metrics for recommender systems . . . . . . . . 72
2.1 Introduction
"The extent to which an organization should utilize cloud computing is a dif-

ficult decision to make, especially in the absence of reliable historical data and
normatively acceptable decision principles" [25]. Because of competition and rapid
technological advancements, the cost of making a bad strategic decision is just too high
to be tolerated.
Existing research has focused on defining and ranking the elements that influence
adoption decisions, "as well as the impact of a predetermined set of criteria on the
intention to utilize cloud computing services" [135], [136], [137], [138], [139], [140],
[141]. Other suggestions attempted to use traditional economic methods by combining
measurable and non-quantifiable components into a single index [142], [143]. The presence
of a linear connection between the influencing factors is implicitly assumed by these
writers, and this is non-provable. The determinants overlap in influence; thus, any
theoretical decision scheme will be very difficult to deduce. On the other hand, inducing
an empirical interaction formula as a function of the determinants would be the most
convenient taking into consideration the limitation of the available data. As will be
discussed later, we debate that non-personalized collaborative filtering recommender
systems are best suitable to induce an empirical relationship between the determinants
and the available cloud technologies/services.
40
2.2. BACKGROUND
A previous contribution of the CBOD project [131] resulted in an exhaustive list of

determinants that participate in the cloud adoption decision. A similar list of determinants
can be seen in [132] The literature has slightly touched the issue of organizational
cloud adoption from a Decision Support Solutions (DSS) perspective. To the best of our
knowledge, his work is the first attempt to offer a DSS on cloud computing adoption
that is holistic, vendor-neutral, general (not confined to particular cloud services), and
non-trivial. The research’s ultimate product is a publicly available Web-based DSS tool
that decision-makers may utilize to assist them decide on their cloud computing approach.
2.2 Background
2.2.1 Decision support solution
Humans spend the majority of their days in front of screens and devices, connected to the
Internet. Whether in their homes, institutions, colleges, hospitals, etc. anywhere and
everywhere technology is integrated. They are required to take a multitude of decisions
that go from simple daily decisions to life-changing ones. However, with the huge stream
of information to which the users are being exposed the problem of information overload
arises. The users are receiving a large quantity of data in a short time: they are not being
able to convert information to useful knowledge. This affects the decisions negatively
to the point of blocking it completely. In order to solve this issue, decision support
systems became widespread solution. Decision support systems (DSS) combine computer
capabilities and human skills in performing advanced operations such as data analytics,
reporting, planning, and modeling. This is not a new field of study, researchers have been
focusing on developing solutions for complex decision-making problems since the 1970s
[144].
DSS are systems that deal with problems that have semi-structured or unstructured
[145] stages. Simon places the problems dealt with DSS on a range from programmed,
easy to solve well-structured problems to non-programmed, difficult, and ill-structured
41
2.2. BACKGROUND
problems [146]. DSS is an incarnation of the concept that computers place an important
role in decision-making. They address the requirement of a certain design strategy
supporting the cognitive process of human decision.
Three types of DSS exist [144]: active, passive, and cooperative. The active DSS
can generate solutions or suggestions. On the other hand, a passive DSS cannot produce
such solutions or suggestions, but it is able to support the process of decision-making.
A cooperative DSS allows the decision-maker to modify the recommendations that the
system provided before being sent back to the system for validation and another round of
decision recommendation.
Figure 2.1 – Decision-Making Process.
The decision-making process can be summarized in a number of steps as shown in

Figure 2.1 [144]:
1. First Step: Problem Requirement Definition Once recognized, the problem needs to
42
2.2. BACKGROUND
be defined. By definition, a clear statement of the issue is meant: a description of

the initial conditions and the desired ones. Afterward, the next part of this step is
the determination of the requirements in order to obtain a list of absolute goals and
exigencies.
2. Second Step: Alternative Generation As the name implies, this stage is in charge of
coming up with alternate answers to the uncertainties of decision-making. One of
the most important characteristics is that the alternatives must fulfill the previously
specified standards. The "Intelligence Phase" is formed when steps one and two
are combined. The system is tasked with locating, identifying, and formulating
the circumstance or issue. A summary of the present state of the detected issue is
provided. As a consequence of this step, a decision statement is issued.
3. Third Step: Model Development Once the alternatives are generated and the fitting
ones are selected, the model needs to be developed. The aim of the model development
is the analysis of the different alternatives. The model that best compares all the
alternatives is the one that achieved the selected criteria and goals the soonest.
4. Fourth Step: Alternative Analysis After this best model is developed, an alternative
analysis is conducted. This analysis concludes the evaluation of alternatives against
criteria. Many tools exist and research is still ongoing to find which method is
appropriate the most to which type of issues. The aim of these researches is to
differentiate between the pros and cons of using a certain method instead of another
and to check if changing the method used affected and changes the decision. At the
end of this step, the solutions are validated against the problem statement.
5. Fifth Step: Choice of Decision Design The choice should be made in a way that the
solution best achieves the objectives of the decision process, satisfies the desired
state, and meets the requirements
6. Sixth Step: Implementation The chosen decision design is implemented. These steps
are the flow of the “Choice Phase” model. The developed alternatives to the design
phase are here selected. A decision or model is the final product of the process.
A DSS possesses several fundamental components as shown in Figure 2.2. As can
43
2.2. BACKGROUND
Figure 2.2 – Conceptual Overview of the Architecture of DSS.
be seen in Figure 2.2 there exist five main applications of DSS that consist of multiple
techniques and sub-technologies for decision making as shown in the coming subsections.
2.2.1.1 Model-driven DSS
For decision support, a model-driven DSS employs optimization, finance, simulation, and
algebraic decision-analytic models. It is intended to assist decision-makers in understand-
ing a particular situation and allowing the user to alter model parameters. In general,
it is not data-intensive: big databases are not required. They may, however, need to be
removed for certain studies. One or more quantitative models that offer functionality are
the key component of this DSS’s architecture.
Analytical tools that use algebraic models assign a basic degree of capability. They
44
2.2. BACKGROUND
are typically created in spreadsheets and are frequently utilized in the development of
model-driven DSS systems. Simulation approaches, optimization, and mathematical
programming models, as well as decision analysis, are used to design and develop increas-
ingly complicated decision-making models. Models in a model-driven DSS are, in general,
a simplified depiction of reality.
Model-driven DSS using simulation techniques conduct a multitude of experiments

to display the effects of alternative courses of action and conditions [147]. "Simulation
methods are commonly used in the area of supply chain management. Several examples
are visual simulation, system dynamics, agent-based and multi-agent simulation, discrete
simulation, and Monte Carlo simulation" [144].
In many environments, optimization models are integrated into DSS. Mostly,

in supply chain management and production & operations management, the area of
optimization models has gained importance. Nowadays, multiple decision support models
exist for the different levels of the supply chain, including logistics planning, production
planning, and scheduling and demand management.
Decision analysis models are the statistical methods and tools such as probabilistic
forecasting, decision tree analysis, AHP (analytical hierarchy process), and multi-criteria
decision analysis. A decision analysis aims at the discovery of the most favorable alterna-
tive under the given situation.
2.2.1.2 Data-driven DSS
A data-driven DSS allows to access and manipulate structured data. It can handle real-
time data and time-series of external and internal company data. "Data-driven DSS are
separated by functionality. Simple file systems are accessed by retrieval tools and queries
that facilitate an elementary level of functionality" [144]. More in-depth functionality is
provided by management reporting systems such as EIS (Executive Information Systems)
and Data Warehouses. They allow computerized tools to manipulate data. The data-
driven DSS that provides the highest mode of decision support and operation is the BI
(Business Intelligence) Systems and the OLAP (Online Analytical Processing). BI systems
45
2.2. BACKGROUND
help in the creation of the decision: they trigger, manipulate, and analyze information
or data stored in historic databases. Their main objective is to ensure a higher quality
of information for decision making. The efficiency of the data-driven DSS is linked to
certain requirements: to have access to a large quantity of data and also high quality, i.e.
accurate, organized, and well-structured data.
2.2.1.3 Communication-driven DSS
Communication-driven DSS connects users through information and resource sharing

environment. Users can communicate and collaborate through a hybrid network and
electronic communication. One main sub-category is group decision making. Over several
years of research, the GDSS (group DSS) are developed [145]. They include problem
structuring techniques like modeling and planning tools.
2.2.1.4 Document-driven DSS
The quantity of documents, videos, images, sounds, correspondence, and hypertext docu-
ments in companies is increasing exponentially. Not only the amount is huge but also
the size of the documents is very large. Document management is now a must. The
worldwide web contains very big document databases which augmented the development
of document-driven DRR. Search engines that are linked to these systems form decision-
aiding tools. It is important to note that documents are not standardized in patterns and
are therefore unstructured. Information retrieval systems are used to extract information
and give a structure for documents to provide better support for decision making.
2.2.1.5 Knowledge-driven DSS
Knowledge-driven DSS originates from AI (Artificial Intelligence), more specifically In-

telligent DSS. In the knowledge-driven DSS communication mechanisms, data mining
technologies, management expert systems, and AI are integrated. Intelligent DS can be
divided into two evolutionary developments.
46
2.2. BACKGROUND
The first type is about rule-based expert systems [148]. It is commonly used in
production systems, for scheduling. These expert systems use heuristics, strategies
leading to the correct solution for a problem. In this type of system, human expert
knowledge is necessary.
The second generation relies on genetic algorithms, fuzzy logic, and neural networks
[149]. These methods are "similar to linear programming models but they conduct random
experiments by selecting variables without identified values to find the fitness function"
[144].
2.2.2 Legacy technology adoption models
2.2.2.1 Technology Adoption Model (TAM)
TAM is a widely used theory in Information Systems that studies the adoption level of a
studied technology under a certain organizational context [150] as shown in Figure 2.3.
It is based on two beliefs which are:
• "Perceived usefulness: the degree to which a user thinks a technology would enhance
performance or productivity in the workplace" [150].
• "Perceived ease of use: the degree of lack of effort required by the user in adopting a
given technology" [150].
These beliefs help in predicting users’ attitudes toward the evaluated technology
and thus its overall effectiveness. The predicted users’ attitudes and intentions to use
the studied technology affect the behavior of actual system usage as shown in Figure 2.3.
2.2.2.2 Technology – Organization – Environment Framework (TOE)
The TOE framework identifies the organization’s adoption influencers to be three elements
of a firm’s context [151] as shown in Figure 2.4. These elements are:
47
2.2. BACKGROUND
Figure 2.3 – Technology Adoption Model.
• Technological context: includes the technology already implemented at the organi-

zation and the technology available by vendors.
• "Organizational context: includes the firm’s resources such as intra-firm communi-

cation processes, firm size, and the amount of slack resources" [151].
• "Environmental context: includes the market’s structure such as the presence or

absence of technology service providers and the regulatory environment" [151].
2.2.2.3 DeLone and McLean Information Systems Success Model
"It is a framework and model for measuring the complex dependent variable in IS research"
[152]. The variables belong to six interrelated dimensions, as shown in Figure 2.5, which
are:
• information
• system and service quality
• (intention to) use
• user satisfaction
48
2.3. RELATED WORK
Figure 2.4 – Technology – Organization – Environment Framework.
• net benefits
"A system can be evaluated in terms of information, system, and service quality; these
characteristics affect the subsequent use or intention to use and user satisfaction. The
net benefits will (positively or negatively) influence user satisfaction and the further use
of the information system" [152].
2.3 Related Work
Cloud computing represents a new field of research in the academic field. In information
systems, for example, 205 ’peer-reviewed’ academic articles were recorded in a review of
descriptive literature on ’cloud computing’ [153]. According to the same review on cloud
computing research, studies on the topic range into four types:
• "Technical issues: data management, cloud performance, data center man-

agement, software development, security, service management" [25]
49
2.3. RELATED WORK
Figure 2.5 – DeLone and McLean IS Success Model.
• "Business / Management issues: cost, pricing, legal issues, ethical issues,

trust, adoption" [25]
• "Conceptualizing cloud computing: foundational, introduction, predic-

tions" [25]
• "Domains and applications: E-government, E-science, education, open-

source, mobile computing, other domains" [25]
In the field of technical issues, the software development subcategory refers to a

stream of software developer-focused research. The topics covered in this area vary from
general discussions on producing distributed and parallel software in cloud computing
settings to specific articles on developing distributed and parallel software in cloud
computing environments [154], [155], [156], to specific analyses of particular cloud-
based programming frameworks such as MapReduce [157]. Novel studies also consider
component-based approaches for developing composite applications [158] and automation
in restructuring traditional applications into distributed/partitioned cloud-based ones
[159].
Indeed, cloud computing adoption is a complicated phenomena with a wide range
50
2.3. RELATED WORK
of potential and difficulties, but cloud computing research is still in its infancy. Much
of the present research focuses on the advantages and hazards of cloud computing, as
well as organizational case studies on cloud adoption and cloud computing designs [160].
More specifically, with respect to business decisions regarding cloud computing, it focuses
primarily on "identifying the determinants that influence the adoption of cloud
computing and the impact of a predefined set of factors on the environment." [25]
"Researchers studying cloud computing adoption have drawn on a variety

of ideas from the Management and Information Systems literature. The Technol-
ogy Acceptance Model (TAM), which [150] was used to offer an integrated picture
of the adoption element, is an example. The model has been criticized for its
narrowness and inability to provide a thorough explanation of adoption when
considering the utility and simplicity of use as factors of usage." [25]
"Another line of research used the DeLone and McLean IS Success Model
to assess the commercial benefits of cloud computing migration. Many attempts
have been made to enhance this model to boost its accuracy while investigating
the adoption of cloud computing [161] [152]. Another stream of research mobi-
lized the TOE (Technology-Organization-Environment) to expand the scope of
analysis by adding new dimensions of explanation [151]. Also, researchers ad-
dressed the topic of cloud computing adoption determinants through the diffu-
sion theory of technology, initially developed by Rogers 1995 aiming at identify-
ing attributes of technology diffusion." [25]
Low et al. [138] addressed the problems of adopting cloud computing using the
TOE framework and argued for better understanding of the adoption of cloud computing
and its environmental and organizational factors.
In fact, the issues of cloud computing adoption constitute, according to [162], one
aspect from four aspects that have dominated the literature on cloud computing, to know:
cloud computing characteristics, cloud computing adoption, governance mechanisms, and
business impact pointing out « a shift from purely technical aspects of research on cloud
computing to a broader understanding of cloud computing as a new IT delivery model»
[162]. We enroll in our work within this perspective.
51
2.3. RELATED WORK
Gap
"Despite the fact that the IS (Information Science) literature have wit-
nessed the development of decision support systems with the objective of guid-
ing managers when deciding about their future investments in cloud comput-
ing services such as [163] [164], [164], and others, we assess a clear lack of an
integrative tool that combine a large mix of heterogeneous and interdependent
decision dimensions. Indeed, the decision support tools have been developed ei-
ther based on a limited number of dimensions like the storage and computing
capacities added to the network [165] or regarding a specific service like [163]."
[25]
"Ours suggests a more integrative view of cloud services choice by organi-

zations. In fact, we propose to both handle a richer range of dimensions that de-
termine the decision and match them with a richer range of cloud computing so-
lutions that does not only take into account traditional services like IaaS, PaaS,
or SaaS, but also consider other solutions like virtualization or grid computing.
Consequently, we answer the needs of a larger sample of organizations by inte-
grating more criteria in the decision model we propose. As argued by Hevner et
al. [166], the design research branch of the research in the information systems
field aims at ‘extending the boundaries of human or organizational capabilities
by creating new and innovative artifacts’ [166]. Our study aims at fulfilling this
objective by proposing an analytical approach, both functional and technical,
to support managers make sense of the environment surrounding them in terms
of cloud computing technologies and capabilities offers. Beyond the commercial
approach, adopted by a variety of ‘Cloud Decision Engines’, we seek to provide
a more comprehensive method that includes 1) objective and critical determi-
nants of cloud computing decision that matches different organizational needs
and 2) rich propositions of cloud computing solutions that range from the clas-
sic solutions to others such as the virtualization, the dedicated servers, the grid
computing, and others." [25]
Indeed, our study fits into the ’Business /Management’ class of the typology pro-
posed above and more particularly in the subject ’Adoption of the Cloud’. Indeed, only
52
2.4. DESIGN CONSTRAINT
21% of the articles listed in the literature review deal with the subject of adoption. As
such, we propose to focus on the decision-making process that companies undertake to
invest in cloud computing and what variables do they consider in this decision-making?
We also propose a decision model (a rule-based modeling approach).
2.4 Design Constraint
The literature presented us with an ad hoc list of relevant factors, with the majority
of studies focusing on specific cloud computing services. We’ll do the following in this
chapter:
• Determine the factors that influence an organization’s choice to use cloud solutions.
• Create a uniform and well-defined classification system for the determinants listed
in the literature.
• Create a vendor-neutral, accurate DSS that meets the demands of businesses. This
tool should be able to meet the organization’s technological requirements.
Every determinant has some effect on an organization’s choice. Each organization’s

demand will be represented by a vector, with each dimension representing an influential
factor. The words determinant and dimension will be used interchangeably throughout
the rest of this chapter. The term degree will relate to the extent to which a factor has an
impact on each business. The phrases "organizational need" and "vector" will be used
interchangeably throughout this document.
Constraints "The design of the DSS tool is very complicated and faces many
constraints. The first constraint is extremely sparse data. To ensure precision
and accuracy, a large number of determinants should be used, each with a range
of degrees. As can be seen in the DSS tool, the number of determinants we identi-
fied is 10, each with approximately 5 degrees. An organization’s need can thus be
represented by a 10-dimension vector with 5 possible values for each dimension.
53
2.4. DESIGN CONSTRAINT
The studied universe (the number of different possible vectors) is 510 = 9, 765, 625.
It is impractical to manually study each possible vector and propose a suitable
technology mix. In addition, the number of manual entries individually studied
by our team is very small compared to the universe’s size. For this reason, the
DSS tool has to adapt to the scarcity of the available data." [25]
"The second constraint is the absence of decision principles. The literature

doesn’t propose well-defined rules linking a vector to a suitable cloud technol-
ogy. We personally believe that if such principles exist, they are emergent and
can’t be deduced without observing combinations of vectors and their suitable
technologies." [25]
"The third constraint is correlation. Since the data is extremely sparse,

it is very difficult to correlate between the organization’s needs and the already
available data. If an organization’s needs match one of the few vectors manually
evaluated, it is easy for the DSS tool to query the suitable technologies; otherwise,
a distance should be defined in order to match the organization’s need to the
closest available vectors and then the suitable technology mix can be deduced."
[25]
"The fourth constraint is distance. Euclidean distance is not suitable in

this case since it considers the magnitude of the distance without considering
the sense. Example: Consider a user who accepts a 3-star hotel, if 2 options
are available; the first is a 2-star hotel and the second is a 5-star hotel. The
Euclidean distance shows that the 2-star hotel is closer to the user’s preference.
Logarithmic distance is not suitable also since equal differences have different
impacts depending on the position. To face this constraint, we had to develop a
new distance as will be shown later." [25]
"The fifth constraint is the absence of well-defined priorities. Different

users can emphasize different dimensions. This prevents the possibility of hav-
ing a fixed distribution of weights /priorities for dimensions." [25]
54
2.5. DESIGN
2.5 Design
2.5.1 Determinants
"Based on literature review and our conducted survey, the determinants influ-
encing the adoption of cloud computing are:" [25]
• "Business Domain: This dimension refers to the affected area of business

activity. This dimension is of high importance to the decision process be-
cause it allows, based on previous content choices, to determine the future
orientations and strategy of the company concerning the management of
its cloud solutions. It forms the basis of other dimensions (such as techni-
cal choices (infrastructure), managerial choices (such as governance and
management), and strategic choices (choice of the cloud service provider)."
[25]
• "Price: The literature suggests that cost savings are manifested in two as-
pects: energy and cost savings" [25]
• "Data Management: Clients must carefully consider these legal require-

ments and how the CSA (Cloud Security Alliance) deals with issues such
as data flow when redundancy on multiple sites involves submitting data
to different jurisdictions at different times. The issue of competence takes
on more complexity when global compliance is taken into consideration
and more than one cloud computing provider is used. In these cases, the
client may have to coordinate negotiations between suppliers to ensure the
management of the necessary data. This includes: Data preservation and
redundancy, Data localization, Data entry, Data confidentiality, and Data
availability." [25]
• "Traffic: It refers to the degree of traffic predictability (general workloads

predictable or not in intensity, fluctuations, frequency, etc.)" [25]
55
2.5. DESIGN
• "Control: This dimension refers to the level of control the company wants
to have on the cloud computing solutions it adopts. It is true that once the
company is deprived of some of its control, by using the services of a cloud
provider, it requires different levels of control over their data, infrastruc-
ture, and processes." [25]
• "Lock-in: This dimension refers to the degree of lock-in or lock-up that the
cloud provider has on the company adopting its solutions. The lock-in is
used not only to choose the solution but the supplier also. The locking situ-
ations that companies experience are the result of the rising difficulty they
experience when it comes to changing cloud providers or replacing cloud
services." [25]
• "Support: Support is an interface provided by the cloud service provider to

address issues and queries raised by the cloud service client. It comprises
response time, support responsiveness, and resolution time." [25] "On the
other hand, 19 possible technologies are covered i.e. could be recommended
by our DSS. The technologies are:" [25]
– "Public Cloud (IaaS, PaaS, SaaS)" [25]
– "Private Cloud (IaaS, PaaS, SaaS)" [25]
– "Managed Private Cloud (IaaS, PaaS, SaaS)" [25]
– "Hybrid Cloud (IaaS, PaaS, SaaS)" [25]
– "Community Cloud (IaaS, PaaS, SaaS)" [25]
– "Dedicated servers / Multi cloud / Virtualization / Grid computing and

Legacy IT" [25]
"The proposed DSS is flexible (as will be described in the coming section)
and can handle other technologies if needed." [25]
56
2.5. DESIGN
2.5.2 DSS objectives
"The DSS tool is an integral part of this work and its final product. Its rec-
ommendation accuracy is highly required since it supports decision-makers in
setting the organization’s strategy. In this section, we describe the design of our
DSS tool, its objectives, and the constraints it faces. The foundational objectives
of the DSS tool are:" [25]
• "Simple user interface: The used dimensions and degrees should be generic
and well formulated. This objective is aligned with the dialog generation
and management system’s requirement of intuitive and easy-to-use inter-
faces [167]." [25]
• "Precision and accuracy: For this objective to be achieved, a sufficient num-

ber of dimensions and a sufficient number of degrees per dimension are re-
quired. This objective has a severe consequence on the constraints as will
be discussed later in this section." [25]
• "Flexibility: The selected dimensions and degrees are not exclusive, thus
future needs and technology innovations might enforce changes. The DSS
tool should be flexible to adopt any change in dimensions and their degrees
without the need for manual intervention." [25]
• "Auto-learning and self-correcting: Any decision model used to automati-

cally propose a technology mix may need fine-tuning and calibration to en-
hance its performance. Auto-learning makes the DSS tool in a continuous
process to optimize its performance." [25]
2.5.3 Components
To meet the constraints and design objectives shown above, we designed a Non-
Personalized Collaborative Filtering Recommender System that receives the user’s re-
57
2.5. DESIGN
quirements and helps to match with the suitable cloud mix. As will be shown in Chapter
3, market factors (external to the organization) will affect the future of cloud adoption
and its success. For this reason, it is impossible to deduce the cloud adoption decision
objectively solely from the organization’s internal factors. For this reason, we decided to
build an RS DSS to help benefit from the subjective influence of the decision-makers and
the market trend.
"The designed DSS tool (Recommender System) is divided into 3 compo-

nents which are the Web component, the backend server, and the database. As
shown in Figure 2.6, the user (decision maker) interacts with the Web component
where he responds to multiple questions. The answers represent the organiza-
tion’s needs and are filled into a vector (each question targets a 1 determinant
and each answer represents its degree). The vector is submitted to the backend
server which in turn retrieves all the stored data from the database server. The
user’s vector and all the retrieved data are processed, and 3 technology mixes
are recommended each with an accuracy level. These 3 technologies are recom-
mended to the user through the Web component. The user will be then asked to
select the technologies he will implement. The user’s selected technologies are
stored in addition to his answers on the database server. The stored data will
help the system increase its accuracy while serving future users." [25]
2.5.4 Design
The flowchart of the “Processing operation” presented in Figure 2.6 is shown in Figure
2.7.
2.5.4.1 Step 1
"As presented in Section 2.4, a distance should be able to calculate the correla-
tion between the user’s answer vector (organization’s need) and any vector stored
in the database. As also presented in Section 2.4, the existing distances do not
58
2.5. DESIGN
Figure 2.6 – The proposed DSS’ message flow diagram.
satisfy our requirements, so, we had to develop a distance formula that satisfies
our needs as shown next." [25]
"The backend server calculates the distance between the user’s vector and
every row in the stored data (row 0 in Figure 2.6 is degree 1, degree 3, . . . , IaaS).
The distance expression we developed is as follows:" [25]
M
X −1
Distance[k] = β[j] × α × (A[j] − S[k, j])
j=0
"Where β[j] is the priority of metric j (in the current version of the DSS tool,
all priorities are set to 1). α is a small positive value when S[k, j] ≥ A[j] (it is set
to 1 in this version of the DSS tool) and a large positive value otherwise (it is
set to 2 in this version of the DSS tool). α is used to represent the higher impact
of a downgrade in comparison with an upgrade. A[j] is the user’s answer for
59
2.5. DESIGN
Figure 2.7 – Flowchart of the processing operation i.e. to select the closest 3 technologies.
dimension j while S[k, j] is the stored value for row k and for dimension j. M is
the number of used metrics." [25]
60
2.5. DESIGN
2.5.4.2 Step 2
"The backend server selects the technology mixes having the smallest distance
(closest to the user’s answer)." [25]
2.5.4.3 Step 3
"The server then calculates the level of confidence in each recommendation. The
equation calculating the level of confidence is: " [25]
3
X Short_List[k, i]
Level of conf idence[k] = W [i] ×
i=1 (accuracy[i]
"Where short_List[k, i] is the number of rows having ith smallest distance

and select technology mix k. accuracy[i] is the number of rows having ith smallest
distance. W [i] has the following values:" [25]
• W [1] = 1 − (weighted_min[1])a ccuracy[0].
• W [2] = weightedm in[1]accuracy[0] × weightedm in[2].
• W [3] = weightedm in[1]accuracy[0] × weightedm in[3].
"Where weighted_min[1] ≤ 0.66 and weighted_min[2] + weighted_min[3] = 1.

weighted_min[1] is set to 0.5 in this version of the DSS tool, while weighted_min[2]
is set to 0.7 and weighted_min[3] to 0.3." [25]
2.5.5 Auto-learning and Self-correcting
"This DSS has 2 levels of auto-learning and self-correcting. The first level is
intrinsic to all RS since the measurement of the new distance and level of con-
fidence will be affected by every new entry to the database. Users’ answers will
61
2.5. DESIGN
normalize the effect of inaccurate entries and the system will have better per-
formance with more accurate entries in the database. The second level of auto-
learning and self-correcting is design extrinsic. The DSS accuracy depends on
its configuration (the values of β[j], α and weighted_min[i]). The tool runs a peri-
odic algorithm to fine-tune its configuration." [25]
2.5.6 Fine-tuning algorithm
The fine-tuning algorithm will try for all possible combinations of (β[j], α and
weighted_min[i]) to predict the technology mix for every input vector. After that, the
algorithm will compare, for every parameter combination, the predicted technology mix
to the selected one. The difference between the predicted and selected technologies is
calculated using RMSE. The parameter combination that scores the smallest RMSE value
will be used by the RS DSS, since it increases the system’s predictive accuracy the most.
The fine-tuning algorithm is presented in the flowchart illustrated in Figure 2.8.
We selected RMSE as an accuracy metric since the DSS will offer one technology
mix for every user, thus rank accuracy and classification accuracy metrics become useless.
We can deduce that predictive accuracy metrics are the only ones suitable. RMSE’s
advantage over MAE and MAUE is its lower tolerance to large deviations between the
predicted and selected mixes.
2.5.7 Data quality
The collaborative filtering recommender system model, no matter what type of field it
is used on, be it e-commerce or education, is not only affected by its parameter and the
tuning methods followed to best fit the training data. To achieve the best accuracy results
from evaluation metrics, it is necessary to analyze and study the quality and the quantity
of the data itself. As shown in [168], working with the data is most of the time risky. Even
with the most efficient algorithms, the output might end up biased, un-expected, and
62
2.5. DESIGN
Figure 2.8 – Fine-tuning algorithm flowchart.
fundamentally flawed. A more interesting work [169] [170] demonstrates how the content
in the datasets can be managed to shape the recommender’s output and immensely affect
hundreds of users of a certain system. It is valid that randomness and noise in data can
not only affect the accuracy results of a model [171], [172] but ultimately affect the quality
of recommendation presented to a certain user, however, this shows that the consequences
63
2.5. DESIGN
of engineering certain data patterns in a recommender dataset can present markedly

more considerable risks on the system’s users.
This leads to two types of problems that are faced when implementing recommender
systems:
• Small datasets (the quantity of the data)
• Noise in the data (the quality of the data such as natural and malicious noise,
obfuscation)
The issue with small datasets when employing recommender models such as
collaborative filtering (CF) approaches is the cold-start problem that it encounters with
data that is inevitably sparse especially in new online systems. That content is just not
dense enough to fill the huge user-item matrix. During the cold-start phase, the knowledge-
based system produces more accurate results, but the collaborative system produces more
accurate results later on. For dealing with the cold-start problem, including knowledge-
based systems is often preferred [173]. Other solutions such as active learning [174] can
also be efficient when addressing such a problem faced by CF recommenders, however,
the user will eventually need to accurately provide insight on the queries presented to
him when he initially connects to the system.
Problems that result from noise in the systems can be tackled by manipulating
data pre-processing methods for noise management before loading the recommender with
the dataset and training it. Various methods exist depending on the type of noise present
in the data, be it malicious [175], natural [168], or obfuscation [170]. Unfortunately, there
is yet to be a unified noise system that fundamentally deals with the noise in a dataset
irrespective of the recommender used, therefore, the data needs to be manually analyzed
and it is more adequate to run tests that cater to all noise types before running tests on
an algorithm. In a CF environment, some easy and efficient noise management solutions
can be adopted to deal with the problem using pre-processing methods such as in [176],
however, for other models that don’t solely consider ratings such as knowledge-based,
other methods need to be considered to overcome this issue.
64
2.6. COLLECTED DATA
2.6 Collected Data
The tool has been publicly available for a period of one year, and we were able to collect
278 responses, mostly from the Middle East and Lebanon specifically. The respondents
have their profiles distribution as shown in Figure 2.9.
Figure 2.9 – Distribution of respondents by position.
As can be seen in Figure 2.9, all the respondents are decision-makers or experts,
and their opinions are considered authoritative. The respondents represent industries
across different market sectors. These market sectors are distributed as shown in Figure
2.10.
As seen in Figure 2.10, a large share of the respondents belongs to the financial
sector, whose members are usually reluctant to adopt a new technology unless proven
reliable. We expect to see in the selected technologies a considerable portion of private
and hybrid technologies selected. The participants represent companies of different sizes.
The distribution of respondents by company size is shown in Figure 2.11.
As seen in Figure 2.11, a large share of the respondents belongs to small companies.
Small non-technology companies usually find variations of SaaS very convenient. The
respondents represent companies of different ages. The distribution of respondents by
company age is shown in Figure 2.12.
65
2.6. COLLECTED DATA
Figure 2.10 – Distribution of respondents by market sector.
Figure 2.11 – Distribution of respondents by company size.
As seen in Figure 2.12, a large share of the respondents belongs to new companies.
Previous expenditures on technology and the quality of a company’s system influence the
adoption decision. The distribution of selected technologies is shown in Figure 2.13.
As seen in Figure 2.13, legacy technologies such as “Dedicated servers” and “Grid
Computing” were selected by 14% of the respondents. Public cloud technologies sum up
66
2.6. COLLECTED DATA
Figure 2.12 – Distribution of respondents by company age.
to 31%, while private and managed cloud technologies sum up to 60% of the respondents’
decisions. The total percentages are different than 100% since we are measuring the
percentage of respondents who selected a specific technology, while the respondent has
the freedom to select any number of technologies simultaneously (Technology Mix).
Figure 2.13 – Technology mixes selected by respondents.
67
2.7. EVALUATION
2.7 Evaluation
The accuracy of the tool’s recommendations, over the process of data collection, has been
measured using MAE and the results are shown in figure ??. The y-axis shows MAE
while the x-axis shows the number of populated answers in the tool.
Figure 2.14 – Evaluation of the recommendation’s accuracy Using Predictive Accuracy Metrics -
MAE
As can be seen in figure 2.14, the system is following a gradient decent towards
low MAE values which means that the tool is achieving good accuracy.
2.8 Summary
This work is our response to the need for a holistic, generic, and non-trivial DSS that
helps decision-makers in their strategy of cloud computing adoption. We tried to do so by:
integrating a richer mix of cloud computing adoption that goes beyond the traditional
price/capacity mix and proposing a richer offer of cloud computing services and similar
solutions; and finally, by targeting a larger sample of users and organizations.
68
2.8. SUMMARY
The designed DSS satisfies CBOD’s task 4. The updated project status is shown in
table 2.1.
Table 2.1 – Chapter 2 contributions to CBOD objectives and tasks
Status Status
Objective 4 Pending Task 4 Done
The designed DSS is based on RS since, as will be shown in the coming chapter,
market factors affect the success of cloud computing in a studied market.
Appendix - A: Active Learning Questions
The questions asked to the user and their respective possible answers are:
• General Questions (Business & Strategy)
1. Does your enterprise have a business strategic vision for the next 2-3 years?
a. Do not know
b. No vision in place or being developed
c. Partially – we may have an initial vision but it is incomplete, not well
formulated or communicated
d. Yes, we have well formulated strategic vision but haven’t taken significant
actions to communicate or implement it yet, we may not have time, resource,
commitment or knowledge to implement it
e. Yes and it is being implemented, it is the foundation of our business direction
and priority-setting
2. Does your organization have a strategy for using cloud technology?
a. Do not know
69
2.8. SUMMARY
b. NO - No strategy, cloud computing is a topic of discussion with the business

and/or within IT but we have yet developed our point of view. No plans to
develop additional strategy
c. Partially - We have an informal approach to cloud, may be doing limited

pilots to get knowledge, we may have a strategy but no written plans or
roadmap at any level. We expect to make improvements in next 1-2 years
d. Yes - We have IT centric cloud strategies, plans, and roadmaps that is

shared across the IT organization
e. Totally – we have strategy, plan and roadmap that is enterprise-wide and

includes It and business.
3. How will your organization’s investments in cloud computing change in the

coming 1-2 years?
a. Do not know
b. Plan to decrease
c. Would be the same
d. Increase
e. Significantly increase
4. What cloud related product or service is of greatest interest to your organiza-

tion?
a. Social business / media
b. Mobile computing
c. Analytics / big data
d. IT service Management
e. DevOps
f. Software, infrastructure or platform as a service
g. Integration including service broker
• Metric Questions
5. How do you characterize your organization’s level of data traffic?
a. Unpredictable
70
2.8. SUMMARY
b. Considerably predictable
c. Predictable
d. Very predictable
6. How do you characterize the data used in cloud operations?
a. Non-sensitive
b. Considerably sensitive
c. Sensitive
d. Very sensitive
7. How are the organization’s operations in which cloud computing would be used?
a. Public facing
b. Critical
c. Mission – critical
d. Specific
e. Combination of above
8. What are your organization’s security requirements?
a. Non-maintained
b. Average
c. Maintained
9. How are your organization’s objectives in terms of cloud performance?
a. Low
b. Maintain existing levels
c. Average cloud services performance
d. Above average
e. High
10. What are your organization’s control requirements over the cloud service?
a. None
b. Maintain the existing control over data and operations
c. High
11. What are your organization’s requirements in terms of legal compliance?
71
2.8. SUMMARY
a. None - accept the provider’s SLA

b. Yes - negotiate the agreements
12. What are your organization’s acceptable levels of lock-in?
a. None
b. Moderate lock-in
c. Complete lock-in
13. Do you think you need a sophisticated IT support to manage cloud solutions?
a. Extremely sophisticated needs

b. Sophisticated needs
c. Average needs
d. No needs
14. How do you characterize the degree technical compliance (between the cloud
technology and your legacy technology)?
a. No compliance problems
b. Some compliance problems
c. Many compliance problems
Appendix - B: Accuracy metrics for recommender

systems
The performance of any RS is measured by one or more metrics depending on the de-
signer’s intentions. "The accuracy metrics are divided into three major classes
[177] [178]:" [25]
• "Predictive accuracy metrics: they measure how close the ratings estimated
by a recommender system are to the true user ratings. They are usually
used to evaluate non-binary ratings. The most popular metrics in this cat-
egory are Mean Absolute Error (MAE), Mean Squared Error (MSE), Root
72
2.8. SUMMARY
Mean Squared Error (RMSE), Normalized Mean Absolute Error (NMAE),

and Mean Absolute User Error (MAUE) [179], [180], as shown in Table 2.2."
[181]
• "Classification accuracy metrics: they ignore the exact rating or ranking
73
2.8. SUMMARY
of items, and measure only the correct or incorrect classification. The most
used metrics in this category are precision and recall, in addition to fall-
out, miss rate, inverse precision, inverse recall, F1-measure, Mean Average
Precision (MAP), Geometric Mean Average Precision (GMAP), Harmonic
Mean Average Precision (HMAP) and Quadratic Mean Average Precision
(QMAP). For more information about these metrics, the reader can refer to
[182]." [181]
• "Rank accuracy metrics: they measure how much the recommender system
is able to estimate the correct order of items with respect to the user’s pref-
erences. These metrics are useful when we care about giving the user the
recommendations in a certain ranking. They use relative ordering regard-
less of the exact values. Therefore, even if the recommender system fails to
predict the ratings accurately, if it gives the items in the correct order of
preference then these metrics will still achieve a great score. Some used
metrics from this category are Kendall’s τ , Spearman’s ρ, DCG, and nDCG
[180]." [181]
Table 2.3 – Confusion Matrix Used in Recommender Systems
Relevant Irrelevant
Recommended tp fp
Not Recommended fn tn
"A major problem with the frequently used metrics presented in Table 2.4
is that they can include great biases. This issue can be solved by three new met-
rics" [181] introduced by Powers [183]:
M arkedness = P recision + InverseP recision − 1 (2.1)
Inf ormedness = Recall + InverseRecall − 1 (2.2)
74
2.8. SUMMARY
Table 2.2 – Details of the Predictive Accuracy Metrics
Metric Definition Formula Comments
"Prediction error of
a heavy rater is very
small compared to that
"It computes the of a cold start user.
deviation be- However, MAE weighs
Mean Absolute tween predicted MP AE = every prediction error
1 n
Error (MAE) ratings and n i=1 |p_i − r_i| in the same way. But in
actual ratings." reality, the heavy rater
[181] is probably satisfied
with the prediction
error while cold start
users are not." [181]
"This method solves

"It computes the any possible bias intro-
MAE for every duced in the calcula-
Mean Absolute E=
single user inde- M AU tion of the MAE. In this
User Error 1 P
pendently, then |U | u∈U M AE(u) way, all the users will
(MAUE) [65]
averages all the have the same weight
MAEs." [181] in the computation."
[181]
"As MAE, it also

computes the de- "Similar to MAE, but
Root Mean SE =
viation between RM it places more empha-
Square Error
q P
1 n
i=1 (p_i − r_i)
2
predicted rat- sis on the larger devia-
(RMSE) n
ings and actual tion." [181]
ratings." [181]
q
M atthewsCorrelation = ± M arkedness.Inf ormedness (2.3)
75
2.8. SUMMARY
Table 2.4 – Details of the Classification Accuracy Metrics Used in this Study
Metric Definition Formula Comments
"Prediction error of
a heavy rater is very
small compared to that
"It computes the of a cold start user.
deviation be- However, MAE weighs
tp
tween predicted P recision = tp+f p every prediction error
Precision
ratings and = gooditemsrecommended
allrecommendations
in the same way. But in
actual ratings." reality, the heavy rater
[181] is probably satisfied
with the prediction
error while cold start
users are not." [181]
"This method solves

"It computes the any possible bias intro-
MAE for every duced in the calcula-
tp
single user inde- Recall = tp+f n tion of the MAE. In this
Recall
pendently, then = gooditemsrecommended
allpref erences
way, all the users will
averages all the have the same weight
MAEs." [181] in the computation."
[181]
"As MAE, it also

tn
Inverse viation between InverseP recision = tn+f n it places more empha-
Precision predicted rat- = baditemsnotrecommended
allnotrecommendeditems
sis on the larger devia-
ratings." [181]
"As MAE, it also

tn
Inverse viation between InverseRecall = tn+f p it places more empha-
Recall predicted rat- = baditemsnotrecommended
allbaditems
ratings." [181]
"As MAE, it also

F1- viation between
F1 = 2 × P recision×Recall it places more empha-
measure predicted rat- P recision+Recall
ratings." [181]
76
Chapter 3
The Network Economics of Cloud

Computing
Contents
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.2.1 Traditional EPC vs. vEPC . . . . . . . . . . . . . . . . . . . . . . . 80
3.2.2 EPCaaS and Multi-tenancy . . . . . . . . . . . . . . . . . . . . . . . 82
3.2.3 Pricing scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.3 Proposed Approach and Evaluation . . . . . . . . . . . . . . . . . . . 93
3.3.1 Expected scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.3.2 Proposed approach and metrics . . . . . . . . . . . . . . . . . . . . 94
3.3.3 Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.3.4 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.4 Conclusion and Future Challenges . . . . . . . . . . . . . . . . . . . . 103
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
77
3.1. INTRODUCTION
3.1 Introduction
Cloud Computing vendors and researchers promise competitive advantage and financial
savings for potential clients. When we consider individual businesses, migrating to the
cloud can be financially feasible and the most rewarding option. When we consider the
ecosystem as a whole, the rewards of cloud computing can become blurry. The literature
considers the individual business’ requirements to decide on cloud adoption without
having a vision into the market’s future. Mobile operators’ decision-makers experience
the same limitation when considering Multitenancy cloud computing for their networks.
Mobile operators are busy planning for the coming 5G era which supports a wide
range of potential applications. A new ecosystem of content provides is growing around
this communication infrastructure and some of the main concerns is how will the mobile
operators slice their spectrum to offer dedicated services to these content providers. 5G
is going to change how business is performed in many industries; according to the Next
Generation Mobile Networks (NGMN) alliance, “5G is an end-to-end ecosystem to enable
a fully mobile and connected society. It empowers value creation towards customers and
partners, through existing and emerging use cases, delivered with a consistent experience,
and enabled by sustainable business models” [184]. It is really a disruptive technology
that has the potential to change even the mobile operator’s business model.
Yousaf et al. claimed that Virtualized Network Functions (VNF), Evolved Packet
Core as a Service (EPCaaS), and Multi-tenancy decrease the total cost of ownership and
increase a mobile operator’s competitiveness [185]. In reality, such a claim is true when
part of the mobile operators adopts these new technologies that give them an advantage
over other operators; but when all the mobile operators migrate to multi-tenancy cloud-
based core services, the claimed advantages are not necessarily maintained. In addition
to that, the promised reduced Capital Expenditures (CAPEX) can considerably reduce the
startup-cost which is an important economic barrier-to-entry that works for the advantage
of already running mobile operators and hinder other players from joining. The Pay-as-
you-go feature of cloud computing also reduces the effect of the economy of scale which is
another important economic barrier-to-entry that also works for the advantage of already
78
3.1. INTRODUCTION
running mobile operators. EPCaaS and other forms of Cloud-based mobile operators can
be invasive to the current stability in the market and current “business may not be able
to recover costs and effectively compete” [186].
Various programmable ABM environments exist in the literature. ABM can also
be performed using custom-made object-oriented programs. In this chapter, Netlogo
[187] will be used as the programmable modeling environment for simulating the cloud
computing market behavior.
Research framework
Market prediction is left for a dedicated few who are experts in their market.
Nevertheless, history is full of experts who failed to successfully predict market behavior,
especially in the technology sector. Using ABM for cloud market prediction is courageous
and risky at the same time. For this reason, we decided to consider a specific application
of cloud computing as a reference.
We are interested in evaluating whether multi-tenancy cloud-based core services

(also called carrier cloud) are viable and economically feasible. We are also interested in
studying its effect on major players especially that this technology frees the market from
oligopoly and small players from the control of dominant players. Although virtual mobile
operators and infrastructure sharing is currently implemented, the business model in
place makes virtual mobile operators dependent on dominant operators who own the
infrastructure and set the prices. In this chapter, we will study the market behavior using
Agent-Based Modeling as a carrier cloud emerges, thus creating a model that predicts
the effect of this new technology on prices, customers, and telecom operators. To the best
of our knowledge, this work is the first to study and predict the mobile network’s market
behavior under the effect of multi-tenancy cloud-based core services.
Existing literature barely touches on the mobile market behavior as EPCaaS and
multitenancy get implemented. As this technology emerges, the current market stability
may diverge into a purely competitive market or converge into a monopolistic market. The
originality of this work can be identified as:
• This work is the first, to the best of our knowledge, to raise the issue of market structure
79
3.2. LITERATURE REVIEW
uncertainty due to the introduction of EPCaaS/multitenancy as part of 5G.
• Existing predictions used in the mobile market are based on historical data. Such
methods are notoriously inaccurate in predicting the implications of emergent tech-
nologies (such as the introduction of EPCaaS in 5G). This work is the first, to the
best of our knowledge, to simulate the market and predict its behavior based on the
interaction between different players, not by interpolating previous behaviors.
• The mobile market differs between countries, thus country-specific studies have limited
impact and may not be valid for different countries. This study is designed to be
country-independent, where the user can configure the simulator (we developed) to
match the studied market’s parameters.
• This study can be used by decision-makers to find the right parameters for their market’s
case to ensure a suitable outcome.
3.2 Literature Review
In this section, we start by presenting the related technologies behind EPCaaS such as
EPC, vEPC, and NFV. We then discuss cloud computing’s business model and pricing
schemes.
3.2.1 Traditional EPC vs. vEPC
Efforts to transform the traditional EPC into a set of virtualized modules started since 2011
[188]. Sama et al. [189] proposed cloud-based network sharing between mobile operators
for the LTE/EPC architecture. This proposal didn’t introduce a new business model since
existing mobile operators were advised to co-invest in a shared cloud network that offers
services to the partnering operators, thus, using the cloud’s community deployment model.
On the contrary, this paper emphasizes the cooperation between mobile operators, thus
inherently causing the competition to stabilize.
80
The term carrier cloud started to be present in research papers such as in [190],
[191], [192], [193], and other papers, and this announced the start of an interest in
cloud-based core services. Similarly, NFV (Network Function Virtualization) emerged
as a facilitator for dynamic mobile network [194] management, carrier cloud, and multi-
tenancy. NFV is also expected to "change the core structure of the telecommunications
infrastructure to be more cost-efficient" [195].
Bagaa et al. [196] designed an algorithm for calculating the most suitable number
and locations of vEPC’s virtual instances over the federated cloud which is suitable
for carrier clouds and EPCaaS. This placement algorithm ensures QoS and maximizes
the profits of cloud operators. Resource pooling has been proposed by Rodriguez et al.
[130] to enhance the performance of vEPCs running in resource-limited general-purpose
datacenters.
Khan [197] proposed to use NFV over SDN (Software Defined Networks) and backed
his proposal by predicting OPEX reductions. These reductions include, among others,
eliminating the takedown and provisioning of network upgrades and eliminating physical
boxes and cable migration. Sama et al. [198] discussed the realization of an operational
VNF/SDN stack and control architecture. Valtulina et al. [199] studied the Mobility
Management in vEPC and proposed a Distributed Mobility Management under VNF/SDN
environment.
Sousa et al. [200] proved the technical feasibility of migrating existing mobile
networks to a fully cloudified environment. They also proposed an architecture that
includes a business lifecycle optimizing CAPEX/OPEX and a technical lifecycle leveraging
a deployment model virtualizing MNO’s core network and RAN elements. The authors of
[201] discussed vEPC from vendors’ point of view, listed EPC market revenue projections,
and compared vEPC suppliers. This paper is one of the rare publications looking at the
mobile network industry from an economic perspective.
81
3.2.2 EPCaaS and Multi-tenancy
Cloudified EPCs received attention from the research community and some unique
services have been proposed which makes vEPC own a competitive advantage against
legacy deployments. Zhao et al. [202] proposed MOBaaS (Mobility and Bandwidth
prediction as a service), uniquely for cloudified EPCs, which provides location predictions
of a single/group user equipment (UEs) in the future moment. This information can be
used for self-adaptation procedures and optimal network function configuration during
run-time operations.
Samdanis et al. [203] envisioned future multi-tenant systems to include Over-

The-Top (OTT) and Vertical industries. The first is simply a Mobile Virtual Network
Operator (MVNO), where the leasing party is not a mobile operator. The second focuses
on providing complementary services to the telecommunication industry. They also
proposed multi-tenancy at the radio level using their 5G Network Slice Broker, which was
defined by 3GPP’s Services Working Group SA1. A required attribute to allow the 5G
Network Slice Broker to work, among other attributes, is Dedicated Core Network (DCN)
where each DCN is designed to serve a different type of user. DCNs are important in
facilitating specialized carrier clouds.
Taleb et al. [204] demonstrated the feasibility of the on-demand creation of cloud-
based elastic mobile core networks. They also described two implementation scenarios for
EPCaaS. The first scenario uses full virtualization where both user plane and control
plane functional entities are implemented in VMs. The second scenario uses partial
virtualization where only control plane functional entities are used in VMs.
Li et al. [205] proposed Caching as a Service as a new feature for cloud-based

5G networks that matches and exceeds the already proposed front-haul and back-haul
caching techniques. The virtualization of Caching as a Service has been proposed in [206]
by the same authors. Additional EPCaaS features are proposed such as resiliency [207].
Jeon et al. [208] proposed a traffic offloading framework between legacy EPC and
vEPC. This work is based on the fact that existing design and deployment initiatives for 5G
82
do not consider that their adoption by operators will need to pass through a co-existence
period with the existing legacy network.
Cau et al. [209] proposed novel solutions for effective subscribers’ state management
in quality-constrained 5G scenarios which will help ensure efficient elasticity provisioning
of EPCaaS stateful components. This work leverages SDN and NFV in addition to
proposed protocols and network nodes to move network control functions close to eNB
(Mobile Edge Computing) which results in robust network decisions. Solozaba et al.
[210] benefitted from the previous work to propose Non-Standalone 5G ETSI MEC-based
architecture for Mission-Critical Push-to-Talk. This work is among many (such as [211],
[212], and [213]) that leverage EPCaaS functionalities at the edge level to offer new
services.
3.2.2.1 Business models
EPCaaS business model hasn’t been touched before in the literature and this work is
considered the first to look at EPCaaS from a business/economic perspective. In this
section, we survey the cloud computing business model as discussed in the literature.
Chang et al. [214] classified the business models under eight dimensions which are:
1- Entertainment and Social Networking. 2- Venture Capitals. 3- In-House Private Clouds.
4- All-In-One Enterprise Cloud. 5- One-Stop Resources and Services. 6- Government
funding. 7- Support and Services Contracts. 8- Service Provider and Service Orientation.
Weinhardt et al. [26] described CBMF (The Cloud Business Model Framework) which
is divided into three main business models which are: 1- Infrastructure as a Service. 2-
Platform as a Service. 3- Software as a Service. They also listed the cloud offerings and
matched them to the business models in CBMF.
Martson et al. [215] surveyed and identified various issues that will affect the
different stakeholders of cloud computing. They issued a set of recommendations for the
practitioners who will provide and manage this technology.
The authors of [216] and [217] studied the effect of cloud computing on the value
chain and listed some of cloud computing’s success factors which include simplicity of
83
service deployment, references, flexible price models, integration and migration interfaces,
and architecture frameworks that support automatic scaling and economies of scale.
3.2.2.2 Cost analysis
Yousaf et al. [185] studied the cost analysis of EPCaaS’ VNF (Virtualized Network
Functions) deployment. The study evaluated cost as a function of CPUs which can be
translated to dollars based on the current pricing of a VM (Virtual Machine).
3.2.3 Pricing scheme
In this section, we will be identifying the pricing schemes, found in literature, of each of
the agents participating in the studied market. It is worth noting that a generic currency
is used to represent pricing as recommended by [218] who defined it as dimensionless
and [77] who defined it as average aggregate quantities.
3.2.3.1 Cloud Service Provider (CSP)
Each CSP (Cloud Service Provider) enhances its competitiveness through value-added
services [26], but pricing remains a decisive factor in offer selection [219]. All available
business models share the same possible pricing models which can be categorized into
two groups:
• Static: The user (the cloud-based mobile operator in our case) has either a fixed resource
cost or fixed lump sum fee. The static pricing models have dominance over dynamic
ones due to accounting and psychological reasons even if it costs more [26], [220].
Static pricing models are:
• Pay-per-use: Resource units are priced at a fixed value. This pricing model is
used for “products whose mass production and widespread delivery have made
price negotiation impractical” [26], [221].
84
• Subscription: The client subscribes for a pre-selected collection of resources

over a fixed period of time for a fixed price [26].
• Dynamic: The user pays a dynamic cost, specified based on supply and demand, for
the used resources. Such models achieve differentiated services and better prices
for low-end services [26], [222]. Dynamic pricing models are:
• Auctions/Bidding: Prices are proposed by clients and the CSP selects the price
that optimizes his revenues.
• Revised pricing: Resource units are priced a new value for each new unit of
time. Existing pricing models found in the literature are:
– RPM1: This RPM is based on the basic principles of economics. In This

RPM the price is specified as follows: Pnew = Pold + ∆system capacity × C where
C is a constant value representing the capacity change to price change ratio
(inverse of price elasticity of demand).
– RPM2 for M/M/1 Queuing system [223]: P = C

(µ−λ)λ
; where C is a constant,
P is the offered price at certain instance, λ is the rate of job arrival (tasks
/sec) at the same instance (following process distribution) and µ is the
average service rate (tasks /sec). The problem in this pricing model is its
severe responsiveness to fluctuating traffic.
– RPM3 [224]: The authors of [224] proposed a pricing model based on the
Black-Scholes-Merton Model and Moore’s law.
3.2.3.2 Cloud-based mobile operator
The literature has limited proposals for pricing models specially designed for virtual or
cloud-based mobile operators. Meidanis et al. [225] studied the pricing scheme of a Mobile
Virtual Network Operator who rents the infrastructure from a Legacy Mobile Operator.
This work uses user profiles from a crowdsourcing platform to maximize revenues for
both operators. Although this study is not identical to our environment (EPCaaS), we
can deduce the pricing correlation between these two types of operators. The correlation
can be simplified as:
85
PM V N O (Ui ) = C(di ) × PLegacy (Ui )
Where Ui is user i, PM V N O (Ui ) is the selling price for user i by the virtual mobile
operator and di is the traffic demand by user i. C(di ) is a linear function and Plegacy (Ui )
is the price charged by the legacy mobile operator against the virtual mobile operator.
The authors have identified a linear function between both pricing schemes even though
Plegacy (Ui ) is a function of di . Thus, PM V N O (Ui ) is a function of the square of di which
makes high traffic users pay extremely high fares.
Ma et al. [226] proposed a dynamic pricing scheme as a function of time and

location. The price increases in hotspots and at peak hours. Such a pricing scheme
will demotivate low-income users from using the network during high-fare periods and
incentives them to communicate during low-traffic periods.
Xiao et al. [227] proposed a non-uniform pricing scheme that tries to motivate users
to utilize the spectrum resources more predictably while ensuring high QoS (Quality of
Service). Predictable user behavior allows the mobile operator to offer consistent service
to its virtual mobile operators.
Zhang et al. [228] proposed a pricing framework for MCA (Mobile Crowdsourced
Access), where the direct traffic and tethering traffic are charged independently according
to a data price and a tethering price, respectively. The literature didn’t touch the pricing
strategies for cloud-based mobile operators specifically, thus the general pricing strategies,
discussed earlier, are also implementable.
3.2.3.3 Legacy mobile operator
Wang et al. [229] listed green pricing schemes for mobile operators and identified the
absence of real green pricing schemes and the existence of a few related research pro-
posals. The listed schemes are [230] and [231]. In the first [230], the authors proposed
a compensation-based pricing model by studying the game where the MVNO invests
in content /advertising to compensate for the quality of service degradation. It is done
by modeling the system as a supply-chain. In the second [231], the authors proposed a
86
pricing mechanism to mediate between the user-centric and network-centric resource

management problems.
Sen et al. [232] surveyed the data pricing strategies and divided them into two cat-
egories, static and dynamic. The static pricing strategies include: Fate-rate, Usage-based,
QoS classes, Negotiated contracts, Application-based, and Time-of-day. The dynamic
pricing strategies include: Raffle-based, Real-time congestion, Auction-based, and Day-
ahead. Although these pricing strategies are ISP oriented but can be extended to legacy
CSPs. Kim et al. [233] studied the own- and cross-price elasticities of SMS and voice
services. They found that voice services and SMS are substitutes with 0.8% increase
in demand on SMS for 10% increase in the price of voice services. Vincenzi et al. [234]
studied the possibility of cooperation among different MNOs that provide service to the
same area by sharing their C-RAN (Cloud Radio Access Network). The authors set rules,
among which is a pricing scheme, for ensuring profitable collaboration. Nicoletta and
Zirulia [235] studied the changes in tariff plans in small and large operators over time
and competition. Tariff plans are not necessarily related to pricing, but rather market
segmentation through service differentiation. Ding et al. [236] studied the effectiveness of
time-dependent pricing mechanisms for large mobile operators and concluded that adding
location factors can enhance its performance. Asghari and yousefi [237] studied monopo-
listic and oligopolistic markets of VMVoIP in which a single VMVoIP and several VMVoIP
operators available in the market respectively using game theory. Koski and Kretschmer
[238] studied the competition between 2G mobile operators across 32 different countries.
They also studied the competition between technologies (substitute technologies) and
their effect on pricing. It was deduced that as different technologies exist, the prices
decrease dramatically. Papai et al. [239] compared the number of broadband operators vs.
the aggressiveness (challenger) of these operators. It was concluded that aggressiveness
seems more important than the mere number of operators. Consul et al. [240] studied the
use of bi-objective heuristics for the construction of Pareto-optimal network topologies that
result in an optimum Pareto between the incumbent operators’ income and the quality of
service deterioration suffered by end-users as a result of tethering was investigated. They
demonstrated that the market interactions between such operators greatly influence the
Pareto-optimal collection of topologies.
87
The related work discussed in this section is summarized in Table 3.1.
Table 3.1 – Related Works
Ref. Summary Comments

The predicted/discussed
Proposed cloud-based network shar-
business model is not com-
[122] ing between mobile operators for the
patible with the current
LTE/EPC architecture.
design of 5G networks.
Traditional
EPC vs.
vEPC
Proposed an algorithm that derives

the optimal number and locations of
The paper’s scope is lim-
[129] vEPC’s virtual instances over the fed-
ited to the VM allocation
erated cloud which is suitable for car-
and didn’t discuss the im-
rier clouds and EPCaaS.
pact of their proposal on
Proposed resource pooling in the pricing scheme.
[130] resource-limited general-purpose
datacenters.
Studied the Mobility Management in

vEPC and proposed a Distributed Mo-
[133]
bility Management under VNF/SDN
environment.
Proposed to use NFV over SDN.

They promised OPEX reductions
which include, (eliminating the take The proposed contribu-
[131]
down and provisioning of network tion is applicable for both
upgrades and eliminating physical legacy and virtual opera-
boxes and cable migration). tors, thus doesn’t affect
the competitiveness of one
operator type over the
other.
88
Proposed an architecture that in-

cludes a business lifecycle optimizing
CAPEX/OPEX and a technical lifecy-
[134]
cle leveraging a deployment model
virtualizing MNO’s core network and
RAN elements.
It helps in benchmarking
Listed EPC market revenue projec- the market situation and
[135]
tions and compared vEPC suppliers. in in calibrating our pre-
dictions.
Proposed MOBaaS (Mobility and

[136] Offer competitive advan-
Bandwidth prediction as a service)
and multi-
EPCaaS
tenancy
tage for cloud-based mo-

bile operators
Envisioned future multi-tenant sys-
[137] tems to include Over-The-Top (OTT)
and Vertical industries
Demonstrated the feasibility of on-

[138] demand creation of cloud-based elas-
tic mobile core networks.
Proposed Caching as a Service as a

[139] new feature for cloud-based 5G net-
works
Proposed virtualization of Caching

[140]
as a Service
89
Proposed novel solutions for effec-

[143] tive subscribers state management
in quality-constrained 5G scenarios
Offer legacy-based mobile

Proposed a traffic offloading frame-
[142] operators a tool to migrate
work between legacy EPC and vEPC.
to cloud-based operations.
Divided the cloud computing busi- All these papers discussed

Business
[148]
Models
ness models into 8 dimensions. cloud adoption without

evaluating its effect on the
mobile market.
Divided cloud computing into 3 main
[149]
business models
Surveyed and identified various is-

[150] sues that will affect the different
stakeholders of cloud computing
Listed some of cloud computing’s suc-

[151]
cess factors
[152]
[156] Static pricing: Pay-per-use (flat rate)
Static pricing: Subscription

[153] Cloud Service Provider
Pnew = Pold + ∆system capacity × C
Pricing Scheme
[157] Dynamic Pricing: Auctions /Bidding

Scheme
Pricing
90
[153] Dynamic Pricing: Revised pricing
Dynamic Pricing: Revised pricing for

[158]
M/M/1 Queuing system P = C
(µ−λ)λ
Dynamic Pricing: Revised pric-

[159] ing based on Black-Scholes-Merton
Model and Moore’s law
[161] Price = f(time, location)

Cloud-based Mobile Oper-
ator Pricing Scheme
Price = f(predictable traffic, non-

[162]
predictable traffic)
Price data=f(data traffic,tether traf-

[163] fic) Price tethering =f(data traf-
fic,tether traffic))
Compensation-based pricing model

where he MVNO invests in con-
[165]
tent/advertising to compensate for
the quality of service degradation.
Pricing mechanism to mediate be- Legacy Mobile Operator

tween the user-centric and network- Pricing Scheme
[166]
centric resource management prob-
lems
91
Studied the own- and cross-price elas-

[168]
ticities of SMS and voice services
Proposed pricing scheme for suc-

[169] cessful collaboration between MNOs
sharing their C-RAN
Studied the changes in tariff plans in

[170] small and large operators over time
and competition
Studied the effectiveness of time-

[171] dependent pricing mechanisms for
large mobile operators
Studied monopolistic and oligopolis-

[172]
tic markets of VMVoIP
Studied the competition between 2G

[173] mobile operators across 32 different
countries
Compared the number of broadband

[174] operators vs. the aggressiveness
(challenger) of these operators
92
3.3. PROPOSED APPROACH AND EVALUATION
Studied he use of bi-objective heuris-

tics for the construction of Pareto-
optimal network topologies that re-
sult in an optimum Pareto between
[175]
the incumbent operators’ income and
the quality of service deterioration
suffered by end-users as a result of
tethering was investigated.
3.3 Proposed Approach and Evaluation
3.3.1 Expected scenario
The current players in the mobile industry are either legacy operators (own and manage
the physical infrastructure) or virtual operators (rent access to the physical infrastructure
of other legacy operators). With the emergence of EPCaaS, two new players will join
the scene. The first new player is the CSP offering its EPCaaS services (Usually mobile
technology vendors will play this role such as Nokia, Ericsson, Huawei, etc.). The second
new player is the cloud-based mobile operator which uses EPCaaS services to support
its business. Finally, the clients are the pool of customers currently served by legacy
operators. Legacy and cloud-based mobile operators will compete on the finite pool of
customers. Figure 3.1 shows the clients, EPCaaS offering CSPs, legacy mobile operators,
and cloud-based mobile operators.
The black link represents the legacy backhaul interface between a legacy mobile
operator’s Radio and Core network. The blue link between a mobile user and a mobile
operator (legacy or cloud-based) represents a client-operator subscription. In Figure 3.1
, all the blue links are connected to legacy mobile operators since current markets do
not have emerging cloud-based mobile operators yet. As cloud-based mobile operators
93
Figure 3.1 – Scenario.
emerge, mobile users will subscribe with the new operators thus some of the mobile users
will connect their blue lines to the new operators.
3.3.2 Proposed approach and metrics
3.3.2.1 Agent-based modeling
The exact number of factors affecting any system is nearly impossible to identify, and
the effect magnitude of every factor also varies. Designing the modeling environment
requires selecting which factors to include, the abstraction level and the detailing level
[241]. The factors we decided to include in this study are:
94
• CSP: Pricing Scheme.
• Cloud-based mobile operator: Pricing scheme, Reputation.
• Legacy mobile operator: Economic model, Pricing scheme, Reputation.
• Mobile user: Degree of knowledge and Readiness to Migrate
Many factors haven’t been included in this study such as:
• Economic/financial level of every user.
• Types of users (personal, corporate, tourist, etc.).
• Market growth.
• Marketing techniques such as targeted bundles, advertising campaigns, and political

and corporate social responsibility advertising.
• Resource capacity of a CSP.
3.3.2.2 Agent modeling
In this section, we specify the included factors and model the market players (agents).
1. EPCaaS offering CSP (CSP)

A cloud-based mobile operator has to pay the “EPCaaS offering CSP” for the resources
used, thus the pricing model of the CSP directly affects the pricing of a cloud-based
mobile operator and the market in general. The “EPCaaS offering CSP” is modeled
as follows:
In the current version of our study, the CSP is modeled as an infinite resource
provider following the pricing scheme RPM3 which is nearest to reality. There are
many general pricing strategies in literature such as: Cost-plus pricing (markup),
Competitive pricing, Price skimming, Penetration pricing, and Price bundling, but
specifically for cloud computing RPM3 is the most realistic. The assumptions behind
this model are that the CSP has no other business lines (and no other source of
95
revenues) and none of the CSPs has a competitive advantage. In a reality, a cloud-
based operator can get services from different CSPs to acquire a wide spectrum of
services, but the model’s complexity and tuning make its realization impractical.
2. Cloud-based mobile operator

A cloud-based mobile operator has to pay the “EPCaaS offering CSP” for the resources
used and then implements a pricing strategy that identifies the selling price. These
strategies differ in their aggressiveness, duration, and profit margin. Factors,
other than price, affect how appealing a cloud-based mobile operator offer looks.
These factors include Reputation and SLA. A more expensive bundle from a well-
established and reputed operator can be a wiser choice compared to a cheaper bundle
from a mobile operator with bad coverage and bad customer service. The used factors
are detailed next.
• Factors: - Reputation and SLAs (RaS): Reputation is a decisive factor other

than price. A client will select a more expensive offer if complemented by a
better reputation and a satisfying SLA. P oP is a decimal number between 0
and 1 representing the percentage of the CSPs with a good reputation. In our
simulator, we consider 2 reputation levels only (with a good reputation, without
a good reputation). RaS is a decimal number between 0 and 1 representing the
maximum difference a user is willing to pay for a better reputation. If RaS = 0
then the user is indifferent to the reputation and uses price as the only decisive
factor. If RaS = 1 then the user will select the provider with a better reputation.
• Model: In the current version of our study, the used pricing scheme is the
markup where all cloud-based mobile operators set their selling price to the
cost they are paying the EPCaaS CSP in addition to a markup margin of a
unified percentage. If all cloud-based operators are using the same markup
percentage, then the value of this percentage will not affect the inter-operator
pricing.
The number of cloud-based mobile operators is set by the user using the vari-
able “N umber_of _CM SP s”. CMSP refers to a Cloud-based Mobile Service
Provider. Since the CMSPs are not currently available, we give the legacy
mobile operators, in this simulator, a certain amount of time to stabilize their
96
prices and their market shares. The time CMSPs start joining the market
is set by a random variable ranging between 0 and 2× “Average_entry_time”.
When the new CMSP joins the market, he uses a predefined price equals
to the variable “Initial_momentum”. This price is going to be used for a
certain time after joining the market. This time is set by the variable
“N umber_of _rounds_bef ore_elasticity”. When this grace period ends, the CMSP
will start using the following pricing model.
((1 + coef f icient_of _price)

P rice = general_cost ×
(N umber_of _active_CM SP s × CM SP _users)
The “general_cost” is a constant that is used as a starting benchmark for

cloud pricing and can be changed to accommodate different currencies. The
variable “coef f icient_of _price” is used to represent the markup percentage.
“coef f icient_of _price” ranges from −1 to +1. Negative values represent ex-
tremely competitive pricing strategies to drive legacy mobile operators out of
business. “N umber_of _active_CM SP ” is the number of current CMSPs using
EPCaaS. As this number increases, it means that the CSP has more clients
and this results in an economy of scale, so the CSP can offer lower pricing. The
variable “CM SP _users” is the number of users connected to this CMSP. The
mobile operator can survive 3 periods without customers and if this period
extends, the operator becomes out of business.
3. Legacy mobile operator
Legacy mobile operators have been around for more than 30 years and are the
driving force behind consecutive developments in mobile technology. Some of the
mobile operators have a presence in more than 10 different countries such as MTN,
Orange, Telenor, Deutsche Telekom, Etisalat, Airtel, Vodafone, América Móvil, and
Telefónica.
• Factors:
– Economic Model and Market Structure Madden and Coble-Neal [242] stud-
ied the relationship between fixed-line and mobile telephony. They deduced
97
that the mobile subscription elasticity with respect to mobile subscription

price is inelastic. Inelasticity means that an increase in price results in a
slight variation in demand.
It is not found in the literature whether the mobile market follows a per-
fect competition, oligopoly, or monopolistic competition. The three market
structures have been studied in [186] and considered applicable in differ-
ent markets. It is deduced by [242] that “a certain level of cooperative
price-setting” exist between mobile operators “but no fully-fledged joint-
profit maximization”. In this chapter, we study the market under perfect
competition.
– Reputation and SLAs (RaS) Reputation is a decisive factor other than price.
A client will select a more expensive offer if complemented by a better
reputation and a satisfying SLA. PoP is a decimal number between 0 and
1 representing the percentage of the CSPs with a good reputation. In our
simulation, we consider 2 reputation levels only (with a good reputation,
without a good reputation). RaS is a decimal number between 0 and 1
representing the maximum difference a user accepts to pay for a better
reputation. If RaS = 0 then the user is indifferent to the reputation and
uses price as the only decisive factor. If RaS = 1 then the user will select
the provider with a better reputation.
• Model: In the current version of our study, the considered market structure
is the perfect competition and the used pricing scheme is markup. The num-
ber of legacy mobile operators is configured by the user using the variable
“N umber_of _DM SP s”. DMSP refers to Legacy Mobile Service Provider. The
price offered by a legacy mobile operator as follows:


 Initial_price_of _DM SP ; where t = 0
P rice =
1
General_cost × ; Otherwise


(DM SPusers )
The “general_cost” is the same as the one used in cloud-based mobile operators
and it represents a constant that is used as a starting benchmark for cloud
pricing and can be changed to accommodate different currencies. The variable
“DM SP _users” is the number of users connected to this DMSP (legacy mobile
98
operator). The variable “Initial_price_of _DM SP ” represents the price that all
legacy mobile operators start with. It is used to represent the current status of
the market where the prices are nearly equal across all mobile operators. Every
legacy mobile operator has a binary reputation (0 or 1) generated as shown
in the previous section. The mobile operator can survive 3 periods without
customers and if this period extends, the operator becomes out of business.
4. Mobile user
• Factors:
– Degree of knowledge (DoK) “Bounded rationality” adopts a restricted search

for alternatives [243], thus the decision-maker has a partial view over the
market offers. DoK is a decimal number between 0 and 1 representing the
search range for each user, where 1 means that the client (decision maker)
has information about all possible alternatives and 0 means that the client
has no access to alternatives or offers.
– Willingness and readiness for migration (RoM) “Bounded rationality”

adopts the tendency to “block out” alternatives that require substantial
changes to the existing position. RoM is a decimal number between 0 and 1
representing the maximum difference a user accepts to pay to stay at the
same position (In-house or CSP). If RoM = 0 then the user has no interest
in changing its position. This term could also be related to customer loyalty.
If RoM = 0 then the user has perfect loyalty.
• Model: The user has access to the list of available operators (both cloud-based
and legacy), but based on the “bounded rationality” concept every user selects
randomly a list of operators to contact for offers. This list is called knowledge_list,
which is of size (N umber_of _active_M SP s × Degree_of _knowledge). When the
variable “Degree_of _knowledge” is set to 1, the user can access all prices of-
fered in the market. Every round, the user searches his knowledge-list and
compares the best-offered price (minimum_price) to his current price. If the
minimum_price × (RoM ) > current_price × (1 − RaS) the user will move to the
new operator and his price becomes equal to the minimum_price. In case the
99
user is to stay with his current operator, the user’s price will be updated to the
operator’s new price.
3.3.3 Metrics
To better understand the market behavior, multiple metrics should be used. The used
metrics are:
1. Average cost paid by clients: The average cost per resource paid by the clients.
2. Percentage of CMSPs in business: Percentage of CSPs having at least 1 cus-

tomer.
3. Percentage of DMSPs in business: Percentage of CSPs having at least 1 cus-

tomer.
4. Average operator size: Average number of clients per operator.
3.3.4 Evaluation
The market behaved very differently under different configurations; thus, we can expect
the impact of EPCaaS to differ between countries. We were able to distinguish 4 different
consequences (outcomes) that could affect the telecom market if EPCaaS was introduced
and cloud-based mobile operators joined. The outcomes are:
3.3.4.1 Outcome 1: All legacy mobile operators driven out of the market with the
final user cost exceeding the initial cost.
As can be seen in Figure 3.2, as new players joined the market, the legacy mobile operator
was driven out of the market to be replaced by new cloud-based mobile operators who
competed on the market share until one dominated the whole market. As can be seen
100
in the average price plot, as a new member joins the market, the prices severely drop
(perfect competition) and rise as one leaves the market and competition stress decreases.
Since one player dominated the market, the user cost elevated and stabilized much higher
than what was during the competition of legacy mobile operators.
3.3.4.2 Outcome 2: All legacy mobile operators driven out of business with the
final user cost less than the initial cost.
As can be seen in Figure 3.3, at tick 2 all legacy mobile operators were in business (100%
DMSPs in business) and the cost was around 70 monetary units. Staring tick 3 legacy
operators started to exit the market as new cloud-based mobile operators entered. The
cost gradually dropped (except for a peak when all legacy operators left the market) as
more users moved from legacy to cloud-based operators. The users were distributed over
the cloud operators who offered the same price which is an all-time low.
3.3.4.3 Outcome 3: Co-existence between legacy and cloud-based mobile

operators. Some of the legacy and cloud-based operators left the market.
As can be seen in Figure 3.4, the cloud-based mobile operators started replacing legacy
operators until they lost their initial-momentum and had to abide by the discussed pricing
scheme. The pricing scheme, with the users distributed among them, prevented the
cloud-based operators from offering more competitive prices and failed to drive the legacy
operator out of business. Since the users are divided over the active legacy and cloud-
based operators, the resultant price is higher than the start price, so we deduce that this
competition resulting from the introduction of EPCaaS is against the financial benefit of
users.
3.3.4.4 Outcome 4: Cloud-based mobile operators join the market and leave later
because of competition from the legacy mobile operators.
As can be seen in Figure 3.5, the cloud-based mobile operators joined the market and left
sequentially as the cost dropped into very competitive rates. Finally, some of the legacy
101
Figure 3.2 – Outcome 1: All legacy mobile operators driven out of business with the final user
cost exceeding the initial cost.
102
3.4. CONCLUSION AND FUTURE CHALLENGES
mobile operators withstood the emergence of EPCaaS which wasn’t surviving competitive
pricing.
3.4 Conclusion and Future Challenges
In this chapter, we raised the issue of market structure uncertainty due to the introduction
of a multitenancy cloud to the 5G mobile network. As delicate as this issue may get, no
historical data is available to guide us during the emergence of multitenancy in the mobile
market. We modeled the 5G network using the Agent-Based-Modeling technique and the
market behavior has been predicted using this model. Since the mobile market differs
between countries, we tested our model for various configurations, and we observed 4
different market behaviors listed as:
1. All legacy mobile operators driven out of the market with the final user cost exceeding
the initial cost.
2. All legacy mobile operators driven out of business with the final user cost less than
the initial cost.
3. Co-existence between legacy and cloud-based mobile operators. Some of the legacy
and cloud-based operators left the market.
4. Cloud-based mobile operators join the market and leave later because of competition
from the legacy mobile operators.
As EPCaaS/multitenancy emerges, mobile markets’ behavior will range from adap-

tation (remain oligopoly) to complete market revolution (changing its nature to monopolis-
tic competition). The conclusions of this chapter can be generalized to all applications of
cloud computing. Thus, any DSS for cloud adoption should not rely solely on the business’
requirements, but consider the market behavior in general and its possible economic
implications. We can deduce that the businesses’ actions towards cloud computing are
active during the cloud’s first stages but becomes reactive with its introduction to the
market.
103
This chapter considered competitive markets only (monopolistic competition and

markup pricing). An important upgrade is investigating the market behavior in oligopoly,
which affects the modeling (pricing scheme) of legacy-mobile operators who will be collab-
orating in countering the impact of cloud-based mobile operators.
As EPCaaS becomes a reality, the collected data will help us validate and fine-tune
the performance of our model. A fine-tuned model will help decision-makers plan their
investments based on the predicted market price.
Finally, this model is expected to be the fundamental framework for testing the
readiness of the mobile market to accept new services/technologies and a decision-support
solution for policy makers and executives.
This chapter satisfies CBOD’s task 3 and objective 3. The updated project status is
shown in table 3.2.
Status Status
Objective 4 Pending Task 4 Done
Appendix
a. Output format explanation

This section shows the output of a sample scenario to introduce the output formats,
which will be used in the coming sections to measure the market behavior.
When the setup button is pressed, a new instance of the simulation is generated.
This instance is based on the inputs set in the left part of the Main view. The visual
representation of the simulation is shown in the black box on the right side of the
Main view in Figure 3.6. The components in the black box are positioned similar to
104
Figure 3.1 where the left side represents the mobile users, the upper-right agents
are cloud-based mobile operators and the lower-right (represented by big houses)
are the legacy mobile operators.
When the “go once” button is pressed, the users start looking for the best offers
and subscribing with their best local operator (based on “bounded-rationality”). In
Figure 3.7, we can see that the users are connected with the legacy mobile operators,
and next to each user, his subscription price is shown in addition to the MSP id
using this format: price – MSP id. The cloud-based operators haven’t entered the
market yet and their printed prices are 0 (not active) using the following format:
Offered price - Reputation. The legacy mobile operators are connected with lines
from the users and their prices are printed using the following format: Reputation –
Number of subscribed users – Offered price.
The fifth sample round, shown in Figure 3.8, shows the outputs plotted in three
windows. The first window shows the average cost paid by the users where we can
deduce that the price peaked at the start of the simulation and then converged
towards a constant price similar to what happens in real-life where the introduction
of the mobile service leads to peak prices and then competition drives the prices
down to a stable price. In the second window, we see that the percentage of legacy
mobile operators (red line) was 100% and then converged to 66%, so we can deduce
that the market settings set in this simulation can’t accept three legacy operators
(since the market is small) and stabilize at just two mobile operators. Increasing
the “number_of _clients” variable can keep the third operator in the business. The
percentage of cloud-based mobile operators (blue line) is still zero, which means
that none joined the market yet. As the third legacy operator was kicked out of
business, the subscribed were distributed over two mobile operators which increase
the average number of subscribers per active operator and this can be seen in the
third window.
The twentieth sample round, shown in Figure 3.9, shows three cloud-based mobile
operators joining the market and driving all legacy mobile operators out of business.
b. Netlogo Code
globals [
105
cout ;; Used as the xcor value for the first Mobile Service Provider.
counter ;; Used to index the IDs of the clients. The value is set to 0
in setup.
counter-DMSP-CMSP ;; Used to index the IDs of the Cloud-based Mobile
Service Providers and Dominant Mobile Service Providers.
General_Cost ;; c = 1000
Number_of_active_CMSPs ;; n
number_of_changes
average_cost
CMSP_count
DMSP_count
List_of_active_MSPs
Number_of_active_MSPs
]
breed [clients client] ;; Breed of clients.
breed [CMSPs CMSP] ;; Breed of CMSPs.
breed [DMSPs DMSP] ;; Breed of CMSPs.
clients-own [
cost ;; Used to represent the cost the company is paying for its IT
services that could be offloaded.
id ;; Client’s ID.
connected_to ;; CMSP’s OR DMSP’s ID
knowledge_list ;; list of CMSPs and DMSPs the user knows
]
CMSPs-own [
price ;; Used to represent the price the CMSP is offering its client.
Price = ( general_cost * ( 1 + coefficient-of-price ) / (
Number_of_active_CMSPs * CMSP_users ) )
CMSP_users ;; Number of users connected to this CMSP
id ;; CMSP’s ID. Starts at 10.
106
entry_period ;; When should enter business. Random number between 0 and

2*Average-entry-time
in_business ;; Boolean value. If tick > entry_period and CMSP_users > 0
Elacticity_starts ;; Before elasticity, CMSP_users will be
Initial_momentum; Elasticity_starts = entry_period + Number-of-
rounds-before-elasticity
Good_Reputation ;; 1 is true, 0 is false
]
DMSPs-own
[
price ;; Initially it is set to Initial-price-of-DMSP; Used to
represent the price the DMSP is offering its client. P =
general_cost / DMSP_capacity
DMSP_users ;; Number of users connected to this user
id ;; CMSP’s ID. Starts at 1
in_business ;; Boolean value. If DMSP_users > 0
]
to fill_List_of_active_MSPs
;;
;; Identify the active MSPs so that a user can search for a suitable
price between the active ones. It will decide the searched list
based on the degree-of-knowledge
;;
set Number_of_active_MSPs 0
set List_of_active_MSPs []
ask DMSPs [if in_business = 1 [set Number_of_active_MSPs
Number_of_active_MSPs + 1
set List_of_active_MSPs sentence List_of_active_MSPs id]]
ask CMSPs [if in_business = 1
[set Number_of_active_MSPs Number_of_active_MSPs + 1
107

;;print "␣Number␣of␣active␣MSPs"
;;print Number_of_active_MSPs
if Number_of_active_MSPs > 0
[
let list1 [id] of DMSPs
let list2 [id] of CMSPs
let list_of_ids sentence list1 list2
;;
;; Fill a knowledge list i.e. the list of MSPs the user contact looking
for the best offer. This is because of bounded rationality
;;
ask clients [
;;
;;Currently the user selects a new knowledge-list every tick, and
the new knowledge-list may not include his current operator. The
best way to be done is to keep the current operator in the list
;;since the user will cover all operators after certain number of
ticks
;;
let start_of_knowledge random Number_of_active_MSPs
ifelse start_of_knowledge + ( Number_of_active_MSPs * Degree-of-
knowledge) < ( Number_of_active_MSPs )
[set knowledge_list sublist List_of_active_MSPs start_of_knowledge (
round (start_of_knowledge + (Number_of_active_MSPs * Degree-of-
knowledge)) )]
[let subl1 sublist List_of_active_MSPs start_of_knowledge (
Number_of_active_MSPs )
let subl2 sublist List_of_active_MSPs 0 (Number_of_active_MSPs *
Degree-of-knowledge - (Number_of_active_MSPs - start_of_knowledge
) )
set knowledge_list sentence subl1 subl2
108
]
]]
end
to setup
clear-all
;;ask patches [set pcolor white]
set-default-shape clients "person"
set-default-shape CMSPs "house"
set-default-shape DMSPs "house"
set cout (100 / (Number-of-CMSPs + Number-of-DMSPs + 1) ) ;; It is a
dynamic value for ycor which is the position of the first CMSP.
Other CMSPs will be shifted by a dynamic value called space
set counter 0 ;; Used to index the IDs of the clients
set counter-DMSP-CMSP 1 ;; Used to index the IDs of the DMSPs and CMSPs
set Number_of_active_CMSPs 0
set General_Cost 1000
create-DMSPs Number-of-DMSPs [
let space ( 90 / ( Number-of-CMSPs + Number-of-DMSPs ) ) ;; Dynamic
value for CMSP locations
set size 5
set price Initial-price-of-DMSP

set DMSP_users 0
set in_business 1
let reputation random 100

set Good_Reputation 0
if reputation < 100 * PoP
109
[set Good_Reputation 1]
set xcor 70
set ycor cout
set cout cout + space
set id counter-DMSP-CMSP
set counter-DMSP-CMSP counter-DMSP-CMSP + 1
set number_of_changes 1
]
set DMSP_count 0
ask DMSPs with [ in_business = 1 ] [ set DMSP_count DMSP_count + 1 ]
create-CMSPs Number-of-CMSPs [
let space ( 90 / ( Number-of-CMSPs + Number-of-DMSPs ) ) ;; Dynamic
value for CMSP locations
set size 1
set entry_period random (2 * Average-entry-time )

set in_business 0
set Elacticity_starts ( entry_period + Number-of-rounds-before-
elasticity )
set CMSP_users Initial_momentum
set price 0


110
set xcor 70
set ycor cout
]
ask CMSPs with [ in_business = 1 ] [ set Number_of_active_CMSPs
Number_of_active_CMSPs + 1 ]
set cout (100 / (number-of-Clients + 1) )

;;print list_of_ids
fill_List_of_active_MSPs
;;print List_of_active_MSPs
create-clients number-of-Clients [
set size ceiling ( 20 / number-of-Clients) ;; be easier to see
let space (90 / number-of-Clients)
set cost 100000

set id counter
set counter counter + 1
set connected_to -1
set xcor 20
111
set ycor cout

;;
;; Fill a knowledge list i.e. the list of MSPs the user contact looking
for the best offer. This is because of bounded rationality
;;
ask clients [
ifelse start_of_knowledge + ( Number_of_active_MSPs * Degree-of-
knowledge) < ( Number_of_active_MSPs )
[set knowledge_list sublist List_of_active_MSPs start_of_knowledge (
round (start_of_knowledge + (Number_of_active_MSPs * Degree-of-
knowledge)) )]
let subl2 sublist List_of_active_MSPs 0 (Number_of_active_MSPs *
Degree-of-knowledge - (Number_of_active_MSPs - start_of_knowledge
) )
]
]]
ask CMSPs [ set plabel (word (precision price 2) "␣-␣" Good_Reputation)

]
ask DMSPs [ set plabel (word (precision price 2) "␣-␣" Good_Reputation)
]
ask clients [ set plabel precision cost 2 ]
reset-ticks
end
to count_customers
112
end
to exiting_business
if ( ticks > 3 ) [
ask CMSPs [ set in_business 0 ]
ask DMSPs [ set in_business 0 ]
set CMSP_count 0
set DMSP_count 0
foreach sort-on [id] Clients

[ ?1 ->
let CMSP_connnected_to [connected_to] of ?1
ask CMSPs with [id = CMSP_connnected_to][set in_business 1]
]
ask CMSPs
[
if in_business = 1
[
set CMSP_count CMSP_count + 1
]
]

[ ?1 ->
let DMSP_connnected_to [connected_to] of ?1
ask DMSPs with [id = DMSP_connnected_to][set in_business 1]
]
113
ask DMSPs
[
if in_business = 1
[
set DMSP_count DMSP_count + 1
]
]
]
set DMSP_count 0
end
to Join_business
ask CMSPs
[
if ( ticks > entry_period - 1 ) AND ( ticks < Elacticity_starts )
[
set in_business 1
print id
set price initial_momentum
]
]
end
to update_pricing
114
ask CMSPs with [in_business = 1 ]

[
let CMSPU 0
ask clients with [ connected_to = id ] [ set CMSPU CMSPU + 1 ]
set CMSP_users CMSPU
ifelse CMSP_users > 0
[
ifelse ticks < Elacticity_starts [set price initial_momentum]
[set price ( general_cost * ( 1 + coefficient-of-price ) / (
Number_of_active_CMSPs * CMSP_users ) )]
]
[
set in_business 0
;;set price 1000
]
ask DMSPs with [in_business = 1]

[
let DID id
let DMSPU 0
ask clients with [ connected_to = DID ] [ set DMSPU DMSPU + 1 ]
set DMSP_users DMSPU
print DMSPU
ifelse DMSP_users > 0
[set price ( general_cost / DMSP_users )]
[set price general_cost ]
]
end
to go
115
;;Update list of active MSPs

;;Linking users
dynamic-revised-pricing
;;Update pricing
update_pricing
;;Linking users
;;Anybody wants to go out of business

exiting_business
;;Anybody to join the market
Join_business
measure-CMSP-involvement
ask CMSPs [ ifelse in_business = 1 [set plabel (word (precision price
2) "␣-␣" Good_Reputation)] [set plabel "" ]]
ask DMSPs [ ifelse in_business = 1 [ set plabel (word id "␣-" (
DMSP_users) "␣-␣" price)] [set plabel "" ] ]
ask clients [ set plabel ( word (precision cost 2 ) "␣-␣" connected_to
)]
tick
end
to measure-CMSP-involvement
end
to optimizeprofits
ask CMSPs[
let checker_price price
let number_of_prices_lower_than_checker 0
let second_lowest_price 2 * Initial-price-of-DMSP
116
foreach sort-on [id] CMSPs

[ ?1 ->
let read_CMSP_price [price] of ?1
if read_CMSP_price < checker_price
[set number_of_prices_lower_than_checker
number_of_prices_lower_than_checker + 1]
]
if number_of_prices_lower_than_checker = 0 ;I am the lowest price. I
can increase price without loosing customers
[
[ ?1 ->
let read_CMSP_price [price] of ?1
if (read_CMSP_price < second_lowest_price) AND (
read_CMSP_price > checker_price)
[
set second_lowest_price read_CMSP_price
]
]
set price second_lowest_price - 1
]
;If I am not the lowest price, I shall not increase it to remain as much
competitive as possible.
]
end
to dynamic-revised-pricing ; Business Model
let read_CMSP_price 0
let read_CMSP_id 0
let read_DMSP_price 0
117
let read_DMSP_id 0
ask clients[
let read_client_connected_to connected_to

let read_client_id id
let read_client_cost cost
;;let read_client_client_capacity 1
let klist knowledge_list
;let node1 end1
let node1 one-of clients with [ id = read_client_id]
let minimum_price 0
;;let Reputation_of_the_CMSP_offering_minimum_price 0
let id_of_CMSP-with_minimum_price 0
;;let Reputation_of_the_CMSP_connected_to 0
ask CMSPs
[
set read_CMSP_price price
set read_CMSP_id id
;;set read_CMSP_Good_Reputation [Good_Reputation] of ?1
if read_client_connected_to = read_CMSP_id ; Updating the cost

[
;;set Reputation_of_the_CMSP_connected_to
read_CMSP_Good_Reputation
set read_client_cost read_CMSP_price ; the client has to pay

for his resourcesBased on the new price the CMSP is
offering
set cost read_CMSP_price ; Both increasing and decreasing
118
costs
]
]
ask DMSPs
[
set read_DMSP_price price
set read_DMSP_id id
if read_client_connected_to = read_DMSP_id ; Updating the cost

[
;;set Reputation_of_the_CMSP_connected_to
set read_client_cost read_DMSP_price ; the client has to pay

for his resourcesBased on the new price the CMSP is
offering
set cost read_DMSP_price ; Both increasing and decreasing
costs
]
]
set minimum_price read_client_cost

set id_of_CMSP-with_minimum_price -1
;;set Reputation_of_the_CMSP_offering_minimum_price
Reputation_of_the_CMSP_connected_to
ifelse ticks = 1 [
119
ask CMSPs
[
set read_CMSP_id id
foreach klist
[
?2 ->
if(read_CMSP_id = ?2)
[
if (read_CMSP_price < minimum_price )
[ set id_of_CMSP-with_minimum_price read_CMSP_id ]
]
]
]
ask DMSPs
[
set read_CMSP_id id
foreach klist
[
?2 ->
[
]
]
]
]
120
[
ask CMSPs
[
set read_CMSP_id id
;;set read_CMSP_capacity [CMSP_users] of ?1
if (minimum_price * (1 - RaS) * (RoM) > read_CMSP_price)

[
foreach klist
[ ??1 ->
let xid ??1
if read_CMSP_id = xid
[
set minimum_price read_CMSP_price
set id_of_CMSP-with_minimum_price read_CMSP_id
]
]
]
]
ask DMSPs
[
set read_DMSP_id id
if (minimum_price * (1 - RaS) * (RoM) > read_DMSP_price)
121
[
foreach klist
[ ??1 ->
let xid ??1
if read_DMSP_id = xid
[
set minimum_price read_DMSP_price
set id_of_CMSP-with_minimum_price read_DMSP_id
]
]
]
]
]

[ ?1 ->
set read_CMSP_price [price] of ?1
set read_CMSP_id [id] of ?1
if (id_of_CMSP-with_minimum_price = read_CMSP_id) ; User

found better price
[
;show connected_to
let was_connected_to connected_to
if connected_to > -1 ; If not first round

[
122
ask links with [ color = 12 + read_client_id] [die] ;

Disabling the connection with the old CMSP
]
set connected_to read_CMSP_id ; Connecting to the new CMSP
set cost read_CMSP_price
let node2 one-of CMSPs with [ id = read_CMSP_id]
ask node1 [ create-link-with node2 [ set color 12 +
read_client_id] ]
;ask CMSPs with [id = iCMSP] [ create-link-with end1 [set color
12 + idinput + idinput]]
]
]
foreach sort-on [id] DMSPs

[ ?1 ->
set read_DMSP_price [price] of ?1
set read_DMSP_id [id] of ?1
;;set read_DMSP_capacity [DMSP_users] of ?1
if (id_of_CMSP-with_minimum_price = read_DMSP_id) ; User

found better price
[
;show connected_to

[
ask links with [ color = 12 + read_client_id] [die] ;

Disabling the connection with the old CMSP
123
]
set connected_to read_DMSP_id ; Connecting to the new CMSP
set cost read_DMSP_price
let node2 one-of DMSPs with [ id = read_DMSP_id]
ask node1 [ create-link-with node2 [ set color 12 +
read_client_id] ]
;ask CMSPs with [id = iCMSP] [ create-link-with end1 [set color
12 + idinput + idinput]]
]
]
]
end
124
Figure 3.3 – Outcome 2: All legacy mobile operators driven out of business with the final user
cost less than the initial cost.
125
Figure 3.4 – Outcome 3: Co-existence between legacy and cloud-based mobile operators. Some of
the legacy and cloud-based operators left the market.
126
Figure 3.5 – Outcome 4: Cloud-based mobile operators join the market and leave later because of
competition from the legacy mobile operators.
127
Figure 3.6 – Simulator’s Main View.
Figure 3.7 – First Sample Round.
128
Figure 3.8 – Fifth Sample Round.
Figure 3.9 – Twentieth Sample Round.
129
Chapter 4
The Oligopoly of Cloud Computing
Contents
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2 Research Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.2.1 Dynamics of oligopoly . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.2.2 Game theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.3 Research Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.4 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.5 Oligopoly of Cloud Computing . . . . . . . . . . . . . . . . . . . . . . . 140
4.5.1 Oligopoly of Cloud Computing: elastic demand use case . . . . . . 141
4.5.2 Oligopoly of Cloud Computing: inelastic demand use case . . . . . 145
4.6 Conclusion and Future Work . . . . . . . . . . . . . . . . . . . . . . . . 146
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.1 Introduction
The adoption of cloud computing divides the market into two categories namely adopters
and non-adopters. This division identifies the organizations’ business models, pricing
131
4.2. RESEARCH BACKGROUND
models and chances of success and failure. We are interested in investigating the market
dynamics when those two categories collide.
In the previous chapter, we studied the market dynamics under the assumption of perfect
competition for a cloud computing scenario namely the adoption of multitenancy by mobile
operators. The legacy mobile operators are few in number, but huge in size and have
the capability to alter the market price, so they match the definition of oligopolistic
players [244]. As current prices are relatively stable, we can safely expect that the market
has reached a form of oligopolistic equilibrium (not necessarily Cournot’s or Betrand’s).
This equilibrium has been reached due to barriers to entry and limited licensed mobile
spectrum which made the market relatively closed in terms of operators. As cloud-based
mobile operators emerge, they will enter a stable market with reached equilibrium which
makes their impact non-trivial in case they succeeded in unbalancing the market.
The dominant oligopoly models are Cournot’s [245] and Bertrand’s [246] where each
is based on a set of assumptions that may or may not be applicable to a certain market. The
oligopoly in the mobile market has no formal representation in literature and this is due to
the incompatibility between the existing oligopoly models’ assumptions and the market’s
condition. In this chapter, we will try to cover this gap by implementing Cournot’s oligopoly
model for the mobile market with additional assumptions and mathematically calculate
its equilibrium price and stability. We will also simulate the oligopolistic behavior of
legacy mobile operators using ABM and draw conclusions.
4.2 Research Background
As new players enter the market, it is expected that legacy mobile operators counter this
new market penetration and either set coordinated competitive strategies or maintain
oligopolistic strategies. Cooperative competitive strategies can be regulated by govern-
ment policies so it is outside the scope of this work. In this work, we will consider that
the legacy mobile operators adopt an oligopolistic strategy.
We are interested in studying the mobile market’s transition state when EPCaaS
132
services get introduced and new cloud-based operators enter the market. This market
penetration is expected to be faced with a non-coordinated opposition from the oligopoly of
legacy mobile operators. To do so, we will study the dynamic interaction between different
players (mobile user, legacy mobile operator, cloud-based mobile operator, and CSP) and
predict the market’s emerging status under oligopolistic rules. In this section, we discuss
the dynamics of oligopoly which were used for studying the interaction between players
in addition to game theory.
4.2.1 Dynamics of oligopoly
Cournot [245], defined oligopoly as the competition between a few sellers, placing it
between monopoly and perfect competition. Monopoly is one seller explicitly dominating
the supply of the market and having total control over the price/cost structure; while
perfect competition is composed of small suppliers having no control over the market
price. In perfect competition, firms with flexible cost functions will lower their cost ratio
to get positive profit or lower their price to become more competitive and the others on
the margin will earn nothing thus causing them to exit the market.
Duopoly, a particular case of an oligopoly with only 2 suppliers, is more complicated

than either monopoly or competition. They take into account the "market demand and try
to equate marginal revenue to cost"[244]. Yet in this case, marginal revenue will not only
depend on a player’s own supply, but also on the supplies of all the competitor(s). In this
scenario "each of them is large enough to influence the market price" [244], thus taking
into account the competitors’ expectations and how they would react back to possible
moves when calculating optimal decisions.
4.2.1.1 Equilibrium
Palander [247] found either a quasi-periodic orbit or a chaotic orbit, which has no dis-
cernible structure at all. The system was characterized by Cournot [245] as a recursive
iterative map, which is a series of motions that leads to a conclusion where the orbit
133
might be a fixed point or equilibrium.
In the case of Cournot’s equilibrium, self-consistency is the driver with an unprob-

lematic situation. The competitors at equilibrium will have the same outcomes for the
same actions, making it risk-free to operate the market while expecting the same results.
When it comes to the chaotic or the quasi-periodic orbit in an oligopoly when in

sufficiently low periods, competitors can learn and adapt to them and generally voiding
the periodicity. In other terms, when competitors learn and adapt to the dynamics of the
market, they are implementing actions that return the periodic orbit into equilibrium.
This is "unrealistic and blows up the concept of rational expectations in current economic
theory."[244]
4.2.1.2 Multiplicity of attractors
Equilibrium can be around one fixed point which is considered to be a static equilibrium,
or range between a set of fixed points. Equilibrium can also be around a set of fixed points
"coexisting with other types of attracting orbits" [244]. The linear model for example
has "one unique attractor, but nonlinearity opens up for coexistence between different
attractors" [244].
Nonlinearities were first suggested for monopoly by Joan Robinson in 1933 [248],
and for duopoly by Tord Palander in 1939 [247]. For the simple reason that prices and
quantities cannot be negative, piecewise linear models are nonlinear. In the case of a
monopoly with linear demand functions, a high-priced product is a amenity restricted
for rich customers; as the price is reduced, other categories of consumers may buy it. As
a result, marginal cost functions can cross at multiple locations, giving the monopolist
multiple coexisting local profit maxima.
In an oligopoly, Palander [247] assumes that each seller will select the global
maximum profit. However, it depends on the competitor’s market share. The reaction
function, representing the anticipated reaction of the competitor, comes in disjoint pieces
leading to multiple attractors. In a duopoly, Palander described also the possibility of two
coexisting "Cournot equilibria with a periodic oscillation" [247].
134
4.2.1.3 Strategy types
In the Cournot setting, a competitor can simply decide his best action if he is knowledgeable
of the competitors’ next move, which is not usually the case. Cournot process assumes
that each agent considers all his competitors’ actions will remain constant and select his
best action based on this assumption. Definitely, the agent will be disappointed by this
assumption, which is generally called naïve or myopic.
Even if orbits are simple and strictly periodic, the agent cannot learn them and
modify his actions because this would affect the orbit’s periodicity (since it is also assumed
that the agents are large enough to change the price-supply structure). Furthermore,
because chaotic and quasiperiodic orbits are hard to learn, the equilibrium is the only
fixed-point orbit where competitors’ expectations are not fooled.
Competitors are expected to practice conservatism, which is sticking to a linear

action plan rather than leaping from their past movements to the current estimated
optimum responses. The adaption weights are then increased from zero to one, trans-
forming the duopoly into a two-dimensional system. If competitors simply react to the
last competitor’s action, a combination of two steps produces a second-order system which
converts the market into iteration-based game.
4.2.1.4 Stability of Cournot Equilibrium
Achieving static equilibrium is very interesting in dynamic settings, for this reason,
linearity is usually considered. Using linear demand functions and constant marginal cost,
the corresponding reaction functions are linear with slope a = − 12 . Under this assumption,
Cournot’s static equilibrium and "Palander’s two concurrent Cournot equilibria are stable"
[244].
4.2.1.5 Stability of Bertrand Equilibrium
Bertrand’s oligopoly model [246] is very similar to that of Cournot, but sets the price as the
controlled variable (instead of quantity as set by Cournot). Bertrand’s model considers the
135
supplied quantity as directly proportional to the demand and intrinsically assumes that
the supplier has the capability of infinitely supplying a product [249]. This is problematic
in markets with limited supply capacity or when the cost function has a fixed component,
because in such cases, Bertrand’s oligopoly model fails to reach nash equilibrium [249].
The mobile network market has limited supply capacity (limited spectrum) and its cost
function has a fixed component (fixed spectrum lease). Accordingly, Bertrand’s oligopoly
model is considered impractical for this work.
4.2.2 Game theory
Game theory is a multi-agent decision-making theory [250] involving decisions and

utilities for these economic agents. It has two important characteristics:
1. Dynamic: where the order of decisions is important.
2. Noncooperative: where agents pursue their interests which are partially conflicting
with others.
We need a technique for thinking of utility maximizing in mathematical terms since

game theory is a technology for formal modeling. A utility function is a technique that
performs this purpose. A utility function is a measure of preferences for a collection of
goods and services in economics. A game is defined as any circumstance in which at least
one agent may only maximize his utility by predicting (either consciously or implicitly in
his conduct) the responses to his actions by one or more other agents. In video games,
agents are referred to as players.
The distinction between games with perfect and imperfect information is due to
differences in how games are represented dependent on the order of play. There is a
distinction to be made between sequential-move games, in which players select their
tactics one by one, and simultaneous-move games, in which players select their plans all
at once.
136
4.3. RESEARCH FRAMEWORK
It’s critical to understand the difference between cardinal and ordinal utility when
discussing utility functions. "An ordinal utility function is a function that represents an
agent’s preferences on an ordinal scale in economics. According to ordinal utility theory,
asking an agent about his choice for one option vs another alternative is only useful if
not possible. A cardinal utility function or scale is a utility index that uniquely retains
preference orderings up to positive affine transformations" [250], allowing the customer
to express satisfaction in cardinal or quantitative values.
4.2.2.1 Nash equilibrium
A set of strategies is named to be Nash equilibrium if none of the participating players

can improve his payoff, given the strategies of all other players in the game. Nash [251]
explicitly defined the equilibrium of a noncooperative game as a profile of strategies, one
for each participant in the game, where each player’s plan optimizes his anticipated utility
reward compared to the other players’ strategies. A network of mutually restricting links
may be described as an economic system, and the equilibrium of such systems are their
stable states.
4.3 Research Framework
In the previous chapter, we evaluated the market behavior when multi-tenancy is intro-
duced under the condition of free competition. Free competition can be enforced through
government policies that are outside the scope of this research. In the absence of such
policies, the market is most likely going to be oligopolistic especially when legacy providers
face an existential threat like multi-tenancy and virtualized mobile operators [244].
In this chapter, we are interested in evaluating market behavior under oligopolistic

conditions. Two cases will be considered both elastic and inelastic demands for cloud
services and in particular mobile services. For the first case, the market stability is going
to be modeled mathematically where legacy mobile operators have a limited capacity
inherited from the nature of their infrastructures; while virtualized mobile operators
137
are free from this constraint [244]. The possibility of network expansion is neglected
from this model because of its extreme mathematical complexity. The market behavior
of the second case will then be modeled using ABM. The originality of this work can be
identified as:
• This work is the first, to the best of our knowledge, to discuss the co-existence of limitless
and limited capacity mobile operators.
• It extends the results of the previous chapter under more realistic assumptions. This
work is the first, to the best of our knowledge, to formally represent an oligopoly in
telecommunication.
4.4 Literature Review
Laffont and Tirole [252] studied the Endogenous Marginal Costs between telecommuni-
cation operators and their effect on the operators’ strategies. In this study we assume
that the terminating access is priced at the marginal cost, thus equating between inter-
operator and intra-operator traffic. This assumption simplifies the model, discussed
in this chapter, by eliminating size proportional pricing and Platform Competition in
Two-sided Markets [253].
Hauge and Jaminson [254] examine common techniques for calculating market
concentration, as well as suppliers’ ability to raise prices, barriers to entrance and
departure, and anticompetitive behavior. They also go through strategies for recognizing
market power, with an emphasis on the telecoms industry, as well as methodologies for
determining if corporations are using their market power.
Jaminson [255] argued that the telecommunication market is a workable competi-

tion, which is a market condition that does not meet the perfect competition model but
has enough perfect competition characteristics that government intervention is unneeded
and potentially destructive.
Vafamehr and Khodayar [256] studied the "oligopolistic competition among cloud
138
providers in electricity and data networks" [256]. At each of the electricity and data
networks system, "the operation strategies of the market participants are determined
by forming bi-level optimization problems in which the upper-level problem maximizes
the payoff of the market participants while the lower-level problem represents the corre-
sponding market settlement process" [256].
Helmes and Schlosser [257] studied oligopoly pricing in isoelastic adoption models
and calculated the formulas for the equilibrium prices as a function of the number of
competing firms. Farias et al. [258] proposed an approximation of dynamic oligopoly
that is suitable for computer simulation and modeling. The authors suggest that their
approach can significantly expand the set of dynamic oligopoly models that can be analyzed
computationally. Desai et al. [259] extended the above work by using a smoothed linear
program which yields more accurate results.
Klemperer and Meyer [260] studied the effect of per-supplier “supply function” on
the equilibria of oligopoly. They proved that under uncertainty, symmetrical oligopoly with
homogeneous production can reach uniqueness if Nash equilibrium in supply functions is
reached.
Competing businesses incur actual resource costs in manipulating their marginal

costs in order to affect the result of the game they intend to play in stage two, according to
Long and Soubeyran [261]. They demonstrated that even for ex-ante symmetric companies,
the best resource allocation inside an oligopoly may be asymmetric.
Ledvina and Sircar [262] are among the few who studied oligopoly between firms
having different costs of production. They concentrated on the linear demand structure
with constant marginal but asymmetric costs.
Chevalier-Roignant et al. [263] explored non-cooperative firms with different cost

structures in the context of uncertainty in strategic investment. They discovered that for
extremely non-linear capacity distributions, "(ex-ante) linear capacity investment cost
asymmetry leads to (ex-post) heterogeneity" [263].
For two consecutive periods, Van den Berg et al. [264] looked at a dynamic Cournot
duopoly in which providers had a restricted number of items available. It was discovered
139
4.5. OLIGOPOLY OF CLOUD COMPUTING
that under commitment, tactics are independent of the rival’s realized sales, that each
allocation of beginning supply has a unique Nash equilibrium, and that prices rise with
time. Sellers can modify their supply choice after the first period if there is no commitment.
A subgame perfect Nash equilibrium may not necessarily exist in this scenario, and prices
may fall over time.
Bischi et al. [265] based their dynamic oligopoly model with R&D externalities
on the work of [231], [232], [233], [234], [235] to formulate a two-stage oligopoly game
evaluating the suitability of sharing R&D results with partner firms as well as gaining
knowledge for free through spillovers. As the mobile technology is guided by the 3GPP
standards, such a model does not apply to our case assuming that mobile operators are
not vendors or manufacturers of telecommunication technology. Mobile operators’ R&D
efforts are limited to the production of differentiating services that, as argued in the
previous chapter, create a competitive advantage against small operators who can’t afford
to develop or to integrate all available services.
Ara et al. [266] studied the government imposing tariffs as a vertical oligopoly.
The discussed case study has the Home market specialized in final goods while the
foreign market specialized in intermediate inputs to mimic the Japan-China case. Other
applications of the mixed Cournot game have been investigated by Zhe et al. [253] and it
showed that private firms aim for profit maximization only while semipublic ones aim at
social welfare and profit maximization.
4.5 Oligopoly of Cloud Computing
One of Cournot’s basic assumptions when studying oligopoly is the existence of an inverse
demand function where the price-quantity relationship is maintained. This assumption
intrinsically includes the assumption of an elastic demand function, which is not always
true. In this chapter, we will consider both elastic and inelastic demands for cloud services
and in particular mobile services. For the first, we will mathematically calculate the
Cournot equilibrium price and the stability constraints. While for inelastic demand, we
140
will simulate the market behavior using ABM.
4.5.1 Oligopoly of Cloud Computing: elastic demand use case
In this section, we will use the mathematical notation of [244]. Let the inverse demand
function be p = f (Q) Where p represents the price, f the inverse demand function and Q
the market supply.
The market supply Q can be broken into the sum of each supplier’s quantity denoted
by Q = Where n represents the number of suppliers, and qi represents their relative
Pn
i=1 qi
supplied quantity. Accordingly, the residual supply can be represented by Qi = Q − qi .
The demand function of telecommunication services is a constant elasticity demand

function [244] which results in a linear inverse demand function. Using the terms of
[244], p = a − bQ where a and b are two positive constants. It was reported by Laffont and
Tirole [244] that details about the cost function of mobile operators are absent.
4.5.1.1 Limitless-capacity suppliers
Assuming a linear cost function, it can be represented by Ci = ci × qi where ci are constant

marginal, equal to average variable costs. The assumption of linear cost function is
verified in section c below. Based on the above assumptions, the reaction function is:

 a−ci

− 21 Qi ; Qi < a−ci
2b b
qi =
0; elsewhere


Considering the equilibria of the general Cournot model, we reach the following
equations:
a − ci
qi = −Q
b
And
141
n a−c
Q=
n+1 b
Where
n
1X
c= ci
n i=1
To calculate the stability of the Cournot equilibria, we need to solve the following n
by n Jacobian Matrix
 
 0 −0.5 −0.5 −0.5
 
−0.5 0 −0.5 −0.5


J =




 ... −0.5 ... −0.5

 
−0.5 −0.5 −0.5 0
The solution of this matrix in its current form calculates the stability for a homo-
geneous limitless-capacity oligopoly which is not the aim of this work. This matrix will be
utilized in Section 4.5.1.5 to calculate the stability of a mixed capacity environment.
4.5.1.2 Limited-capacity suppliers
Legacy operators are limited-capacity suppliers, but have already provisioned their
infrastructure to accommodate the current traffic size. We can thus assume that with
the introduction of cloud-based mobile operators, legacy mobile operators will run in the
range of constant marginal costs without reaching the jump to infinity when the limit is
reached. Assuming a constant cost function, it can be represented by ci = Ci /qi where Ci
is the running cost of operator i. Based on the above assumptions, the reaction function
is:

 a

− 12 Qi ; Qi < a
2b b
qi = 
 0; elsewhere
142
Considering the equilibria of the general Cournot model, we reach the following
equations:
a
qi = −Q
b
And
m a
Q=
m+1b
a2
For the supplier to remain profitable, p × qi > ci then ci < b(m+1)2
To calculate the stability of the Cournot equilibria, we need to solve the following n
by n Jacobian Matrix
 
 0 −0.5 −0.5 −0.5
 
−0.5 0 −0.5 −0.5


J =




 ... −0.5 ... −0.5

 
−0.5 −0.5 −0.5 0
4.5.1.3 Mixed capacity environment
It is safe to represent the mobile market during the transition period when cloud-based
mobile operators join the market as a mixed capacity oligopoly where legacy operators
with their existing infrastructure represent the limited-capacity suppliers and the cloud-
based operators represent the limitless-capacity suppliers. Legacy operators have nearly
fixed costs when the supply is within their capacity and peaks when new infrastructure
investment is required. As this peak increases, the jump in cost can tend to infinity and
thus satisfy the definition of limited-capacity suppliers.
The cost function of limited-capacity mobile operators can be represented either

with a piecewise function with a fixed cost below supply capacity and infinity otherwise or
with constant elasticity of substitution (CES) function. As in the case of most technology-
143
based service provides, the CES function is not a good representation of its cost function,
thus a piecewise function of 2 steps will be used to represent legacy mobile operators.
Cloud-based operators, like all cloud clients, can be represented by a linear cost function
as a simplification of the granular incremental piecewise cost function where the range of
each step equals the capacity of 1 virtual server.
4.5.1.4 Exit conditions in mixed capacity environment
For limitless-capacity suppliers, the price is p = a

n+1
+ n
n+1
C. The suppliers will remain
profitable while P × qi > ci × qi . Neglecting the opportunity cost, the supplier will remain
profitable while a > ci . The profitability is independent from the quantity sold, thus the
cloud service provider will remain profitable and will only exit the market if he acquired
0 customers.
For limitled-capacity suppliers, the price is p = a

n+1
. The suppliers will remain
profitable while P × qi > ci . Neglecting the opportunity cost, the supplier will remain
profitable while qi > ci
a
.
The profitability is dependent on the quantity sold, thus, a legacy mobile operator
will exit the market if he maintained a number of customers less than or equal to ci
a
.
4.5.1.5 Stability of Cournot equilibrium in mixed capacity environment
Studying the long-run stability of the mixed process, assuming that n legacy operators
exist in the market (n limited-capacity suppliers) and m cloud-based operators joining
the market (m limitless-capacity suppliers) we can compose the Jacobian for the mixed
system as an (n + m) by (n + m) matrix:
 
 0 −0.5 −0.5 −0.5
 
−0.5 0 −0.5 −0.5


J =




 ... −0.5 ... −0.5

 
−0.5 −0.5 −0.5 0
144
Luckily enough, the resulting matrix is similar to the limited-capacity and limitless-
capacity cases and its eigenvalues are:
1 1
λ1 ...λn+m−1 = ; λn+m = − (n + m − 1)
2 2
For the system to be in stability, all the eigenvalues should be in the unit circle of
the complex plan which is always true for all the eigenvalues except the last. For λn+m to
fall in the unit circle of the complex plane, n + m < 3. This means that the market will
only be stable if less than 3 operators remain in the market.
4.5.2 Oligopoly of Cloud Computing: inelastic demand use case
In this section, the user is considered to have full market exposure without differentiation
between operator service (Rationality) while neglecting the transfer cost (the cost and
inconvenience to move from one operator to another). Operators can reject sharing the
Porting Authorisation Code [[267], [268], [269]], thus, preventing their customers from
switching to new operators while maintaining their phone numbers. This inconvenience
can be represented by a transfer cost which reduces the customer’s willingness to migrate
between operators. In this section, we consider the user to have a complete willingness to
move between operators. For these reasons, discrepancies in market price are usually
punished and an equilibrium price is expected to be achieved. The relative source code is
listed in the appendix and the simulation results are shown below.
In the absence of a demand function, the operators can’t predict their best move
and consequently, the overall behavior is unstable. Under oligopolistic pricing schemes,
the cloud-based mobile operators will remain in the market if the selling price is higher
than their customer price. On the other hand, the legacy mobile operators will remain
in business if the selling price is higher than their cost divided by the number of their
customers.
As competitive pricing push the prices towards the marginal cost (and in our
145
4.6. CONCLUSION AND FUTURE WORK
case to the operating cost since we neglect the opportunity cost), if we consider that all
legacy mobile operators have the same operational cost and similarly for the cloud-based
operators, we will end up with one type of operators remaining (Figure 4.1 and Figure
4.2). If the operational costs are different, the market will develop into a monopoly with
the operator having the lowest operational cost remaining (Figure 4.3).
Figure 4.1 – Oligopolistic equilibrium with legacy operators out of business (cloud-based operators
have the same operational cost which is lower than that of legacy operators for the considered
number of clients).
4.6 Conclusion and Future Work
In this chapter, we raised the issue of oligopoly in the mobile network especially during the
transition phase when cloud-based mobile operators emerge. We mathematically modeled
the market behavior under the assumption of elastic demand. Cournot Equilibrium
and its stability were calculated for the mixed environment of limitless suppliers (cloud-
146
Figure 4.2 – Oligopolistic equilibrium with cloud-based operators out of business (cloud-based
operators have the same operational cost which is higher than that of legacy operators for the
considered number of clients).
based mobile operators) and limited-capacity suppliers (legacy mobile operators). We

also modeled the 5G network using the Agent-Based-Modeling technique and the market
behavior has been predicted for inelastic demand. The equilibrium and stability of the
mobile market can’t be ensured neither for elastic nor inelastic demand. The potential
states have been presented in this chapter.
The outcomes of this chapter reinforce the outcomes of the previous chapter, which
can be summarized as:
• Any DSS for cloud adoption should not rely solely on the business’ requirements, but
consider the market behavior in general and its possible economic implications.
• Businesses’ actions towards cloud computing are active during the cloud’s first stages
but become reactive with its introduction to the market.
147
Figure 4.3 – Oligopolistic equilibrium ending up in monopoly.
This chapter covers one of the gaps faced in the previous chapter which is the
oligopolistic nature of the mobile market, but it introduced many more challenges that
are to be faced in our future work. Cournot’s oligopoly model is equipped with additional
tools such as investment and time series that help users predict the market behavior as
new players join the market over time. We have neglected these tools in this work, but
are considered as future work for their importance.
We considered in this chapter Cournot’s oligopoly model without taking into con-
sideration Betrand’s oligopoly (although many economists refuse to call it so) or other
models that could be more convenient for the mobile and cloud markets and disruptive
technologies in general.
We have also considered in this chapter that inter-operator and intra-operator

traffic have equal cost because of the mathematical complexity representing the more
generalized form. Generalizing this work to include per-operator pricing is also important
148
for the advancement of this field.
This chapter satisfies CBOD’s task 5 and objectives 4, 5 and 6. The updated project
status is shown in table 4.1.
Status Status
Objective 6 Done Task 6 Pending
Appendix
globals [
cout ;; Used as the xcor value for the first Mobile Service Provider.
counter ;; Used to index the IDs of the clients. The value is set to 0 in
setup.
counter-DMSP-CMSP ;; Used to index the IDs of the Cloud-based Mobile Service
Providers and Dominant Mobile Service Providers.
General_Cost_DMSP ;; c = 1000
General_Cost_CMSP ;; c = 200
constant_a ;;market’s inverse demand function constant a
constant_b ;;market’s inverse demand function constant b
Number_of_active_CMSPs ;; n
number_of_changes
average_cost
CMSP_count
DMSP_count
List_of_active_MSPs
Number_of_active_MSPs
149
breed [clients client] ;; Breed of clients.

breed [CMSPs CMSP] ;; Breed of CMSPs.
breed [DMSPs DMSP] ;; Breed of CMSPs.
clients-own
[
cost ;; Used to represent the cost the company is paying for its IT services
that could be offloaded.
id ;; Client’s ID.
connected_to ;; CMSP’s OR DMSP’s ID
knowledge_list ;; list of CMSPs and DMSPs the user knows
]
CMSPs-own
[
price ;; Used to represent the price the CMSP is offering its client. Price
= ( general_cost * ( 1 + coefficient-of-price ) / (
Number_of_active_CMSPs * CMSP_users ) )
CMSP_users ;; Number of users connected to this CMSP
id ;; CMSP’s ID. Starts at 10.
entry_period ;; When should enter business. Random number between 0 and 2*
Average-entry-time
in_business ;; Boolean value. If tick > entry_period and CMSP_users > 0
Elacticity_starts ;; Before elasticity, CMSP_users will be Initial_momentum;
Elasticity_starts = entry_period + Number-of-rounds-before-elasticity
]
DMSPs-own
[
150
price ;; Initially it is set to Initial-price-of-DMSP; Used to represent the

price the DMSP is offering its client. P = general_cost / DMSP_capacity
DMSP_users ;; Number of users connected to this user
id ;; CMSP’s ID. Starts at 1
in_business ;; Boolean value. If DMSP_users > 0
]
to fill_List_of_active_MSPs
;;
;; Identify the active MSPs so that a user can search for a suitable price
between the active ones. It will decide the searched list based on the
degree-of-knowledge
;;
set List_of_active_MSPs []
ask DMSPs [if in_business = 1 [set Number_of_active_MSPs
Number_of_active_MSPs + 1
ask CMSPs [if in_business = 1
[set Number_of_active_MSPs Number_of_active_MSPs + 1
;;print "␣Number␣of␣active␣MSPs"
;;print Number_of_active_MSPs
if Number_of_active_MSPs > 0
[
;;
;; Fill a knowledge list i.e. the list of MSPs the user contact looking for
the best offer. This is because of bounded rationality
151
;;
ask clients [
;;
;;Currently the user selects a new knowledge-list every tick, and the
new knowledge-list may not include his current operator. The best way
to be done is to keep the current operator in the list
;;since the user will cover all operators after certain number of ticks
;;
ifelse start_of_knowledge + ( Number_of_active_MSPs * Degree-of-knowledge)
< ( Number_of_active_MSPs )
[set knowledge_list sublist List_of_active_MSPs start_of_knowledge ( round
(start_of_knowledge + (Number_of_active_MSPs * Degree-of-knowledge)) )
]
let subl2 sublist List_of_active_MSPs 0 (Number_of_active_MSPs * Degree-
of-knowledge - (Number_of_active_MSPs - start_of_knowledge ) )
]
]]
end
to setup
clear-all
;;ask patches [set pcolor white]
set-default-shape clients "person"
set-default-shape CMSPs "house"
set-default-shape DMSPs "house"
set cout (100 / (Number-of-CMSPs + Number-of-DMSPs + 1) ) ;; It is a dynamic
value for ycor which is the position of the first CMSP. Other CMSPs will
be shifted by a dynamic value called space
152
set counter 0 ;; Used to index the IDs of the clients

set counter-DMSP-CMSP 1 ;; Used to index the IDs of the DMSPs and CMSPs
set General_Cost_DMSP 1000
set General_Cost_CMSP 200
set constant_a 1000
set constant_b 20
create-DMSPs Number-of-DMSPs [
let space ( 90 / ( Number-of-CMSPs + Number-of-DMSPs ) ) ;; Dynamic value
for CMSP locations
set size 5
set price Initial-price-of-DMSP

set DMSP_users 0
set in_business 1


set xcor 70
set ycor cout
]
153
set DMSP_count 0
create-CMSPs Number-of-CMSPs [
let space ( 90 / ( Number-of-CMSPs + Number-of-DMSPs ) ) ;; Dynamic value
for CMSP locations
set size 1
set entry_period random (2 * Average-entry-time )

set in_business 0
set Elacticity_starts ( entry_period + Number-of-rounds-before-elasticity
)
set CMSP_users Initial_momentum
set price 0


set xcor 70
set ycor cout
]
154

set cout (100 / (number-of-Clients + 1) )

;;print list_of_ids
;;print List_of_active_MSPs
create-clients number-of-Clients [
set size ceiling ( 20 / number-of-Clients) ;; be easier to see
let space (90 / number-of-Clients)
set cost 100000

set id counter
set counter counter + 1
set connected_to -1
set xcor 20
set ycor cout
;;
;; Fill a knowledge list i.e. the list of MSPs the user contact looking for
the best offer. This is because of bounded rationality
;;
ask clients [
155
ifelse start_of_knowledge + ( Number_of_active_MSPs * Degree-of-knowledge)

< ( Number_of_active_MSPs )
[set knowledge_list sublist List_of_active_MSPs start_of_knowledge ( round
(start_of_knowledge + (Number_of_active_MSPs * Degree-of-knowledge)) )
]
let subl2 sublist List_of_active_MSPs 0 (Number_of_active_MSPs * Degree-
of-knowledge - (Number_of_active_MSPs - start_of_knowledge ) )
]
]]
ask CMSPs [ set plabel (word (precision price 2) "␣-␣" Good_Reputation)]

ask DMSPs [ set plabel (word (precision price 2) "␣-␣" Good_Reputation)]
ask clients [ set plabel precision cost 2 ]
reset-ticks
end
to exiting_business
if ( ticks > 3 ) [
ask DMSPs [ set in_business 0 ]
set DMSP_count 0

[ ?1 ->
let DMSP_connnected_to [connected_to] of ?1
ask DMSPs with [id = DMSP_connnected_to][set in_business 1]
]
156
ask DMSPs
[
if in_business = 1
[
set DMSP_count DMSP_count + 1
]
]
]
set DMSP_count 0
end
to Join_business
ask CMSPs
[
if ( ticks > entry_period - 1 ) AND ( ticks < Elacticity_starts )
[
set in_business 1
print id
set price initial_momentum
]
]
end
to update_pricing
ask DMSPs with [ in_business = 1 ] [ set Number_of_active_MSPs
Number_of_active_MSPs + 1 ]
157
;;Every operator in business has to check whether increasing his price or

decreasing it up to 100 units would increase his profits assuming all other
operators are constant.
;;The price should be based on the old prices.
let Total_Q 0
ask clients [ set Total_Q (Total_Q + 1) ]
ask CMSPS with [in_business = 1]

[
let q 0
ask clients with [ connected_to = id ] [ set q q + 1 ]
set CMSP_users q
print q
ifelse CMSP_users > 0
[
[
ifelse (constant_a - constant_b * (q + Total_Q + 1)- General_Cost_CMSP
) > 0
[set price ( constant_a - constant_b * ( Total_Q + 1 ))]
[ ifelse (constant_a - constant_b * (q + Total_Q - 1) +
General_Cost_CMSP) > 0
[set price ( constant_a - constant_b * ( Total_Q - 1 ))]
[set price ( constant_a - constant_b * ( Total_Q ))]]
]
]
[
[
ifelse (constant_a - constant_b * (q + Total_Q + 1) - General_Cost_CMSP
) > 0
158

[set price ( constant_a - constant_b * ( Total_Q ))]
]]]
ask DMSPs with [in_business = 1]

[
let DID id
let q 0
ask clients with [ connected_to = DID ] [ set q q + 1 ]
set DMSP_users q
print q
ifelse DMSP_users > 0
[ifelse (constant_a - constant_b * (q + Total_Q + 1) ) > 0
[ ifelse (constant_a - constant_b * (q + Total_Q - 1)) > 0
[set price ( constant_a - constant_b * ( Total_Q - 1 ))]
[set price ( constant_a - constant_b * ( Total_Q ))]]]
[ifelse (constant_a - constant_b * (q + Total_Q + 1) - General_Cost_CMSP)
> 0
[set price ( constant_a - constant_b * ( Total_Q ))] ]
]
end
to go
;;Update list of active MSPs

;;Linking users
;;Update pricing
update_pricing
;;Linking users
159
;;Anybody wants to go out of business

exiting_business
;;Anybody to join the market
Join_business
ask CMSPs [set plabel (word (precision price 2) "␣-␣" Good_Reputation)]

ask DMSPs [ ifelse in_business = 1 [ set plabel (word id "␣-" (DMSP_users) "
␣-␣" price)] [set plabel "" ] ]
ask clients [ set plabel ( word (precision cost 2 ) "␣-␣" connected_to )]
tick
end
to dynamic-revised-pricing ; Business Model
let read_CMSP_price 0
let read_CMSP_id 0
let read_DMSP_price 0
let read_DMSP_id 0
ask clients[
let read_client_connected_to connected_to

let read_client_id id
let read_client_cost cost
;;let read_client_client_capacity 1
let klist knowledge_list
;let node1 end1
let node1 one-of clients with [ id = read_client_id]
let minimum_price 0
;;let Reputation_of_the_CMSP_offering_minimum_price 0
160
let id_of_CMSP-with_minimum_price 0
;;let Reputation_of_the_CMSP_connected_to 0

[ ?1 ->
if read_client_connected_to = read_CMSP_id ; Updating the cost

[
;;set Reputation_of_the_CMSP_connected_to read_CMSP_Good_Reputation
set read_client_cost read_CMSP_price ; the client has to pay for

his resourcesBased on the new price the CMSP is offering
set cost read_CMSP_price ; Both increasing and decreasing costs
]
]
[ ?1 ->
if read_client_connected_to = read_DMSP_id ; Updating the cost

[
;;set Reputation_of_the_CMSP_connected_to read_CMSP_Good_Reputation
set read_client_cost read_DMSP_price ; the client has to pay for

his resourcesBased on the new price the CMSP is offering
161
set cost read_DMSP_price ; Both increasing and decreasing costs

]
]
set minimum_price read_client_cost

set id_of_CMSP-with_minimum_price -1
Reputation_of_the_CMSP_connected_to
ifelse ticks = 1 [
ask CMSPs
[
set read_CMSP_id id
foreach klist
[
?2 ->
[
] ] ]
ask DMSPs
[
set read_CMSP_id id
foreach klist
[
162
?2 ->
[
]
]
]
]
[
ask CMSPs
[
set read_CMSP_id id
if (minimum_price * (1 - RaS) * (RoM) > read_CMSP_price)

[
foreach klist
[ ??1 ->
let xid ??1
if read_CMSP_id = xid
[
set minimum_price read_CMSP_price
set id_of_CMSP-with_minimum_price read_CMSP_id
]]]]
ask DMSPs
[
163
set read_DMSP_id id
if (minimum_price * (1 - RaS) * (RoM) > read_DMSP_price)

[
foreach klist
[ ??1 ->
let xid ??1
if read_DMSP_id = xid
[
set minimum_price read_DMSP_price
set id_of_CMSP-with_minimum_price read_DMSP_id
]]]]]

[ ?1 ->
if (id_of_CMSP-with_minimum_price = read_CMSP_id) ; User found

better price
[
;show connected_to

[
164
ask links with [ color = 12 + read_client_id] [die] ; Disabling the

connection with the old CMSP
]
set connected_to read_CMSP_id ; Connecting to the new CMSP
set cost read_CMSP_price
let node2 one-of CMSPs with [ id = read_CMSP_id]
ask node1 [ create-link-with node2 [ set color 12 + read_client_id] ]
;ask CMSPs with [id = iCMSP] [ create-link-with end1 [set color 12 +

idinput + idinput]]
]
]

[ ?1 ->
;;set read_DMSP_capacity [DMSP_users] of ?1
if (id_of_CMSP-with_minimum_price = read_DMSP_id) ; User found

better price
[
;show connected_to

[
ask links with [ color = 12 + read_client_id] [die] ; Disabling the

connection with the old CMSP
165
]
set connected_to read_DMSP_id ; Connecting to the new CMSP
set cost read_DMSP_price
let node2 one-of DMSPs with [ id = read_DMSP_id]
ask node1 [ create-link-with node2 [ set color 12 + read_client_id] ]
;ask CMSPs with [id = iCMSP] [ create-link-with end1 [set color 12 +

idinput + idinput]]
]
]
]
end
166
Chapter 5
General Discussion
Contents
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.2 Impact on Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.3 Impact on Managerial Practices . . . . . . . . . . . . . . . . . . . . . . 171
5.4 Future Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.4.1 Simulation using system dynamics . . . . . . . . . . . . . . . . . . 173
5.4.2 Optimizing the recommender system . . . . . . . . . . . . . . . . . 173
5.4.3 Investment and time series . . . . . . . . . . . . . . . . . . . . . . . 174
5.4.4 Oligopoly models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5.1 Introduction
Organization’s success is based on the correctness of its decisions. Correct decisions

require accurate and relevant information. Too much data and the decision-maker will
fail to detect relevant information and thus take a mal-informed decision.
Technology is a driving force for current businesses and is among the key deci-
sions and investments directly supervised by senior executives. Large organizations are
167
5.1. INTRODUCTION
searching the industry for technologies developed by startups and small companies that
can complement their arsenal of services and technologies. Those startups and small
companies receive generous offers for acquisition in the hope that their technologies offer
market superiority for the buyer. Selecting winner technologies is a tedious task, contains
many ambiguities and variables, and is relatively not very accurate.
Epistemologically, the organization cannot know a winner technology ahead of

time but can use existing evidence, predictions, statistics, surveys, and other techniques
to justify its belief in the correctness of its decision. Due to the sheer number of emerg-
ing technologies, Business executives try to assess hyped emerging technologies, they
sometimes overlook disruptive technologies.
Cloud computing is typically one of those disruptive technologies that divide the
market between adopters vs. non-adopters. Choosing whether to adopt cloud computing
must consider its impact on the organization’s success which, in turn, depends on internal
and external factors to the organization. The existing literature addressed organizational
cloud computing adoption using two distinct approaches. The first one considered the
internal factors to help an organization select the most suitable type of cloud computing,
while the second approach took into account the external factors that set the market price
and service standard. This thesis provides in-depth coverage of both approaches.
In this thesis, we tried to generate automated tools that help decision-makers

decide on the adoption of disruptive technologies and more specifically cloud computing.
The generated tools offer two portals, the first portal hides the organization’s internal
factors, shows the most relevant information, and recommends decisions based on the
organization’s needs. The second portal hides the external (economic) factors, shows the
most relevant information, and recommends decisions based on the predicted economic
trend.
This work was intentionally designed to be modular and applicable to other disrup-
tive and emerging technologies. As shown in Section 1.3 of Chapter 1 (general discussion),
the need for studying the corporate adoption of blockchain is very relevant and is expected
to be in high demand next year. The contributions of this work can be easily migrated to
the blockchain which makes this work very relevant and one of the very first, the first to
168
5.2. IMPACT ON LITERATURE
the best of our knowledge, to cover the economic factors of blockchain adoption.
This work has yielded the first open-source DSS, to the best of our knowledge, to
study the economic factors affecting cloud computing adoption and predicted its economic
trend. Additionally, this work is the first to use a collaborative filtering recommender
system to match experts’ answers concerning cloud computing adoption. This approach
yields considerable accuracy enhancement compared to the approaches found in the
literature.
This work is also the first to study the market risks resulting from the introduction
of multitenancy in 5G mobile networks. It also paved the way for policy studies regulating
incoming 5G networks.
This chapter rediscusses the main findings and contributions of this work in a more
general context and transversally summarizes them as follows. Section 5.2 analyzes how
this contribution adds to the literature. Section 5.3 analyzes how this contribution adds
to managerial practices. Lastly, Section 5.4 describes some operational future research
directions.
5.2 Impact on Literature
Cloud computing can be represented by a three-layer business model framework. Each

layer is related to one of its core services which are SaaS, PaaS, and IaaS. More advanced
and complex business models evolved to respond to the new features of cloud computing.
Like all disruptive technologies, it is susceptible to failures and downtime. Thus, consid-
ering adopting it involves risks, sophisticated contracts, and complex decision-making
processes. Adopting cloud computing or other disruptive technologies is a core decision
that should be taken at the highest corporate levels because it can lead to unexpected
losses, if not bankruptcy. The severity of the decision’s consequences and the value-at-risk
motivated researchers to help decision-makers select the suitable technology.
The literature is divided into two distinct methodologies for deciding on the adoption
169
5.2. IMPACT ON LITERATURE
of cloud computing. The first methodology relies on questionnaires sent to experts and
decision-makers to extract a set of determinants for cloud computing adoption. The second
methodology relies on modeling the market dynamics and predicting the success of the
adoption.
In the first methodology, researchers tried to identify all the determinants that
affect the cloud’s business models in order to design a realistic model for a viable process.
Including all the determinants in a business model is not a guarantee for success. The
second methodology considers that strategies and clients’ business models, suppliers, and
competitors’ processes are very important in deciding the success of the user’s strategy.
For this reason, simulating the interaction between business models is important to
fine-tune the designed model and decrease failure risk by preventing any miscalculated
behavior.
As shown in Figure 1.1 (Cloud adoption literature), the first approach reached a
maturity level without being able to overcome its weaknesses which are implementation
weaknesses due to limitations in the determinant’s selection, questionnaire distribution
and rigidity of the outcomes and Conceptual weaknesses due to the nature of the method-
ology such as: experts’ biased recommendations, decision-makers’ biased answers and
corporate officials’ biased perception to the used adoption model or determinants.
Biases influence the experts’ recommendations, decision-makers’ answers, and

corporate officials’ perception of the used adoption model or determinants. We next
provide a couple of examples that highlight the different factors at play for choosing the
right IT-based strategy for a business. For instance, CEOs with fixed salaries have less
tendency to migrate to the cloud while CEOs with shares have a higher tendency to adopt
this technology. In other circumstances, we notice that Strategic IT Alignment’s (SITA)
interaction with IT Investment (ITI) is negative. SITA’s unmalleability lowers the good
influence of ITI on firm performance in stable, uncomplicated, and abundant surroundings.
Based on the abovementioned concerns, the second methodology is becoming dominant
since 2017.
The second methodology considers that strategies and clients’ business models,
suppliers, and competitors’ processes are very important in deciding the success of the
170
5.3. IMPACT ON MANAGERIAL PRACTICES
user’s strategy. For this reason, simulating the interaction between business models is
important to fine-tune the designed model and decrease failure risk. Several approaches
were proposed to study the interaction between processes and to predict the success
rate under different environments, such as mathematically modeling the optimal pricing
strategies of a cloud service provider, studying the mathematical equilibrium in a market
for data-based services, or using game theory to compare different selling strategies under
monopolistic and competitive environments.
This work contributed to the literature by solving the implementation weakness

of the first methodology by using a collaborative filtering recommender system. It also
contributed to the expansion of the second methodology by being the first to introduce
Agent-Based Modeling to simulate the market behavior for cloud computing. Additionally,
this work is the first to discuss oligopolistic behavior in cloud computing markets.
5.3 Impact on Managerial Practices
The aim of this work, first and above all, is increasing the manager’s awareness of his
corporation’s capabilities and needs and making sense of the surrounding environment
in terms of cloud computing technologies and capabilities offers.
The decision-making process for selecting a technology to adopt is very complex

since it requires information from different departments and different skillsets. The
decision-maker has to have the solid technical mastery to be able to fully understand an
incoming technology and assess its suitability. He also has to have an economics skillset
and accurate forecasts and analytics on the market to be able to assess the market’s
readiness for the new technology. He also has to have finance exposure and accurate
predictions on the corporation’s finances to be able to decide whether the corporation
can afford it. The decision-making process is a multi-party process, and any inaccurate
information or calculation will result in damaging consequences. Additionally, the decision
requires skillsets unavailable in many SMEs. This work offers managers with limited
technical or economic exposure the ability to decide on the adoption of cloud computing
171
5.3. IMPACT ON MANAGERIAL PRACTICES
and other emerging technologies.
The commercial approach in existing âĂŸCloud Decision Engines’ found in liter-

ature and market may include deliberate and indeliberate biases. Those biases affect
the correctness and fairness of the decisions by favoring one solution over the other or
faithfully holding optimistic expectations. As long as the managers are concerned, our
work provides DSS tools that are open-source, vendor-neutral, and free. Our designed
tools save managers from the subjective evaluation of individual experts/consultants and
replace their subjective evaluations with normalized subjective evaluations that can be
considered objective. Using those tools, the consultants’ role changes from providing
subjective expert (sometimes biased) opinion into helping extend the capabilities of the
tools by adding additional factors, technologies, and features.
The DSS tools provided in this work are transparent, i.e. the logic behind the
decision-making process can be traced and investigated. Consequently, managers can
now take informed decisions powered by the dynamic logic generated in both tools. Trans-
parency is legally required [270] for autonomous systems and shareholders required in
corporate environments.
In addition to transparency, we offer accountability, a feature predicting our confi-

dence in the recommended decision. None of the experts or DSS can confidently recom-
mend technologies /architectures/systems for every faced scenario. Although consultants
are esteemed to be experts and express confidence in their recommendations as a sales
tactic, it is very important to keep the managers aware of the fact that emerging technolo-
gies introduce high margins of uncertainty due to a plethora of technical and economic
factors. Accountability benefits the managers by providing them with indicators to assess
the reliability of a recommendation and the need for additional opinion.
Ambiguity resulting from low-confidence recommendations or conflicting ones

should automatically flag the need for deep investigation. Such case usually happens
because of wrong evaluation of organizational needs. It is well-known that technologies
and architectures provide a compromise between different needs. Overdemanding can
confuse the consultants and DSS alike. In the case of low-confidence recommendations, the
investigation should start with ensuring the correctness of the organizational needs. As
172
5.4. FUTURE RESEARCH DESIGN
highlighted earlier, a major contribution to the proposed tools is increasing the manager’s
awareness of his corporation’s capabilities and needs and making sense of the surrounding
environment. Finally, low-confidence recommendations can pinpoint errors in reports
generated by different teams in the organization.
5.4 Future Research Design
The contributions of this manuscript can still be developed in multiple directions and we
are planning to pursue the following:
5.4.1 Simulation using system dynamics
We modeled the market using Agent-Based Modeling which is very efficient in representing
the interactions between agents. Complementing this work with a system dynamics
simulation would allow us to build a much better understanding of market behavior.
5.4.2 Optimizing the recommender system
As discussed in the limitations, the proposed recommender system satisfies the require-
ments we set in Chapter 2, but it hasn’t been evaluated for accuracy against other
recommender systems. We are planning to work on optimizing the design of the used
algorithm and hopefully overcome the cold-start problem even with small datasets.
173
5.4. FUTURE RESEARCH DESIGN
5.4.3 Investment and time series
We are planning to mathematically study the emergence of new suppliers over time
using time series instead of evaluating the market as a closed network of suppliers and
customers.
5.4.4 Oligopoly models
In this work, we only considered Cournot’s oligopoly model although not very suitable
for representing our use case. Other models are available, such as Bertrand’s duopoly
model. It would be much more enriching if other models are considered and benchmarked
against the Agent-based modeling and System dynamics simulations.
174
Chapter 6
Conclusion
Contents
6.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.1.1 Generic and vendor-neutral decision support solution for cloud

computing adoption . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.1.2 Modeling 5G networks under perfect competition conditions . . . 178
6.1.3 Modeling 5G networks under oligopolistic conditions . . . . . . . . 179
6.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.2.1 Cold-Start problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.2.2 Recommender system . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.2.3 Parameter fine-tuning . . . . . . . . . . . . . . . . . . . . . . . . . . 180
Disruptive technologies, including cloud computing, are double-edge swords. Many

organizations adopted the new disruptive technologies to boost their revenues and suc-
ceeded such as Amazon’s cloud services and Kaspersky Lab’s Next Generation Anti-Virus.
Others failed to adopt and lost market share such as Nokia’s failure to adopt smartphone
technology and finally some organizations adopted disruptive technologies that turned
out to be a bubble such as solar cars.
The common factor between all the above examples is that they all considered their
organizational factors when deciding on the adoption of the respective disruptive technol-
ogy. The factors deciding the decision’s success or failure were outside the organizational
175
boundaries especially that all the above names are major players in their respective
markets. Based on this observation, we studied in this dissertation the organizational as
well as the economic factors affecting cloud computing’s corporate adoption.
Selecting winner technologies is a tedious task, contains many ambiguities and

variables, and is relatively not very accurate. The literature still misses a successful
formula for predicting the future of a technology and organizational learning leads to
organization-specific decision schemes which lack community testing and evaluation and
thus potentially not very accurate. Epistemologically, the organization cannot know a
winner technology ahead of time, but can use existing evidence, predictions, statistics,
surveys, and other techniques to justify its belief in the correctness of its decision. Very
successful corporations, such as Google and Microsoft, invested in and developed mul-
tiple services and technologies that turned to be a complete failure and later ended in
suspension.
Almost all emerging technologies will not endure time and market competition.
Their impact and market penetration will be marginal and will eventually fade away.
Unlike lawmakers, business leaders must assess all emerging technologies and devote
their very limited resources to study any potential technology. Since early adopters achieve
the highest revenue rate, managers can be overwhelmed with the new technologies and
this can lead them to neglect real disruptive technologies, and therefore, to become late
adopters. While late adoption of emerging technologies removes many of the inherent
risks associated with them, a significant part of the profit margin will also be eliminated.
Cloud computing is a distinctive disruptive technology that divides the market into
two parts: adopters and non-adopters. Depending on the type of cloud used, the adopters
themselves can be evenly divided. This important division is success versus failure. The
decision of an organization to adopt or not to adopt cloud computing should be based on
the impact of this technology on the success of that organization. Several internal and
external factors influence this success, therefore, adopting a specific technology is not a
trivial decision, it depends on future opportunities and market dynamics, and not just on
the internal state of the organization.
The impact of economic factors on the success of corporate decisions is consistent
176
across all emerging technologies, thus corporations should follow an exocentric approach to
decision-making. Exocentric approach is more difficult than endocentric because the latter
relies on internal historical data, internal processes, and internal needs. Corporations
are usually very efficient in accessing and analyzing internal factors. Exocentric approach
requires external historical data, knowledge of competitors’ intentions, new player’s
intentions and most importantly suppliers’ plans. Corporations usually find the analysis
and extraction of complicated trends from external data difficult, resource consuming and
very sensitive to random factors not included in the observations.
Missed factors, even negligible, can lead to radical changes in the predictions fol-
lowing the butterfly effect. This makes any designed model in a continuous enhancement
loop. Artificial Intelligence (AI) and Machine Learning (ML) gained momentum in nearly
every field, but not in our case. Training effective AI and ML models require a relatively
large training dataset which we miss in this case. For this reason, we decided to design
a parameterized model relying Agent-Based Modelling and economics equations. Each
decision maker can insert the parameters representing the market situation he is facing.
The designed model definitely requires loops of enhancement to become more

accurate as will be shown next.
This chapter recapitulates the main findings and contributions and is organized as
follows. Section 6.1 indicates the important contributions and relevant implications of
this thesis, while Section 6.2 lists some of this work’s limitations.
It also acts as the result summarization and dissemination as requested in CBOD

task 6. The updated project status is shown in table 6.1.
Status Status
177
6.1. CONTRIBUTIONS
6.1 Contributions
6.1.1 Generic and vendor-neutral decision support solution for cloud

computing adoption
Considering both organizational and economic factors in a corporate decision is not a trivial
task especially that market trends differ from our mental models. We solved this challenge
by proposing an RS based DSS since one of RS’s strength is extracting trends from datasets
and user interactions. These trends are very important for providing decision-makers
with an insight into the external factors. Interestingly enough, Collaborative Filtering
Recommender Systems profile similar organizations and utilize the trends only from the
coherent group. We consider this DSS to be the first generic and vendor-neutral decision
support solution for cloud computing adoption.
6.1.2 Modeling 5G networks under perfect competition conditions
In this contribution, we simulated two categories of organizations, one didn’t adopt cloud
computing (legacy mobile operators), and the second adopted cloud technology (cloud-
based mobile operators). The adoption decision is not related to internal organizational
factors. The simulation showed that the market behavior (outcome) could result in:
1. All legacy mobile operators driven out of the market with the final user cost exceeding
the initial cost.
2. All legacy mobile operators driven out of business with the final user cost less than
the initial cost.
3. Co-existence between legacy and cloud-based mobile operators. Some of the legacy
and cloud-based operators left the market.
178
6.2. LIMITATIONS
4. Cloud-based mobile operators join the market and leave later because of the compe-
tition from the legacy mobile operators.
More important than the number of possible outcomes is that we were able to
prove that economic factors decide whether cloud adoption is beneficial or not. This proof
reinforces our decision of using RS as the engine inside DSS that we designed in this
manuscript.
6.1.3 Modeling 5G networks under oligopolistic conditions
In this contribution, we evaluated the transition phase from a stable market (in oligopoly
terms) into a disrupted market (when disruptive technologies emerge). We considered two
cases: elastic and inelastic demands for cloud services. For the first case, the market sta-
bility was modeled mathematically where legacy mobile operators have a limited capacity
inherited from the nature of their infrastructures; while cloud-based mobile operators
are free from this constraint. The market behavior of the second case was modeled using
Agent-Based-Modeling. In addition to the outcomes of the previous contribution, we
concluded that Businesses’ actions towards cloud computing are active during the cloud’s
first stages, but become reactive with its introduction to the market.
6.2 Limitations
6.2.1 Cold-Start problem
Collaborative Filtering has a well-known problem called the cold-start problem that
affects the accuracy of the recommender system when having a very small dataset. This
problem is automatically overcome as the dataset grows. For this case, the DSS has to go
through a test phase for collecting answers from practitioners.
179
6.2. LIMITATIONS
6.2.2 Recommender system
The proposed recommender system is not necessarily the best algorithm available, but it
has been developed to cover the requirements of the DSS. Further investigation in the
computer science discipline can help develop a more optimized algorithm.
6.2.3 Parameter fine-tuning
The ABM simulation is not benchmarked with valid data because the data aren’t available
yet and the simulation’s role is to predict the future market behavior. More investigation
is required to calibrate the simulator equipped with market data.
180
References
[1] Phil Johnson and Joanne Duberley. Understanding management research: An

introduction to epistemology. Sage, 2000.
[2] Jacques Bou Abdo and Sherali Zeadally. Neural network-based blockchain decision
scheme. Information Security Journal: A Global Perspective, pages 1–15, 2020.
[3] Peeyush Tugnawat and Mohamed E Fayad. Advanced peer to peer discovery and
interaction framework. In 18th Annual ACM SIGPLAN Conference on Object-
Oriented Programming, 2003.
[4] Rajkumar Buyya, Chee Shin Yeo, Srikumar Venugopal, James Broberg, and Ivona
Brandic. Cloud computing and emerging it platforms: Vision, hype, and reality
for delivering computing as the 5th utility. Future Generation computer systems,
25(6):599–616, 2009.
[5] Rajkumar Buyya, James Broberg, and Andrzej M Goscinski. Cloud computing:
Principles and paradigms, volume 87. John Wiley & Sons, 2010.
[6] Luis M Vaquero, Luis Rodero-Merino, Juan Caceres, and Maik Lindner. A break in
the clouds: towards a cloud definition, 2008.
[7] Will Forrest and Charlie Barthold. Clearing the air on cloud computing. Discussion
Document from McKinsey and Company, 2009.
[8] Jacques Bou Abdo. Efficient and secure mobile cloud networking. PhD thesis, Paris
6, 2014.
[9] Antonio Celesti, Francesco Tusa, Massimo Villari, and Antonio Puliafito. Three-
phase cross-cloud federation model: The cloud sso authentication. In 2010 Second
181
REFERENCES
International Conference on Advances in Future Internet, pages 94–101. IEEE,

2010.
[10] Rafael Moreno-Vozmediano, Rubén S Montero, and Ignacio M Llorente. Iaas cloud
architecture: From virtualized datacenters to federated cloud infrastructures.
Computer, 45(12):65–72, 2012.
[11] Michael Miller. Cloud computing: Web-based applications that change the way you
work and collaborate online. Que publishing, 2008.
[12] Peter Mell, Tim Grance, et al. The nist definition of cloud computing. 2011.
[13] Surya Nepal, Rajiv Ranjan, and Kim-Kwang Raymond Choo. Trustworthy process-
ing of healthcare big data in hybrid clouds. IEEE Cloud Computing, 2(2):78–84,
2015.
[14] Rayane El Sibai, Nader Gemayel, Jacques Bou Abdo, and Jacques Demerjian.
A survey on access control mechanisms for cloud computing. Transactions on
Emerging Telecommunications Technologies, 31(2):e3720, 2020.
[15] Subashini Subashini and Veeraruna Kavitha. A survey on security issues in service
delivery models of cloud computing. Journal of network and computer applications,
34(1):1–11, 2011.
[16] Tharam Dillon, Chen Wu, and Elizabeth Chang. Cloud computing: issues and
challenges. In 2010 24th IEEE international conference on advanced information
networking and applications, pages 27–33. Ieee, 2010.
[17] Prince Kwame Senyo, Erasmus Addae, and Richard Boateng. Cloud computing
research: A review of research themes, frameworks, methods and future research
directions. International Journal of Information Management, 38(1):128–139, 2018.
[18] Mariya Shyshkina. The hybrid service model of electronic resources access in the
cloud-based learning environment. arXiv preprint arXiv:1807.09264, 2018.
[19] Alexander Lenk, Markus Klems, Jens Nimis, Stefan Tai, and Thomas Sandholm.
What’s inside the cloud? an architectural map of the cloud landscape. In 2009
ICSE workshop on software engineering challenges of cloud computing, pages 23–31.
IEEE, 2009.
182
REFERENCES
[20] Wassim Itani, Ayman Kayssi, and Ali Chehab. Privacy as a service: Privacy-aware
data storage and processing in cloud computing architectures. In 2009 Eighth IEEE
International Conference on Dependable, Autonomic and Secure Computing, pages
711–716. IEEE, 2009.
[21] Jacques Bou Abdo, Jacques Demerjian, Hakima Chaouchi, Kabalan Barbar, and
Guy Pujolle. Broker-based cross-cloud federation manager. In 8th International
Conference for Internet Technology and Secured Transactions (ICITST-2013), pages
244–251. IEEE, 2013.
[22] May Al-Roomi, Shaikha Al-Ebrahim, Sabika Buqrais, and Imtiaz Ahmad. Cloud
computing pricing models: a survey. International Journal of Grid and Distributed
Computing, 6(5):93–106, 2013.
[23] Dan C Marinescu. Cloud computing: theory and practice. Morgan Kaufmann, 2017.
[24] Gonzalo Huerta-Canepa and Dongman Lee. A virtual cloud computing provider for
mobile devices. In proceedings of the 1st ACM workshop on mobile cloud computing
& services: social networks and beyond, pages 1–5, 2010.
[25] Jacques Bou Abdo, Najma Saidani, and Ahmed Bounfour. OrganizationsâĂŹ
investment in cloud computing: Designing a decision support platform. In 14th
International Conference on Data Science (ICDATAâĂŹ18), pages 100–104, 2018.
[26] Christof Weinhardt, Arun Anandasivam, Benjamin Blau, Nikolay Borissov, Thomas
Meinl, Wibke Michalk, and Jochen Stößer. Cloud computing–a classification, busi-
ness models, and research directions. Business & Information Systems Engineering,
1(5):391–399, 2009.
[27] Jacques Bou Abdo, Jacques Demerjian, Hakima Chaouchi, Kabalan Barbar, and
Guy Pujolle. Cloud federation means cash. In The Third International Conference
on e-Technologies and Networks for Development (ICeND2014), pages 39–42. IEEE,
2014.
[28] Jacques Bou Abdo, Jacques Demerjian, Hakima Chaouchi, Rami Yared, and Talar
Atechian. Micro-economy effect on cloud federation. In 2015 Global Summit on
Computer & Information Technology (GSCIT), pages 1–4. IEEE, 2015.
183
REFERENCES
[29] Jacques Bou Abdo, Jacques Demerjian, Hakima Chaouchi, and Talar Atechian.
Enhanced revenue optimizing sla-based admission control for iaas cloud networks.
In 2015 3rd International Conference on Future Internet of Things and Cloud, pages
225–230. IEEE, 2015.
[30] Jacques Bou Abdo, Jacques Demerjian, Hakima Chaouchi, Kabalan Barbar, Guy
Pujolle, and Talar Atechian. Cloud federation? we are not ready yet. In 2014
IEEE Intl Conf on High Performance Computing and Communications, 2014 IEEE
6th Intl Symp on Cyberspace Safety and Security, 2014 IEEE 11th Intl Conf on
Embedded Software and Syst (HPCC, CSS, ICESS), pages 831–834. IEEE, 2014.
[31] Loretta Mastroeni and Maurizio Naldi. Insurance pricing and refund sustainability
for cloud outages. In International Conference on the Economics of Grids, Clouds,
Systems, and Services, pages 3–17. Springer, 2017.
[32] Young Bong Chang, Vijay Gurbaxani, and Kiron Ravindran. Information technology
outsourcing: Asset transfer and the role of contract. MIS Q., 41(3):959–973, 2017.
[33] Maurizio Naldi and Loretta Mastroeni. Economic decision criteria for the migration
to cloud storage. European Journal of Information Systems, 25(1):16–28, 2016.
[34] Maurizio Naldi. Evaluation of customer’s losses and value-at-risk under cloud
outages. In 2017 40th International Conference on Telecommunications and Signal
Processing (TSP), pages 12–15. IEEE, 2017.
[35] Loretta Mastroeni, Alessandro Mazzoccoli, and Maurizio Naldi. Service level
agreement violations in cloud storage: Insurance and compensation sustainability.
Future Internet, 11(7):142, 2019.
[36] Maurizio Naldi. Bagging estimation of availability in public cloud storage. In

2018 UKSim-AMSS 20th International Conference on Computer Modelling and
Simulation (UKSim), pages 163–168. IEEE, 2018.
[37] Matteo Adriani and Maurizio Naldi. Whose fault is it? correctly attributing out-
ages in cloud services. In 2019 Federated Conference on Computer Science and
Information Systems (FedCSIS), pages 433–440. IEEE, 2019.
184
REFERENCES
[38] Maurizio Naldi. Non-parametric bootstrap detection of availability service level

objective violations in cloud storage. In 2018 Network Traffic Measurement and
Analysis Conference (TMA), pages 1–7. IEEE, 2018.
[39] Saskia Bayerl, Kristina Lauche, and Carolyn Axtell. Revisiting group-based tech-
nology adoption as a dynamic process: The role of changing attitude-rationale
configurations. Mis Quarterly, 40(3):775–784, 2016.
[40] Rajiv Sabherwal, Sanjiv Sabherwal, Taha Havakhor, and Zach Steelman. How does
strategic alignment affect firm performance? the roles of information technology
investment and environmental uncertainty. MIS Quarterly, 43(2):453–474, 2019.
[41] Ricky YK Chan and Katherine HY Ma. Impact of executive compensation on

the execution of it-based environmental strategies under competition. European
Journal of Information Systems, 26(5):489–508, 2017.
[42] Adel Ben Youssef, Walid Hadhri, and Téja Maherzi. Adoption of cloud computing
in emerging countries: the role of the absorptive capacity. Systemes d’information
management, 20(4):117–142, 2015.
[43] Sébastien Tran and Emmanuel Bertin. Changing organizational models of it

departments as a result of cloud computing: proposal for a typology. Systemes
d’information management, 20(4):51–87, 2015.
[44] Michael Lang, Manuel Wiesche, and Helmut Krcmar. Criteria for selecting cloud
service providers: a delphi study of quality-of-service attributes. Information &
Management, 55(6):746–758, 2018.
[45] Gongtao Zhang and MN Ravishankar. Exploring vendor capabilities in the cloud
environment: A case study of alibaba cloud computing. Information & Management,
56(3):343–355, 2019.
[46] Erick Leroux and Pierre-Charles Pupion. Modelling cloud computing adoption
in major french local public authorities. Systemes d’information management,
20(4):11–50, 2015.
185
REFERENCES
[47] Mahdi Fahmideh, Farhad Daneshgar, Fethi Rabhi, and Ghassan Beydoun. A
generic cloud migration process model. European Journal of Information Systems,
28(3):233–255, 2019.
[48] Sabine Khalil, Valérie Fernandez, and Valerie Fautrero. Cloud impact on it gov-
ernance. In 2016 IEEE 18th Conference on Business Informatics (CBI), volume 1,
pages 255–261. IEEE.
[49] Valérie Fautrero, Valérie Fernandez, and Sabine Khalil. Le paradigme du cloud
computing: au-delà de nouvelles solutions informatiques, un enjeu de gouvernance
renouvelée des technologies numériques. In Annales des Mines-Gerer et comprendre,
number 3, pages 13–24. FFE, 2018.
[50] Adam Saunders and Erik Brynjolfsson. Valuing it-related intangible assets. MIS
Quarterly, Forthcoming, 2015.
[51] German F Retana, Chris Forman, Sridhar Narasimhan, Marius Florin Niculescu,
and DJ Wu. Technology support and post-adoption it service use: Evidence from
the cloud. MIS Quarterly, 42(3):961–978, 2018.
[52] Philipp Wunderlich, Daniel J Veit, and Saonee Sarker. Adoption of sustainable tech-
nologies: a mixed-methods study of german households. MIS Quarterly, 43(2):673–
691, 2019.
[53] Manuel Trenz, Jan Huntgeburth, and Daniel Veit. Uncertainty in cloud service
relationships: Uncovering the differential effect of three social influence processes
on potential and current users. Information & Management, 55(8):971–983, 2018.
[54] Soumya Sen, Roch Guérin, and Kartik Hosanagar. Shared versus separate networks:
the impact of reprovisioning. In Proceedings of the 2009 workshop on Re-architecting
the internet, pages 73–78, 2009.
[55] Roch Guerin, Kartik Hosanagar, Xinxin Li, and Soumya Sen. Shared or dedicated
infrastructures: On the impact of reprovisioning ability. MIS Quarterly, 43(4):1059–
1079, 2019.
186
REFERENCES
[56] Maurizio Naldi, Marta Flamini, and Giuseppe D’Acquisto. Negligence and sanctions
in information security investments in a cloud environment. Electronic Markets,
28(1):39–52, 2018.
[57] Alok Gupta, Karthik Kannan, and Pallab Sanyal. Economic experiments in infor-
mation systems. MIS Quarterly, 42(2):595–606, 2018.
[58] Zhiling Guo and Dan Ma. A model of competition between perpetual software and
software as a service. MIS Quarterly, 42(1):1, 2018.
[59] Luisanna Cocco, Giulio Concas, and Michele Marchesi. Simulation of the competi-
tion among traditional and on-demand software vendors. Simulation, 92(1):33–45,
2016.
[60] Guofang Nan, Zan Zhang, and Minqiang Li. Optimal pricing for cloud service
providers in a competitive setting. International Journal of Production Research,
57(20):6278–6291, 2019.
[61] Haiyang Feng, Zhengrui Jiang, and Dengpan Liu. Quality, pricing, and release
time: Optimal market entry strategy for new software-as-a-service vendors. MIS
Quarterly, Forthcoming, 2017.
[62] Maurizio Naldi and Marta Flamini. Dynamics of the hirschman–herfindahl index
under new market entries. Economic Papers: A journal of applied economics and
policy, 37(3):344–362, 2018.
[63] Zhe Song, Ziyuan Zhang, Xiaolin Xu, and Chunlin Liu. An agent-based model to
study the market dynamics of perpetual and subscription licensing. Journal of the
Operational Research Society, 66(5):845–857, 2015.
[64] Douwe Postmus, Jacob Wijngaard, and Hans Wortmann. An economic model to
compare the profitability of pay-per-use and fixed-fee licensing. Information and
Software Technology, 51(3):581–588, 2009.
[65] Loretta Mastroeni, Maurizio Naldi, and Pierluigi Vellucci. An agent-based model
on scale-free networks for personal finance decisions. In WOA, pages 77–83, 2019.
187
REFERENCES
[66] Wolfgang Ketter, Markus Peters, John Collins, and Alok Gupta. A multiagent com-
petitive gaming platform to address societal challenges. Mis Quarterly, 40(2):447–
460, 2016.
[67] Luis Guijarro, Vicent Pla, Jose R Vidal, and Maurizio Naldi. Competition in data-
based service provision: Nash equilibrium characterization. Future Generation
Computer Systems, 96:35–50, 2019.
[68] Luis Guijarro, Jose R Vidal, Vicent Pla, and Maurizio Naldi. Economic analysis of
a multi-sided platform for sensor-based services in the internet of things. Sensors,
19(2):373, 2019.
[69] Luis Guijarro, Jose R Vidal, Vicent Pla, and Maurizio Naldi. Wireless sensor
network-based service provision: A three-sided market model. In 2017 13th In-
ternational Conference on Network and Service Management (CNSM), pages 1–7.
IEEE, 2017.
[70] Shantanu Bhattacharya, Andreas Robotis, and Luk N Van Wassenhove. Installed
base management versus selling in monopolistic and competitive environments.
European Journal of Operational Research, 273(2):596–607, 2019.
[71] Bo Feng, Qiwen Ye, and Brian J Collins. A dynamic model of electric vehicle adoption:
The role of social commerce in new transportation. Information & Management,
56(2):196–212, 2019.
[72] Thomas J Housel and Sarah K Nelson. Knowledge valuation analysis: Applications
for organizational intellectual capital. Journal of Intellectual Capital, 2005.
[73] José Cintrón, Luis C Rabelo, and Thomas J Housel. Estimating the knowledge
value-added of information technology investments. In IIE Annual Conference.
Proceedings, page 1849. Institute of Industrial and Systems Engineers (IISE), 2008.
[74] Johnathan Mun. Real options analysis: Tools and techniques for valuing strategic
investments and decisions, volume 137. John Wiley & Sons, 2002.
[75] Davide Antonioli, Grazia Cecere, and Massimiliano Mazzanti. Information com-
munication technologies and environmental innovations in firms: joint adoptions
188
REFERENCES
and productivity effects. Journal of Environmental Planning and Management,

61(11):1905–1933, 2018.
[76] Grazia Cecere, Nicoletta Corrocher, Cédric Gossart, and Muge Ozman. Lock-in
and path dependence: an evolutionary approach to eco-innovations. Journal of
Evolutionary Economics, 24(5):1037–1065, 2014.
[77] W Brian Arthur. Foundations of complexity economics. Nature Reviews Physics,

3(2):136–145, 2021.
[78] W Brian Arthur. Complexity and the economy. science, 284(5411):107–109, 1999.
[79] Jianxi Gao, Baruch Barzel, and Albert-László Barabási. Universal resilience
patterns in complex networks. Nature, 530(7590):307–312, 2016.
[80] Jean-François Arnoldi, Bart Haegeman, Tomas Revilla, and Michel Loreau. Partic-
ularity of âĂĲuniversal resilience patterns in complex networksâĂİ. bioRxiv, page
056218, 2016.
[81] Jacobo Aguirre, David Papo, and Javier M Buldú. Successful strategies for compet-
ing networks. Nature Physics, 9(4):230–234, 2013.
[82] Cbod’s website. http://www.cbod.u-psud.fr/about/. Accessed 6/3/2019.
[83] Xiaolin Cheng. Cloud computing and decision-making: determinants, modelling

and impacts. PhD thesis, 2017.
[84] Gideon Greenspan. Avoiding the pointless blockchain project, 2015. URL
https://www. multichain. com/blog/2015/11/avoiding-pointless-blockchain-
project.
[85] Bart Suichies. Why blockchain must die in 2016. Retrieved from, 2015.
[86] Kunal Nandwani. Do you really need to use blockchain for your application, 2016.
[87] Price Water House Coopers. Blockchain: The $5 billion opportunity for rein-
surers. https://www.pwc.com/gx/en/financial-services/publications/assets/
blockchain-for-reinsurers.pdf, 2016.
189
REFERENCES
[88] David Birch, Richard G Brown, and Salome Parulava. Towards ambient account-
ability in financial services: Shared ledgers, translucent transactions and the
technological legacy of the great financial crisis. Journal of Payments Strategy &
Systems, 10(2):118–131, 2016.
[89] D. Saiko. Blockchain technology. http://www.cambridgefx.com/blog/

blockchain-technology/, 2016.
[90] IBM. How to decide when to use blockchain. https://www.ibm.com/

developerworks/community/blogs/gcuomo/resource/BLOGS_UPLOADED_IMAGES/
HowToDecideWhenToUseBlockchain.jpg, 2016.
[91] A. Lewis. Blockchain cheat sheet v0.1. https://bitsonblocks.files.wordpress.

com/2016/01/2016-01-26-fintech-finals-hk.pdf?lipi=urn%3Ali%3Apage%3Ad_
flagship3_pulse_read%3BQH2LGbmzRKy9rZY9N4nfsA%3D%3D., 2016.
[92] S. Meunier. When do you need blockchain? decision models. https://medium.

com/@sbmeunier/when-do-you-need-blockchain-decision-models-a5c40e7c9ba1,
2016.
[93] M. Quindazzi. Do you really need a blockchain? https://twitter.com/

mikequindazzi/status/787760892783894528, 2016.
[94] Deloitte. I. Blockchain a new model for health informa-

tion exchanges. https://www.semanticscholar.org/paper/I.
-Blockchain%E2%80%94A-New-Model-for-Health-Information-I../
b99277c3eecfe6d3dd784fe572a45780ffd040e2, 2016.
[95] Morgen E Peck. Blockchain world-do you need a blockchain? this chart will tell you
if the technology can solve your problem. IEEE Spectrum, 54(10):38–60, 2017.
[96] Lixar. Blockchain part 2. https://lixar.com/lixar-blog/tech/

blockchain-part-2/, 2017.
[97] K. Verslype. Beslissingsmodel: Wanneer blockchain gebruiken? https://www.

smalsresearch.be/beslissingsmodel-wanneer-blockchain-gebruiken/, 2017.
190
REFERENCES
[98] Sin Kuang Lo, Xiwei Xu, Yin Kia Chiam, and Qinghua Lu. Evaluating suitability
of applying blockchain. In 2017 22nd International Conference on Engineering of
Complex Computer Systems (ICECCS), pages 158–161. IEEE, 2017.
[99] Verified ICOs. Is a blockchain really required? https://medium.com/

@VerifiedICOs/is-a-blockchain-really-required-1a68c7791fa1., 2017.
[100] Yu-Pin Lin, Joy R Petway, Johnathen Anthony, Hussnain Mukhtar, Shih-Wei Liao,
Cheng-Fu Chou, and Yi-Fong Ho. Blockchain: The evolutionary next step for ict
e-agriculture. Environments, 4(3):50, 2017.
[101] T. Mueller. Will blockchain solve my business problem? https://medium.com/

contractus/do-i-need-a-blockchain-for-my-business-project-8f8cada7f3ac.,
2017.
[102] B. Henkel. Beginning blockchain: Key questions to get-

ting started. https://www.ca.com/en/blog-mainframeai/
beginning-blockchain-key-questions-to-getting-started.html., 2017.
[103] Roger Maull, Phil Godsiff, Catherine Mulligan, Alan Brown, and Beth Kewell.
Distributed ledger technology: Applications and implications. Strategic Change,
26(5):481–489, 2017.
[104] Xiwei Xu, Ingo Weber, Mark Staples, Liming Zhu, Jan Bosch, Len Bass, Cesare
Pautasso, and Paul Rimba. A taxonomy of blockchain-based systems for architecture
design. In 2017 IEEE International Conference on Software Architecture (ICSA),
pages 243–252. IEEE, 2017.
[105] Cap gemini - sai trends. https://sai.be/UserContent/FPAX7KUQ45F7B9C4KPZQ_SAI%

20trends%20-%20December%202017%20-%20handout.pdf, 2017.
[106] Sandra Klein and Wolfgang Prinz. A use case identification framework and use
case canvas for identifying and exploring relevant blockchain opportunities. In
Proceedings of 1st ERCIM Blockchain Workshop 2018. European Society for Socially
Embedded Technologies (EUSSET), 2018.
[107] N El Ioini, C Pahl, and Sven Helmer. A decision framework for blockchain platforms
for iot and edge computing. SCITEPRESS, 2018.
191
REFERENCES
[108] Florian Wessling, Christopher Ehmke, Marc Hesenius, and Volker Gruhn. How
much blockchain do you need? towards a concept for building hybrid dapp archi-
tectures. In 2018 IEEE/ACM 1st International Workshop on Emerging Trends in
Software Engineering for Blockchain (WETSEB), pages 44–47. IEEE, 2018.
[109] World Economic Forum. Blockchain beyond the hype. http://www3.weforum.org/

docs/48423_Whether_Blockchain_WP.pdf., 2018.
[110] Karl Wüst and Arthur Gervais. Do you need a blockchain? In 2018 Crypto Valley
Conference on Blockchain Technology (CVCBT), pages 45–54. IEEE, 2018.
[111] J. Gardner. Do you need a blockchain? https://twitter.com/Disruptepreneur/

status/755857596423077888/photo/1?ref_src=twsrc%5Etfw, 2018.
[112] Valentina Gatteschi, Fabrizio Lamberti, Claudio Demartini, Chiara Pranteda, and
Victor Santamaria. To blockchain or not to blockchain: That is the question. IT
Professional, 20(2):62–74, 2018.
[113] Tommy Koens and Erik Poll. What blockchain alternative do you need? In Data
Privacy Management, Cryptocurrencies and Blockchain Technology, pages 113–129.
Springer, 2018.
[114] Brian A Scriber. A framework for determining blockchain applicability. IEEE

Software, 35(4):70–77, 2018.
[115] M. Chand. Do you need a blockchain. https://www.c-sharpcorner.com/article/

do-you-need-a-blockchain2/., 2020.
[116] Mohammed YA Rawwas, Surendra Arjoon, and Yusuf Sidani. An introduction of

epistemology to business ethics: A study of marketing middle-managers. Journal
of Business Ethics, 117(3):525–539, 2013.
[117] Łukasz Sułkowski. Two paradigms in management epistemology. Journal of

intercultural management, 2(1):109–119, 2010.
[118] Andrzej K Koźmiński and Dominika Latusek-Jurczak. Rozwój teorii organizacji.

Wolters Kluwer, 2011.
192
REFERENCES
[119] Yves Levant and Henri Zimnovitch. Epistemology and management science: Is
accounting history still a legitimate subject of study? Accounting History, 22(4):450–
471, 2017.
[120] Hilary Putnam and Abel Gerschenfeld. Raison, vérité et histoire. 1986.
[121] Philippe DEVAUX, Nicol THYSSEN-RUTTEN, Karl R POPPER, and Jacques

Monod. La logique de la découverte scientifique. Editions Payot, 1973.
[122] Armand Hatchuel. Towards an epistemology of collective action: management

research as a responsive and actionable discipline. European Management Review,
2(1):36–47, 2005.
[123] Albert David et al. La recherche-intervention, cadre général pour la recherche en

management? Technical report, 2012.
[124] Sue Llewelyn. What counts as ’theory’ in qualitative management and accounting
research? introducing five levels of theorizing. Accounting, Auditing & Account-
ability Journal, 2003.
[125] Romain Laufer and Catherine Paradeise. Le prince bureaucrate: Machiavel au pays
du marketing. Flammarion, 1982.
[126] Claudine Tiercelin. Le doute en question: parades pragmatistes au défi sceptique.

Editions de l’éclat, 2005.
[127] Russell Keat and John Urry. Social science as theory, 1982.
[128] Christian De Cock. " it seems to fill my head with ideas" a few thoughts on post-
modernism, tqm, and bpr. Journal of Management Inquiry, 7(2):144–153, 1998.
[129] Paul Dicken. Constructive empiricism: epistemology and the philosophy of science.
Springer, 2010.
[130] Bas C Van Fraassen et al. The scientific image. Oxford University Press, 1980.
[131] Xiaolin Cheng, Alessandro Solimando, Ahmed Bounfour, and Emmanuel Waller.
Cloud computing adoption: A rule-based modeling. Cloud Computing and Decision-
Making Determinants, Modelling and Impacts, page 64, 2017.
193
REFERENCES
[132] Euripidis Loukis, Marijn Janssen, and Ianislav Mintchev. Determinants of software-
as-a-service benefits and impact on firm performance. Decision Support Systems,
117:38–47, 2019.
[133] Maxime Guériau, Frédéric Armetta, Salima Hassas, Romain Billot, and Nour-Eddin
El Faouzi. A constructivist approach for a self-adaptive decision-making system:
application to road traffic control. In 2016 IEEE 28th International Conference on
Tools with Artificial Intelligence (ICTAI), pages 670–677. IEEE, 2016.
[134] Mariana MX Lima and Regina C Ruschel. Proposition of an architectural design pro-
cess model based on a constructivist decision support approach. In ANNUAL CON-
FERENCE OF THE INTERNATIONAL GROUP FOR LEAN CONSTRUCTION-
IGLC, volume 21, pages 399–407, 2013.
[135] Merve Bayramusta and V Aslihan Nasir. A fad or future of it?: A comprehen-
sive literature review on the cloud computing research. International Journal of
Information Management, 36(4):635–644, 2016.
[136] Marian Carcary, Eileen Doherty, Gerard Conway, and Stephen McLaughlin. Cloud
computing adoption readiness and benefit realization in irish smesâĂŤan ex-
ploratory study. Information Systems Management, 31(4):313–327, 2014.
[137] Haibo Yang and Mary Tate. A descriptive literature review and classification of
cloud computing research. Communications of the Association for Information
Systems, 31(1):2, 2012.
[138] Chinyao Low, Yahsueh Chen, and Mingchang Wu. Understanding the determinants
of cloud computing adoption. Industrial management & data systems, 2011.
[139] Hasimi Sallehudin, Razli Che Razak, and Mohammad Ismail. Factors influencing
cloud computing adoption in the public sector: an empirical analysis. Journal of
Entrepreneurship and Business, 3(1):30–45, 2015.
[140] Maricela-Georgiana Avram. Advantages and challenges of adopting cloud comput-

ing from an enterprise perspective. Procedia Technology, 12(0):529–534, 2014.
194
REFERENCES
[141] Mark Stieninger, Dietmar Nedbal, Werner Wetzlinger, Gerold Wagner, and
Michael A Erskine. Impacts on the organizational adoption of cloud computing: A
reconceptualization of influencing factors. Procedia Technology, 16:85–93, 2014.
[142] Subhas Chandra Misra and Arka Mondal. Identification of a company’s suitability
for the adoption of cloud computing and modelling its corresponding return on
investment. Mathematical and Computer Modelling, 53(3-4):504–521, 2011.
[143] Íñigo Goiri, Jordi Guitart, and Jordi Torres. Economic model of a cloud provider
operating in a federated cloud. Information Systems Frontiers, 14(4):827–843, 2012.
[144] Andreas Felsberger, Bernhard Oberegger, and Gerald Reiner. A review of decision
support systems for manufacturing systems. In SAMI@ iKNOW, 2016.
[145] Robert H Bonczek, Clyde W Holsapple, and Andrew B Whinston. Foundations of

decision support systems. Academic Press, 2014.
[146] Herbert A Simon. The new science of management decision. 1960.
[147] Per Hilletofth, Olli-Pekka Hilmola, and Yacan Wang. Simulation based decision
support systems in the supply chain context. Industrial Management & Data
Systems, 2016.
[148] Aziza Chakir, Meriem Chergui, S Elhasnaou, Hicham Medromi, and Adil Sayouti.
A decision approach to select the best framework to treat an it problem by using
multi-agent system and expert systems. In International Symposium on Ubiquitous
Networking, pages 499–511. Springer, 2015.
[149] Arpan Kumar Kar. A hybrid group decision support system for supplier selection
using analytic hierarchy process, fuzzy set theory and neural network. Journal of
Computational Science, 6:23–33, 2015.
[150] Fred D Davis. Perceived usefulness, perceived ease of use, and user acceptance of
information technology. MIS quarterly, pages 319–340, 1989.
[151] Louis G Tornatzky and Katherine J Klein. Innovation characteristics and innovation
adoption-implementation: A meta-analysis of findings. IEEE Transactions on
engineering management, (1):28–45, 1982.
195
REFERENCES
[152] Pamila Dembla, Charles Flack, and Stacie Petter. Extending the delone and mclean
is success model to cloud computing. 2015.
[153] Haibo Yang and Mary Tate. Where are we at with cloud computing?: a descriptive
literature review. 2009.
[154] George Lawton. Moving the os to the web. Computer, 41(3):16–19, 2008.
[155] Panos Louridas. Up in the air: Moving your applications to the cloud. IEEE
software, 27(4):6–11, 2010.
[156] Peng Wang, Dan Meng, Jizhong Han, Jianfeng Zhan, Bibo Tu, Xiaofeng Shi, and
Le Wan. Transformer: A new paradigm for building data-parallel programming
models. IEEE micro, 30(4):55–64, 2010.
[157] Yang Liu, Maozhen Li, Nasullah Khalid Alham, and Suhel Hammoud. Hsim: a
mapreduce simulator in enabling cloud computing. Future Generation Computer
Systems, 29(1):300–308, 2013.
[158] Maciej Malawski, Jan Meizner, Marian Bubak, and Paweł Gepner. Component
approach to computational applications on clouds. Procedia Computer Science,
4:432–441, 2011.
[159] Alexander Böhm and Carl-Christian Kanne. Demaq/transscale: automated distri-

bution and scalability for declarative applications. Information Systems, 36(3):565–
578, 2011.
[160] Anol Bhattacherjee and Sang Cheol Park. Why end-users move to the cloud: a
migration-theoretic analysis. European Journal of Information Systems, 23(3):357–
372, 2014.
[161] Imran Khan Azeemi, Mike Lewis, and Theo Tryfonas. Migrating to the cloud:
lessons and limitations of âĂŸtraditional’is success models. Procedia Computer
Science, 16:737–746, 2013.
[162] Patrick Hoberg, Jan Wollersheim, and Helmut Krcmar. The business perspective
on cloud computing-a literature review of research on cloud computing. 2012.
196
REFERENCES
[163] Ali Khajeh-Hosseini, Ian Sommerville, Jurgen Bogaerts, and Pradeep Teregowda.
Decision support tools for cloud migration in the enterprise. In 2011 IEEE 4th
International Conference on Cloud Computing, pages 541–548. IEEE, 2011.
[164] Miranda Zhang, Rajiv Ranjan, Armin Haller, Dimitrios Georgakopoulos, and Peter
Strazdins. Investigating decision support techniques for automating cloud service
selection. In 4th IEEE International Conference on Cloud Computing Technology
and Science Proceedings, pages 759–764. IEEE, 2012.
[165] Adrian Juan-Verdejo, Steffen Zschaler, Bholanathsingh Surajbali, Henning Baars,

and Hans-Georg Kemper. Inclouder: A formalised decision support modelling
approach to migrate applications to cloud environments. In 2014 40th EUROMICRO
Conference on Software Engineering and Advanced Applications, pages 467–474.
IEEE, 2014.
[166] Alan Hevner and Samir Chatterjee. Design science research in information systems.
In Design research in information systems, pages 9–22. Springer, 2010.
[167] Marek J Druzdzel and Roger R Flynn. Decision support systems. encyclopedia of
library and information science. a. kent. Marcel Dekker, Inc. Last Login, 10(03):2010,
1999.
[168] Wissam Al Jurdi, Jacques Bou Abdo, Jacques Demerjian, and Abdallah Makhoul.
Critique on natural noise in recommender system. ACM Transactions on Knowledge
Discovery (Under Review), 2020.
[169] Helen Nissenbaum and Finn Brunton. Obfuscation: A user’s guide for privacy and
protest. Hachette Audio, 2015.
[170] Wissam Al Jurdi, Jacques Bou Abdo, Jacques Demerjian, and Abdallah Makhoul.
Obfuscation in recommender systems: Desperate and strategic attack. ACM Trans-
actions on Knowledge Discovery (Under Review), 2020.
[171] Xavier Amatriain, Josep M Pujol, and Nuria Oliver. I like it... i like it not: Evaluating
user ratings noise in recommender systems. In International Conference on User
Modeling, Adaptation, and Personalization, pages 247–258. Springer, 2009.
197
REFERENCES
[172] Xavier Amatriain, Josep M Pujol, Nava Tintarev, and Nuria Oliver. Rate it again:
increasing recommendation accuracy by user re-rating. In Proceedings of the third
ACM conference on Recommender systems, pages 173–180, 2009.
[173] Charu C Aggarwal. An introduction to recommender systems. In Recommender

systems, pages 1–28. Springer, 2016.
[174] Neil Rubens, Mehdi Elahi, Masashi Sugiyama, and Dain Kaplan. Active learning
in recommender systems. In Recommender systems handbook, pages 809–846.
Springer, 2015.
[175] Miriam El Khoury Badran, Wissam Jurdi, and Jacques Bou Abdo. Survey on shilling
attacks and their detection algorithms in recommender systems. In Proceedings of
the International Conference on Security and Management (SAM), pages 141–146.
The Steering Committee of The World Congress in Computer Science, Computer âĂę,
2019.
[176] Raciel Yera Toledo, Yailé Caballero Mota, and Luis Martínez. Correcting noisy
ratings in collaborative recommender systems. Knowledge-Based Systems, 76:96–
108, 2015.
[177] Félix Hernández Del Olmo and Elena Gaudioso. Evaluation of recommender
systems: A new approach. Expert Systems with Applications, 35(3):790–804, 2008.
[178] Jonathan L Herlocker, Joseph A Konstan, Loren G Terveen, and John T Riedl.
Evaluating collaborative filtering recommender systems. ACM Transactions on
Information Systems (TOIS), 22(1):5–53, 2004.
[179] Paolo Massa and Paolo Avesani. Trust metrics in recommender systems. In
Computing with social trust, pages 259–285. Springer, 2009.
[180] Guy Shani and Asela Gunawardana. Evaluating recommendation systems. In

Recommender systems handbook, pages 257–297. Springer, 2011.
[181] Georges Chaaya, Elisabeth Métais, Jacques Bou Abdo, Raja Chiky, Jacques De-
merjian, and Kablan Barbar. Evaluating non-personalized single-heuristic active
learning strategies for collaborative filtering recommender systems. In 2017 16th
198
REFERENCES
IEEE International Conference on Machine Learning and Applications (ICMLA),

pages 593–600. IEEE, 2017.
[182] Gunnar Schröder, Maik Thiele, and Wolfgang Lehner. Setting goals and choosing
metrics for recommender system evaluations. In UCERSTI2 workshop at the 5th
ACM conference on recommender systems, Chicago, USA, volume 23, page 53, 2011.
[183] David MW Powers. Evaluation: from precision, recall and f-measure to roc, in-
formedness, markedness and correlation. arXiv preprint arXiv:2010.16061, 2020.
[184] NGMN Alliance. 5g white paper. Next generation mobile networks, white paper, 1,
2015.
[185] Faqir Zarrar Yousaf, Paulo Loureiro, Frank Zdarsky, Tarik Taleb, and Marco
Liebsch. Cost analysis of initial deployment strategies for virtualized mobile core
network functions. IEEE Communications Magazine, 53(12):60–66, 2015.
[186] Costas Courcoubetis and Richard Weber. Pricing and communications networks.
Wiley-Interscience series in systems and optimization, page 3, 2003.
[187] U. Wilensky. Netlogo - center for connected learning and computer-based modeling,
northwestern university, evanston, il. http://ccl.northwestern.edu/netlogo/.,
1999.
[188] Timothy K Forde, Irene Macaluso, and Linda E Doyle. Exclusive sharing & vir-
tualization of the cellular network. In 2011 IEEE International Symposium on
Dynamic Spectrum Access Networks (DySPAN), pages 337–348. IEEE, 2011.
[189] Malla Reddy Sama, Yvon Gourhant, and Lucian Suciu. Cloud based mobile network
sharing: A new model. EAI Endorsed Trans. Indust. Netw. & Intellig. Syst., 2(3):e2,
2015.
[190] Paul Resnick and Hal R Varian. Recommender systems. Communications of the
ACM, 40(3):56–58, 1997.
[191] Dietmar Jannach. Finding preferred query relaxations in content-based recom-

menders. In Intelligent techniques and tools for novel system architectures, pages
81–97. Springer, 2008.
199
REFERENCES
[192] 3leafnodes. Apache spark with a recommender system, 2018.
[193] Francesco Ricci, Lior Rokach, and Bracha Shapira. Introduction to recommender
systems handbook. In Recommender systems handbook, pages 1–35. Springer, 2011.
[194] Bo Han, Vijay Gopalakrishnan, Lusheng Ji, and Seungjoon Lee. Network function
virtualization: Challenges and opportunities for innovations. IEEE Communica-
tions Magazine, 53(2):90–97, 2015.
[195] Hassan Hawilo, Abdallah Shami, Maysam Mirahmadi, and Rasool Asal. Nfv: state
of the art, challenges, and implementation in next generation mobile networks
(vepc). IEEE Network, 28(6):18–26, 2014.
[196] Miloud Bagaa, Tarik Taleb, Abdelquoddouss Laghrissi, and Adlen Ksentini. Ef-
ficient virtual evolved packet core deployment across multiple cloud domains. In
2018 IEEE Wireless Communications and Networking Conference (WCNC), pages
1–6. IEEE, 2018.
[197] Farhan Ahmad Khan. Virtualized epc: Unleashing the potential of nfv and sdn.
2014.
[198] Malla Reddy Sama, Luis M Contreras, John Kaippallimalil, Ippei Akiyoshi, Haiyang
Qian, and Hui Ni. Software-defined control of the virtualized mobile packet core.
IEEE Communications Magazine, 53(2):107–115, 2015.
[199] Luca Valtulina, Morteza Karimzadeh, Georgios Karagiannis, Geert Heijenk, and
Aiko Pras. Performance evaluation of a sdn/openflow-based distributed mobility
management (dmm) approach in virtualized lte systems. In 2014 IEEE Globecom
Workshops (GC Wkshps), pages 18–23. IEEE, 2014.
[200] Bruno Sousa, Luis Cordeiro, Paulo Simoes, Andy Edmonds, Santiago Ruiz,
Giuseppe A Carella, Marius Corici, Navid Nikaein, Andre S Gomes, Eryk Schiller,
et al. Toward a fully cloudified mobile network infrastructure. IEEE Transactions
on Network and Service Management, 13(3):547–563, 2016.
[201] vepc for communications service providers. intel white-paper., 2016.
200
REFERENCES
[202] Zhongliang Zhao, Morteza Karimzadeh, Torsten Braun, Aiko Pras, and Hans
van den Berg. A demonstration of mobility prediction as a service in cloudified
lte networks. In 2015 IEEE 4th International Conference on Cloud Networking
(CloudNet), pages 78–80. IEEE, 2015.
[203] Konstantinos Samdanis, Xavier Costa-Perez, and Vincenzo Sciancalepore. From

network sharing to multi-tenancy: The 5g network slice broker. IEEE Communica-
tions Magazine, 54(7):32–39, 2016.
[204] Tarik Taleb, Marius Corici, Carlos Parada, Almerima Jamakovic, Simone Ruffino,
Georgios Karagiannis, and Thomas Magedanz. Ease: Epc as a service to ease
mobile core network deployment over cloud. IEEE Network, 29(2):78–88, 2015.
[205] Xiuhua Li, Xiaofei Wang, Chunsheng Zhu, Wei Cai, and Victor CM Leung. Caching-
as-a-service: Virtual caching framework in the cloud-based mobile networks. In
2015 IEEE Conference on Computer Communications Workshops (INFOCOM WK-
SHPS), pages 372–377. IEEE, 2015.
[206] Xiuhua Li, Xiaofei Wang, Keqiu Li, and Victor CM Leung. Caas: Caching as a
service for 5g networks. IEEE Access, 5:5982–5993, 2017.
[207] Tarik Taleb, Adlen Ksentini, and Bruno Sericola. On service resilience in cloud-
native 5g mobile systems. IEEE Journal on Selected Areas in Communications,
34(3):483–496, 2016.
[208] Seil Jeon, Daniel Corujo, and Rui L Aguiar. Virtualised epc for on-demand mobile
traffic offloading in 5g environments. In 2015 IEEE Conference on Standards for
Communications and Networking (CSCN), pages 275–281. IEEE, 2015.
[209] Eleonora Cau, Marius Corici, Paolo Bellavista, Luca Foschini, Giuseppe Carella,
Andy Edmonds, and Thomas Michael Bohnert. Efficient exploitation of mobile edge
computing for virtualized 5g in epc architectures. In 2016 4th IEEE international
conference on mobile cloud computing, services, and engineering (MobileCloud),
pages 100–109. IEEE, 2016.
[210] Ruben Solozabal, Aitor Sanchoyerto, Eneko Atxutegi, Bego Blanco, Jose Oscar
Fajardo, and Fidel Liberal. Exploitation of mobile edge computing in 5g distributed
201
REFERENCES
mission-critical push-to-talk service deployment. IEEE Access, 6:37665–37675,

2018.
[211] Bruno Sousa, Zhongliang Zhao, Morteza Karimzadeh, David Palma, Vitor Fonseca,
Paulo Simoes, Torsten Braun, Hans Van Den Berg, Aiko Pras, and Luis Cordeiro.
Enabling a mobility prediction-aware follow-me cloud model. In 2016 IEEE 41st
Conference on Local Computer Networks (LCN), pages 486–494. IEEE, 2016.
[212] Ibrahim Afolabi, Miloud Bagaa, Tarik Taleb, and Hannu Flinck. End-to-end network
slicing enabled through network function virtualization. In 2017 IEEE Conference
on Standards for Communications and Networking (CSCN), pages 30–35. IEEE,
2017.
[213] Morteza Karimzadeh, Zhongliang Zhao, Luuk Hendriks, Ricardo de O Schmidt, Se-
bastiaan la Fleur, Hans van den Berg, Aiko Pras, Torsten Braun, and Marius Julian
Corici. Mobility and bandwidth prediction as a service in virtualized lte systems.
In 2015 IEEE 4th International Conference on Cloud Networking (CloudNet), pages
132–138. IEEE, 2015.
[214] Victor Chang, David Bacigalupo, Gary Wills, and David De Roure. A categorisa-
tion of cloud computing business models. In 2010 10th IEEE/ACM International
Conference on Cluster, Cloud and Grid Computing, pages 509–512. IEEE, 2010.
[215] Sean Marston, Zhi Li, Subhajyoti Bandyopadhyay, Juheng Zhang, and Anand
Ghalsasi. Cloud computingâĂŤthe business perspective. Decision support systems,
51(1):176–189, 2011.
[216] Abbas Strømmen-Bakhtiar and Amir R Razavi. Cloud computing business models.
In Cloud Computing for Enterprise Architectures, pages 43–60. Springer, 2011.
[217] Michael Jaekel and Achim Luhn. Cloud computing–business models, value creation
dynamics and advantages for customers. Siemens IT Solutions and Services, 2009.
[218] Geoffrey B West. Scale: the universal laws of growth, innovation, sustainability,
and the pace of life in organisms, cities, economies, and companies. Penguin, 2017.
202
REFERENCES
[219] Artan Mazrekaj, Isak Shabani, and Besmir Sejdiu. Pricing schemes in cloud
computing: an overview. International Journal of Advanced Computer Science and
Applications, 7(2):80–86, 2016.
[220] Peter C Fishburn and Andrew M Odlyzko. Competitive pricing of information goods:
Subscription pricing versus pay-per-use. Economic Theory, 13(2):447–470, 1999.
[221] Gabriel Bitran and René Caldentey. An overview of pricing models for revenue
management. Manufacturing & Service Operations Management, 5(3):203–229,
2003.
[222] Kevin Lai. Markets are dead, long live markets. ACM SIGecom Exchanges, 5(4):1–
10, 2005.
[223] George Darzanos, Iordanis Koutsopoulos, and George D Stamoulis. Economics

models and policies for cloud federations. In 2016 IFIP Networking Conference
(IFIP Networking) and Workshops, pages 485–493. IEEE, 2016.
[224] Bhanu Sharma, Ruppa K Thulasiram, Parimala Thulasiraman, Saurabh K Garg,

and Rajkumar Buyya. Pricing cloud compute commodities: A novel financial
economic model. In 2012 12th IEEE/ACM International Symposium on Cluster,
Cloud and Grid Computing (ccgrid 2012), pages 451–457. IEEE, 2012.
[225] Charalampos Meidanis, Ioannis Stiakogiannakis, and Maria Papadopouli. Pricing

for mobile virtual network operators: The contribution of u-map. In 2014 IEEE
International Symposium on Dynamic Spectrum Access Networks (DYSPAN), pages
133–136. IEEE, 2014.
[226] Qian Ma, Ya-Feng Liu, and Jianwei Huang. Time and location aware mobile data
pricing. IEEE Transactions on Mobile Computing, 15(10):2599–2613, 2015.
[227] Xun Xiao, Rui Zhang, Jianping Wang, Chunming Qiao, and Kejie Lu. An optimal
pricing scheme to improve transmission opportunities for a mobile virtual network
operator. Computer Networks, 99:51–65, 2016.
[228] Meng Zhang, Lin Gao, Jianwei Huang, and Michael Honig. Cooperative and
competitive operator pricing for mobile crowdsourced internet access. In IEEE
203
REFERENCES
INFOCOM 2017-IEEE Conference on Computer Communications, pages 1–9. IEEE,

2017.
[229] Xiaofei Wang, Athanasios V Vasilakos, Min Chen, Yunhao Liu, and Ted Taekyoung
Kwon. A survey of green mobile networks: Opportunities and challenges. Mobile
Networks and Applications, 17(1):4–20, 2012.
[230] Hélène Le Cadre, Mustapha Bouhtou, and Bruno Tuffin. A pricing model for a
mobile network operator sharing limited resource with a mobile virtual network
operator. In International Workshop on Internet Charging and QoS Technologies,
pages 24–35. Springer, 2009.
[231] Nan Feng, Siun-Chuon Mau, and Narayan B Mandayam. Pricing and power con-
trol for joint network-centric and user-centric radio resource management. IEEE
Transactions on Communications, 52(9):1547–1557, 2004.
[232] Soumya Sen, Carlee Joe-Wong, Sangtae Ha, and Mung Chiang. A survey of smart
data pricing: Past proposals, current plans, and future trends. Acm computing
surveys (csur), 46(2):1–37, 2013.
[233] Youngsoo Kim, Rahul Telang, William B Vogt, and Ramayya Krishnan. An empirical
analysis of mobile voice service and sms: a structural model. Management Science,
56(2):234–252, 2010.
[234] Matteo Vincenzi, Angelos Antonopoulos, Elli Kartsakli, John Vardakas, Luis Alonso,
and Christos Verikoukis. Cooperation incentives for multi-operator c-ran energy
efficient sharing. In 2017 IEEE International Conference on Communications (ICC),
pages 1–6. IEEE, 2017.
[235] Nicoletta Corrocher and Lorenzo Zirulia. Demand and innovation in services: The
case of mobile communications. Research Policy, 39(7):945–955, 2010.
[236] Jingtao Ding, Yong Li, Pengyu Zhang, and Depeng Jin. Time dependent pricing
for large-scale mobile networks of urban environment: Feasibility and adaptability.
IEEE Transactions on Services Computing, 2017.
204
REFERENCES
[237] Masoud Asghari and Saleh Yousefi. Pricing mechanism for interconnection between
phone operators and virtual mobile voip operators. Telecommunication Systems,
67(1):133–147, 2018.
[238] Heli Koski and Tobias Kretschmer. Entry, standards and competition: Firm strate-
gies and the diffusion of mobile telephony. Review of Industrial Organization,
26(1):89–113, 2005.
[239] Zoltan Papai, Peter Nagy, and Bertalan Papp. Does the number or the composition
of players matter on the mobile broadband markets?-lessons from a benchmarking
study of the largescreen mobile broadband prices in the european union. 2017.
[240] Jone Consul, Cristina Perfecto, Miren Nekane Bilbao, and Javier Del Ser. An
analysis of coalition-competition pricing strategies for multi-operator mobile traffic
offloading using bi-objective heuristics. In International Conference on Harmony
Search Algorithm, pages 157–167. Springer, 2017.
[241] Uri Wilensky and William Rand. An introduction to agent-based modeling: modeling
natural, social, and engineered complex systems with NetLogo. Mit Press, 2015.
[242] Gary Madden and Grant Coble-Neal. Economic determinants of global mobile
telephony growth. Information Economics and Policy, 16(4):519–534, 2004.
[243] John Arnold and Tony Hope. Accounting for management decisions. Prentice Hall,
1990.
[244] Tönu Puu. Oligopoly: old ends-new means. Springer Science & Business Media,
2010.
[245] Antoine Augustin Cournot. Researches into the Mathematical Principles of the
Theory of Wealth. Macmillan, 1897.
[246] J Bertrand. ibook review of theorie mathematique de la richesse sociale and of

recherches sur les principles mathematiques de la theorie des richessesj. Journal
de Savants, 67:1067–1082, 1883.
[247] Tord Palander. Konkurrens och marknadsjämvikt vid duopol och oligopol. i. ful-
lkomlig marknad och" autonomt" handlande. Ekonomisk Tidskrift, (r 3):222–250,
1939.
205
REFERENCES
[248] Joan Robinson. The theory of imperfect competition. Quarterly Journal of Eco-
nomics, 1933.
[249] Martin J Osborne et al. An introduction to game theory, volume 3. Oxford university
press New York, 2004.
[250] Tamer Başar and Geert Jan Olsder. Dynamic noncooperative game theory. SIAM,
1998.
[251] John F Nash Jr. The bargaining problem. Econometrica: Journal of the econometric
society, pages 155–162, 1950.
[252] Jean-Jacques Laffont and Jean Tirole. Competition in telecommunications. MIT

press, 2001.
[253] Xiaolong Zhu, Weidong Zhu, and Lei Yu. Analysis of a nonlinear mixed cournot
game with boundedly rational players. Chaos, Solitons & Fractals, 59:82–88, 2014.
[254] Janice Hauge and Mark Jamison. Analyzing telecommunications market competi-
tion: foundations for best practices. Public Utility Research Center, University of
Florida, 29, 2009.
[255] Mark Jamison. Methods for increasing competition in telecommunications markets.

Gainesville, Florida: Public Utility Research Center, 2012.
[256] Ali Vafamehr, Mohammad E Khodayar, and Khaled Abdelghany. Oligopolistic com-
petition among cloud providers in electricity and data networks. IEEE Transactions
on Smart Grid, 10(2):1801–1812, 2017.
[257] Kurt Helmes and Rainer Schlosser. Oligopoly pricing and advertising in isoelastic
adoption models. Dynamic Games and Applications, 5(3):334–360, 2015.
[258] Vivek Farias, Denis Saure, and Gabriel Y Weintraub. An approximate dynamic
programming approach to solving dynamic oligopoly models. The RAND Journal
of Economics, 43(2):253–282, 2012.
[259] Vijay V Desai, Vivek F Farias, and Ciamac C Moallemi. Approximate dynamic
programming via a smoothed linear program. Operations Research, 60(3):655–674,
2012.
206
REFERENCES
[260] Paul D Klemperer and Margaret A Meyer. Supply function equilibria in oligopoly
under uncertainty. Econometrica: Journal of the Econometric Society, pages 1243–
1277, 1989.
[261] Ngo Van Long and Antoine Soubeyran. Cost manipulation games in oligopoly, with
costs of manipulating. International Economic Review, 42(2):505–533, 2001.
[262] Andrew Ledvina and Ronnie Sircar. Oligopoly games under asymmetric costs
and an application to energy production. Mathematics and Financial Economics,
6(4):261–293, 2012.
[263] Benoıt Chevalier-Roignant, Christoph M Flath, and Lenos Trigeorgis. Dynamic

asymmetric cournot oligopoly under capacity constraints.
[264] Anita Van den Berg, Iwan Bos, P Jean-Jacques Herings, and Hans Peters. Dynamic
cournot duopoly with intertemporal capacity constraints. International Journal of
Industrial Organization, 30(2):174–192, 2012.
[265] Gian Italo Bischi and Fabio Lamantia. A dynamic model of oligopoly with r&d
externalities along networks. part i. Mathematics and Computers in Simulation,
84:51–65, 2012.
[266] Tomohiro Ara, Arghya Ghosh, and Hongyong Zhang. Tariffs, vertical oligopoly and
market structure. Technical report, Working Paper, 2017.
[267] Jumoke Lambo. Telecommunications-nigeria. 2013.
[268] Rosa-Branca Esteves. Behavior-based price discrimination with retention offers.

Information Economics and Policy, 27:39–51, 2014.
[269] Marco Bernardi and Jack Nuijten. Final report on number portability for mobile
networks, 2000.
[270] Jatinder Singh, Christopher Millard, Chris Reed, Jennifer Cobbe, and Jon
Crowcroft. Accountability in the iot: Systems, law, and ways forward. Computer,
51(7):54–65, 2018.
207

Corporate Decision-Making For The Adoption of Cloud Computing: Economic and Organizational Factors

Transféré par

Droits d'auteur :

Formats disponibles

Corporate Decision-Making For The Adoption of Cloud Computing: Economic and Organizational Factors

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Corporate Decision-Making For The Adoption of Cloud Computing: Economic and Organizational Factors

Transféré par

Droits d'auteur :

Formats disponibles

Corporate Decision-Making for the Adoption of Cloud

Computing : Economic and Organizational Factors

To cite this version:

HAL Id: tel-03631249

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

Thèse de doctorat de l’Université Paris-Saclay

École doctorale n◦ 630, Droit Economie Management (DEM)

Thèse présentée et soutenue à Sceaux, le 07/12/2021, par

Jacques BOU ABDO

Grazia Cecere Présidente

Professor, Institut Mines-Télécom Business School

I can’t but thank all my colleagues at “Université Paris-Saclay, Réseaux

Love of science and knowledge is not comparable to the love of my angel

Above all, I am thankful for God’s grace.

Ad maiorem Dei gloriam

2 Decision Support Solution for Cloud Computing Adoption 39

3 The Network Economics of Cloud Computing 77

4 The Oligopoly of Cloud Computing 131

5 General Discussion 167

0.1 Cloud stack (Well-known layers). . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.1 Cloud adoption literature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.1 Decision-Making Process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.1 Oligopolistic equilibrium with legacy operators out of business (cloud-based

0.1 Advantages and disadvantages of cloud computing. . . . . . . . . . . . . . . . 5

1.1 Previous contributions to CBOD objectives and tasks . . . . . . . . . . . . . . 25

2.1 Chapter 2 contributions to CBOD objectives and tasks . . . . . . . . . . . . . 69

3.1 Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

4.1 Chapter 4 contributions to CBOD objectives and tasks . . . . . . . . . . . . . 149

6.1 Chapter 5 contributions to CBOD objectives and tasks . . . . . . . . . . . . . 177

Selecting winner technologies is a tedious task, contains many ambiguities and

executives, in contrary to legislators, have to assess all emerging technologies, thus

Cloud computing is a typical disruptive technology that divides the market as

Blockchain is the technology supporting many applications in financial, healthcare,

Background on Cloud Computing

• "a parallel and distributed computing system consisting of a collection of inter-connected

development platforms and/or services). These resources can be dynamically recon-

Cloud computing can be better described by two complementary definitions, one

• “Pool of resource under the management of a centralized entity that is capable of

The last component to define is the "Cloud Manager". It provides higher-level

• Transferring the Virtual Machines and extending the hypervisor.

Table 0.1 – Advantages and disadvantages of cloud computing.

The advantages and disadvantages of CC as described in [11] are summarized in

cost-effective. "However, it is less secured: among all the Cloud deployment

• Hybrid: It is a mix of several distinct Cloud infrastructures. It is a single architecture

• Community: "A Cloud of type community is a collaborative and shared Cloud

"The Cloud architecture stack is composed of main layers such as SaaS

• Software as a Service (SaaS): is a web-based application interface where the end-

• Platform as a Service (PaaS): is a platform on which software and applications can

Figure 0.1 – Cloud stack (Well-known layers).

Cloud computing adopters experience the following issues in variable extents.

• Lock-in: It is one of the principal constraints of the Cloud environment. It concerns

to the cloud. Finally, it enables the delivery of e-payment services.

Figure 0.2 – Mobile Cloud computing (MCC).

• Chapter 2 - Decision Support Solution for Cloud Computing Adoption

lock-in. This chapter presents a generic and vendor-neutral decision sup-

• Chapter 3 - The Network Economics of Cloud Computing

• Chapter 4 - The Oligopoly of Cloud Computing

• Chapter 5 - General Discussion

1.2 Research Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23