JOURNAL OF LATEX CLASS FILES, VOL. 14, NO.

8, AUGUST 2015 1

Cloud computing as a platform for monetizing data

services: A two-sided game business model
Ahmed Saleh Bataineh, Jamal Bentahar, Rabeb Mizouni, Omar Abdel Wahab, Gaith Rjoub, May El Barachi

Abstract—With the unprecedented reliance on cloud comput- favoring the access to cloud-based infrastructure over pur-
ing as the backbone for storing today’s big data, we argue in chasing their own computing and storage platforms, find in
this paper that the role of the cloud should be reshaped from the enormous and varied number of data consumers that deal
being a passive virtual market to become an active platform
arXiv:2104.12762v1 [cs.GT] 26 Apr 2021

for monetizing the big data through Artificial Intelligence (AI) with the cloud an extra motivation to store their data on
services. The objective is to enable the cloud to be an active this platform to improve their exposure and increase their
platform that can help big data service providers reach a wider market shares. This indirectly makes, as shown in Figure
set of customers and cloud users (i.e., data consumers) to be 1, the cloud computing platform a mediator between data
exposed to a larger and richer variety of data to run their providers and data consumers and a principal player in the
data analytic tasks. To achieve this vision, we propose a novel
game theoretical model, which consists of a mix of cooperative whole big data analytics process. This opens the door for
and competitive strategies. The players of the game are the big new and innovative business models to take advantage of this
data service providers, cloud computing platform, and cloud scenario to increase the profits of all the involved parties, apart
users. The strategies of the players are modeled using the two- from the traditional business models which treat the cloud
sided market theory that takes into consideration the network as being a passive virtual market for offering services via
effects among involved parties, while integrating the externalities
between the cloud resources and consumer demands into the the Internet. Specifically, the literature on business-oriented
design of the game. Simulations conducted using Amazon and data trading can be classified into two main categories, i.e.,
google clustered data show that the proposed model improves pure merchant approaches and collaborative approaches. The
the total surplus of all the involved parties in terms of cloud proposals under the pure merchant approach such as [3, 4] and
resources provision and monetary profits compared to the current [2] adopt classic economic approaches, mainly the demand-
merchant model.
supply model and one-sided game theory/auction-based pric-
Index Terms—Cloud computing business model; game theory; ing to model the interactions among data providers, data
big data, two-sided-market. consumers and third-party platforms (i.e., information service
providers). In this approach, the third-party platform aims to
maximize its revenue through buying data from their owners,
I. I NTRODUCTION
reprocessing them, extracting useful information and selling
LOUD computing is witnessing a striking increase in the
C number of enterprises and manufacturers that are relying
on this paradigm to store and process their data. For example,
this information to consumers. This approach suffers from
several limitations when applied in cloud computing scenarios.
The first limitation is related to the diversity in the data
the study reported in [1] revealed that one million customers consumers’ interests, which entails higher processing costs
deploy their own enterprises on Amazon, spending 30 billion (in terms of information extraction for different customers’
USD on persistent storage on Amazon EC2 instances and interests) for the third-party platform. Moreover, under the
generating 600 ZB of data per year [2]. This explosive amount pure merchant model, data providers aim to maximize their
of data generated and stored on cloud resources forms the revenue of using their data commodities while the third-party
backbone for Artificial Intelligence (AI) services and opens platform aims to minimize the cost of raw data bought from
the door for a new cloud business paradigm, enabling the latter these providers. In parallel, from the data consumers’ side,
to be an active platform for monetizing data that benefit AI the third-party platform aims to maximize its revenue from
services. However, the cloud is not the actual owner for these selling the processed information while consumers aim to
big chunks of data, and has no right to trade and use these minimize the cost of information commodities, considering
data without considering its actual owners. the maximum available quality and quantity of information.
Motivated by the vision of the cloud as platform for The resulting equilibrium from such aggressive competitions
monetizing data services, we propose in this paper a novel among the different involved parties leads to less and coarse
cloud business model which allows data consumers (e.g., distribution of the total surplus. In addition, data differ from
market research enterprises) to run their data analytics on other economic goods for its potential of being (re)-sold to
the huge and diverse data that are stored on the cloud. This many consumers at the same time. In the pure merchant model,
not only gives data consumers the opportunity to extract since economic goods cannot be resold, the equilibrium of
valuable patterns from massive data coming from multiple the market lies at the intersection of the demand and supply
data providers, but also releases them from having to search curves. This means that the quantities of goods needed by
and discover appropriate providers for each particular type consumers is equal to the quantities of goods provided by
of data they need to analyze. Data providers, in addition to the sellers. This however does not hold in our case since
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 2

dŚĞŚƵŐĞĚĂƚĂ
ƌĞƐŝĚŝŶŐŝŶƚŚĞ
ĐůŽƵĚĞŶƚŝĐĞƐƚŚĞ
ĞƉůŽLJĞĚ
ĚĂƚĂĐŽŶƐƵŵĞƌƐƚŽ
ĞŶƚĞƌƉƌŝƐĞƐ dŚĞƌŽůĞŽĨƚŚĞĐůŽƵĚƉƌŽǀŝĚĞƌƚƵƌŶƐĂƌŽƵŶĚĨƌŽŵĂŶ/d ĐŽŶŶĞĐƚƚŚĞĐůŽƵĚ
ďĞĐŽŵĞĚĂƚĂ ƐƵƉƉůŝĞƌƚŚĂƚƐĞůůƐͬƌĞŶƚƐŽŶůLJĐůŽƵĚĐŽŵŵŽĚŝƚŝĞƐ;/d ƉƌŽǀŝĚĞƌƚŽďĞŶĞĨŝƚ
ƉƌŽǀŝĚĞƌƐ͘dŚĞLJ ŝŶĨƌĂƐƚƌƵĐƚƵƌĞͬƐŽĨƚǁĂƌĞͿƚŽĂƚŚŝƌĚͲƉĂƌƚLJͬďƌŽŬĞƌĂŐĞŶƚ ĨƌŽŵƚŚĞƐĞĚĂƚĂ͘
ĐŽŶŶĞĐƚƚŽƚŚĞ ƚŚĂƚĨĂĐŝůŝƚĂƚĞƐƚŚĞƚƌĂĚĞŽĨĨďĞƚǁĞĞŶƚǁŽĚŝƐƚŝŶĐƚ
ĐůŽƵĚƚŽĚĞůŝǀĞƌ ŐƌŽƵƉƐŽĨƵƐĞƌƐ͖ĂƚĂƉƌŽǀŝĚĞƌƐ;ŝŶĐůƵĚŝŶŐ/ŽdƐĞƌǀŝĐĞ
ƚŚĞŝƌŽǁŶƐĞƌǀŝĐĞƐ ƉƌŽǀŝĚĞƌƐͿĂŶĚĚĂƚĂĐŽŶƐƵŵĞƌƐ͘
ĨŽƌƚŚĞŵĂdžŝŵƵŵ
ŶƚĞƌƉƌŝƐĞƐ
ŶƵŵďĞƌŽĨĚĂƚĂ
ĐŽŶƐƵŵĞƌƐ͘

ĂƚĂ
ŐĞŶĞƌĂƚĞĚ
ďLJ
ĐŽŵƉƵƚĂƚŝŽŶ
ŽĨ ZĞŶƚŝŶŐͲͲ
ĞŶƚĞƌƉƌŝƐĞƐ DĂĐŚŝŶĞůĞĂƌŶŝŶŐʹ
ŽŵƉƵƚŝŶŐ ŽŵƉƵƚŝŶŐ ĂƚĂĂŶLJůƐŝƐ
/ŶĨƌĂƐƚƌƵĐƚƵƌĞ /ŶĨƌĂƐƚƌƵĐƚƵƌĞ

ůŽƵĚWƌŽǀŝĚĞƌ

Figure 1: Overview of the new cloud business model

the same data can be shared with more than one consumer for these players; and (3) it does not clarify how cooperating
at a time, which leads to an aggressive competition among entities would share their earned revenues.
data providers to sell their data even at lower prices. The Adopting traditional game theory concepts (e.g., Shapley
drawbacks of the merchant model are deply discussed in [5] value and Nash equilibrium) to distribute the revenue that
which alternatively proposes a two sided market model for results from the cooperation among the different parties suffers
monetizing personal data. In more detail, Bataineh et al.[5] from several limitations when applied in dynamic data trad-
propose an open market model in which individuals (actual ing scenarios over the cloud. Specifically, 1) although such
data owners) and data consumers trade data over a third party concepts might be highly efficient in scenarios wherein all
platform that helps them discover each other. The authors show the involved parties are rational, their effectiveness starts to
that the two-sided market outperforms the merchant model in decrease in the presence of parties that are heterogenious and
maximizing the total surplus. However, the main limitation prefer to deviate from the equilibrium points. For example, re-
of this approach is that it is based on a static analysis of cent studies have revealed that only 37% of the players tend to
consumer’ demand and data prices, which makes it unsuitable accept the Nash equilibrium in cooperative games (interested
for dynamic cloud markets. readers can consult behavioral games and ultimatum games
Under the umbrella of collaborative approaches, some pro- [8] for further details); and 2) even though the Shapley value
posals, for instance [6] and [7], tried to model the interactions approach fairly splits the revenues among the cooperative
among three entities in the domain of business-oriented IoT. entities based on their contributions, it becomes inapplicable in
In [6], the authors propose a model in which client peers are cases wherein the contributions of entities cannot be measured
interested in sharing video content with the help of the cloud. (which applies to the cloud scenario considered in this work).
In [7], the authors propose game theoretical models among Specifically, the cloud provider adds an ethereal/intangible,
IoT sensors, IoT service providers and data consumers. In yet significant, contribution to the coalition via introducing
these games, two entities (i.e., IoT sensors and IoT service the wide social networks of data consumers to those of data
providers) cooperate together in one game and then compete providers. On other hand, data providers own the data which
as one entity against data consumers. Such an approach suffers forms the core of this new business. This creates a continuous
from three drawbacks: (1) it does not consider the cross-group dilemma between data providers and cloud providers about
externalities (e.g., the mutual impact of the clientele size of who makes the most significant contribution to the coalition
one party on that of the other party) among the involved and hence who deserves the biggest share of the revenues.
parties, which makes it unable to capture the whole and more Equal distribution, so-called fifty-fifty, is one approach to split
concrete and realistic picture of the three-sided economical the revenues between the cloud provider and data provider.
model; (2) the cooperation and competition strategies adopted However, as mentioned before, the rationality and greediness
by the different players are highly impacted by the cross-group of the involved parties (i.e., the cloud provider and data
externalities which might not always lead to the best outcome provider) prohibit the success of such a strategy. This leads
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 3

us to the conclusion that we are dealing with a behavioral pump out more infrastructure units to increase the consumers’
and ultimatum game in which two players (proposer and demand so as to guarantee the highest possible level of revenue
responder) argue to split a certain amount of revenue. The portion.
proposer is endowed with a sum of revenue and is responsible To validate our solution, we conduct empirical experiments
for splitting this sum with the responder. The responder may using real-world data from Google and Amazon. Experimental
accept or reject the sum. In the case the responder accepts the results show that by following our solution, all the involved
sum, the revenue is split as per the proposal; otherwise, both parties (i.e., cloud platform, data providers and data con-
players receive nothing. sumers) achieve higher revenues than those achieved by the
Contributions. To solve the aforementioned problems, the traditional cloud computing business model.
two-sided market model [9], which is praised for its success WƌŝĐĞ
in modeling situations that involve brokers and cross-group
externalities, is investigated to study the cloud-based data
trading problem. The main idea of our solution is that the
cloud computing platform tries to attract data consumers ĂƚĂƉƌŽǀŝĚĞƌ ĂƚĂĐŽŶƐƵŵĞƌƐ
by offering them higher amounts of computing resources to
deploy their data analytic tasks. This in turn contributes in
attracting a larger number of data providers to reach the
cloud’s network of data consumers. Consequently, the data ůŽƵĚƉƌŽǀŝĚĞƌ
providers have incentives to offer higher portions of their
revenues to the cloud computing platform. Two-sided market
provides effective solution concepts for situations that are
characterized by a third-party platform connecting two other
parties. However, the main limitation of the two-sided market
theory is that it is effective in modeling scenarios in which Figure 2: Overview of the proposed two-sided game
the demand is static, but becomes less effective in elastic
environments that characterize cloud computing where the
demand is subject to dynamic and continuous changes. To II. R ELATED W ORK
address this problem, we integrate a novel game theoretical In this section, we provide a literature review on cloud
model, as shown in Figure 2, on top of the two-sided market computing business models. The existing proposals can be
model. The players of our game are (multiple) independent classified into two main categories: classical market and game
competing service providers (followers) and the cloud comput- theoretic-based pricing models. The proposals under the clas-
ing platform (leader). The players opt for hybrid cooperative sical market category such as [10, 11, 12] tackle the pricing of
and non-cooperative strategies, where strategies are modeled the cloud services using simple pricing models including those
as closed loops of dependencies. Data consumers and the of cost-based pricing, differential pricing, Ramsey pricing, and
cloud platform exhibit cross group externalities between each demand curve function. They model the pricing of cloud com-
other, where a higher demand from consumers leads to a puting resources as an optimization problem among multiple
revenue increase for the cloud platform and a higher supply cloud providers and cloud users. However, the main drawback
of computing resources from the cloud creates more demand of these approaches lies in their static pricing strategy which
from consumers. does not suit the highly variable and dynamic environment of
In the first stage of the game, the leader (cloud platform) cloud computing.
announces the desired portion of returned revenues out of On the other hand, the game theoretical models consider the
the data providers’ gain, and then in the second stage, data instantaneous interactions that might occur among the involved
providers decide about their pricing strategy for data con- entities and their effects on each party’s welfare. The objective
sumers. The resulting equilibrium forces the cloud platform is to dynamically capture the optimal price and distribution of
to offer higher and reasonable supply of computing resources the cloud computing resources. Many proposals such as [13,
to guarantee maximal levels of revenues, while not showing 14, 15, 16, 17] applied different approaches including games
greedy behavior in terms of its share of data providers’ and machine learning [18] to the cloud resource allocation
revenue. Moreover, following our solution, the data providers and pricing problem. In [13], the authors propose an economic
are forced to offer the cloud platform a higher portion of model based on a Stackelberg game to trade video contents and
their revenues to ensure appropriate Quality of Service (QoS) movies over a cloud platform. The proposed model formulates
delivered to data consumers. In the case of a greedy behavior the interactions between a service provider (e.g., Netflix) and
from the cloud, our game uses a subsidizing mechanism. This end users. The service provider acts as the game leader and
mechanism pushes data providers to increase the shared por- aims to minimize the cloud bandwidth consumption while
tion offered for the cloud to sustain high and reasonable levels guaranteeing at the same time users’ satisfaction. The work
of computing infrastructure so as to guarantee high levels of in [19] models the interactions among multiple Software as
consumers’ demand. Similarly, in cases where data providers a Service (SaaS) providers and Infrastructure as a Service
behave greedily by offering small portions of revenues to (IaaS) provider as a two-stage Stackelberg game. In the first
the cloud, the subsidizing mechanism pushes the cloud to stage of the game, SaaS providers determine the number of
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 4

required VM instances while accounting for both the QoS 1

delivered to their users and the associated costs. In the second Ds1
stage, the IaaS providers seek to maximize their revenues in P1
the light of the bids done by the SaaS providers [20, 21]. (CS1)
The author in [6] proposes an economic model in which
1 (SP1) 1
cloud users seek to share video content with other users
over the cloud. The model is solved using both cooperative .
.
and non-cooperative games between the cloud and its users. .
Similar studies are investigated in [7, 16] for different cloud 
applications. The authors in [22] propose a game theoretical Dsi
model to deliver a bundle of complementary IoT services. The Pi
proposed solution studies the merchant-consumer scenario in (CSi)
i
which the IoT services are directly traded between the service (SPi)

providers and service consumers without the intervention from
any third party. However, this solution cannot be adopted in .
.
our case, where the cloud computing is not the actual data .
owner and hence it cannot monetize the data directly for n
the consumers. Nevertheless, the cloud computing (the third
Dsn
party in our paper) is considered as a global market where Pn
the data services and data consumers meet each other, thus
(CSn)
increasing their market shares. The authors in [23] and [24] Data
Interfaces Servers
n
introduce a market model for managing, trading, and pricing Data
n
(SPn)
Consumers
big data services. Both proposals use the two-sided market Cloud Pla
orm
(CS) Data Service
theory in order to provide incentives for both cloud providers Providers (SP) (CP)
and users to increase their data shares. The work presented in
Figure 3: Two-sided model
[5] extends the work proposed in [23] and comprehensively
studies the two-sided market model as a successful model for
monetizing personal data. However, these proposals consider
a static environment in which the demands on cloud resources denotes Consumers of Service i), big data service providers
are computed in a static manner, which makes them unable to SP (a Service Provider providing service i is denoted SPi )
accommodate the cloud’s elasticity property. and a typical Cloud Platform (CP). The cloud platform, such
To the best of our knowledge, the proposed work is the first as Google and Amazon, is a market leader with huge com-
that addresses big data services monetization, while consider- puting and storage capabilities, capitals, and social consumer
ing the cross-group externalities among the involved entities. networks. In our model, a certain big data service provider SPi
Unlike the classical cloud computing business model (where that provides a service i deploys its service on the cloud and
the main challenge is how to optimize the cloud utilization receives a monetary value of Pi for each consumer access to
while incorporating only operational cost and QoS metrics), its service i. The cloud platform CP is in charge of sustaining
our approach : 1) supports and helps junior big data service the consumer access through providing the needed computing
providers especially those that have limited monetary budgets; and storage infrastructure including hardware, software and
2) uses the two-sided market theory to model the interac- security services. The relationship between consumer CSi ’s
tions among the involved parties, while all above-discussed demand, denoted by Dci , and the computing and storage
proposals use the classical merchant model; (3) includes a resources Dsi supplied to CSi is modeled using the two-sided
subsidizing technique to push the resulting equilibrium toward market model as cross group externalities α and β . Here, α
a Pareto optimal point. On the other hand, the above-discussed represents the benefits that a consumer obtains when some
proposals adopt the fairness criterion that rewards the involved new computing and storage resources are added to Dsi and β
parties based on their contributions. We also differ from the represents the amount of benefits that the cloud platform earns
other proposals that adopt the two-sided market theory by when more new consumers are added to Dci . The parameters
providing a dynamic pricing method, instead of a static game α and β are dependant on the service i. However, instead
theoretic-based pricing strategy. of using the notations αi and βi , the index i is omitted to
simplify the equations where the service i is understood from
III. P ROPOSED B IG DATA S ERVICES M ONETIZATION the context. The same simplification is used for the other
M ODEL OVER THE C LOUD : A T WO - SIDED G AME M ODEL parameters that appear as powers (exponents) in our equations.
We explain in this section the details of our proposed Big The interaction between SP and CP is modeled as a two-
data services monetization. stage game where CP acts as the game leader and SP are the
followers. In the first stage of the game, each service provider
A. Solution Architecture and Game Formulation providing service i SPi observes the amount of money returns
The proposed cloud market platform, depicted in Figure χi requested by CP, in order to adjust the price to be charged to
3, consists of three entities: consumers of services CS (CSi CSi . In quest of the price specified by SPi , CP determines the
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 5

optimal amount of computing and storage resources Dsi that used γ and ψ (see Table I). These two parameters depend on
should be supplied to CSi . The model forms a closed loop of the service i, which is omitted from the notations for simplicity
dependencies that involves techniques from Stackelberg and as mentioned earlier. In our model, the consumer’s demand
Ultimatum game theory as well as a subsidizing technique. (Dci ) is a function of Pi and Dsi as shown in Equation (1).
In the Stackelberg game, the interactions take place in two
stages where the leader (CP) makes the first move and then −γ
Dci = k1 Pi Dαsi (1)
does each follower (SPi ) after having observed the leader’s
move. In the ultimatum game, the first player (CP) proposes Dsi is given in Equation (2). Clearly, higher consumers’
a strategy to divide the amount of returned revenue with demands would have an influence on the quantity of supplied
the second player (SPi ). In case SPi rejects the offer, neither resources. The cloud platform CP uses more computing and
player gains anything. Otherwise, the first player gets the storage resources to keep up with the increasing number of
amount it requested and the second player gets the rest. In consumer accesses, to maintain a high quality level. The
φ
the subsidizing technique, SPi may chose to subsidize CP by parameter χi represents the cloud platform’s preferences (i.e.,
an extra amount of payment that exceeds the contribution of desired profit) and implicitly captures the rationality of both
this CP. The objective is to keep an optimal level of Dsi that CP and SPi . In fact, it reflects the level of perfect/imperfect
maximizes the return revenues Pi ∗ Dci . Alternatively, CP may information that CP and SPi have about one another. High
subsidize SPi by low portion of the resulting revenues to keep elasticity φ is caused either by a greedy monopolist cloud
an optimal level of Pi . The different parameters and symbols platform or by a weak service with few capitals accepting
used in our proposed solution are depicted in Table I. small portions of returns on profits. The parameter φ depends
on the service i, but as mentioned earlier, the index i is omitted
Model Parameters Descriptions. when the service i is understood from the context. The charged
SPi Service provider providing service i.
price Pi also positively contributes to Dsi . We can arguably
CP A typical cloud platform.
claim that charging consumers with higher prices Pi forces
CSi Consumers of service i.
CP to provide more computing and storage resources so as
Dci CSi ’s demand.
to satisfy the consumers’ needs. Modeling Dsi as a function
Dsi IT-infrastructure supply to CSi .
of χi and Pi with different elasticity values connects CP and
Pi Service i’s price.
SPi strategies with each other, which captures the sensitivity
φ Dsi ’s elasticity with respect to χi .
of CP to SPi ’s strategy (i.e., structure of the charged price
χi Portion of revenue required by CP from SPi .
and shared portion), and highlights the importance of the
γ Dci ’s elasticity with respect to Pi .
subsidizing technique. This aspect is illustrated and discussed
β The Network effects (externality) on Dsi by Dci
further in the simulation section (Section IV-D).
ψ Dsi ’s elasticity with respect to Pi .
φ ψ
α The Network effects (externality) on Dci by Dsi . Dsi = k2 χi Pi Dβci (2)
fc Associated costs per service consumer.
By substituting Equation (2) into Equation (1) and vice versa,
k1 , k2 Constant multipliers.
we can express Dci and Dsi as functions of Pi and χi as follows:
πi SPi ’s payoff.
fs Associated costs per IT-infrastructure unit.
π Cloud platform’s payoff.
Dci = (k1 k2α Pi−a1 χia4 )1/a2 (3)
a1 = γ − αψ.
a
= 1 − αβ .
Dsi = (k2 k1 Pi 3 χi )1/a2
a2 β φ
(4)
a3 = ψ − γβ .
a4 = αφ Each big data service provider SPi is subject to a fixed cost
fc per each consumer access. SPi aims to maximize its payoff
Table I: Model parameters
as given in Equation (5). We express the service provider’s
payoff πi as a function of Pi and χi by substituting Equation
(3) into Equation (5) and taking the log for both sides as shown
B. Players’ Demand and Utility Functions in Equation (6).
A precise estimation of the needed computing and storage
resources requires a price estimation mechanism for the num- πi = ((Pi )(1 − χi ) − fc )Dci (5)
ber of consumers and the variation of their demand with re-
spect to the provided QoS. To do so, we define the consumer’s
demand and supply using the Cobb-Douglas function that log πi = log(Pi (1 − χi ) − fc ) + (1/a2 )(log k1 k2α
effectively captures the elasticity of the computing and storage (6)
− a1 log Pi + a4 log χi )
resources supply (Dsi ) and its variations for each specific user’s
demand. This elasticity is a characteristic property of cloud The cloud platform CP is subject to a fixed cost fs per
computing environments. The demand functions we use are each unit of computing and storage resources. The CP aims
continuous, concave or convex, and capture the elasticity with to maximize its payoff as given in Equation (7). We express
respect to each input parameter. Two elasticity parameters are the cloud platform’s payoff π as a function of Pi and χi by
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 6

substituting Equations (3) and (4) into Equation (7) as shown

in Equation (8). ∂ Dci −a1
= Dc (15)
∂ Pi a2 Pi i
π = Pi χi Dci − fs Dsi (7) By substituting Equation 15 into Equation 14, then
∂ πi a1
a a4 a3
= (1 − χi )Dci − (Pi (1 − χi ) − fc )Dci (16)
1 1− a1 a2 +1 β 1
a2
φ
a2
∂ Pi a2 Pi
π = (k1 k2α ) a2
Pi 2 χi − fs ((k2 k1 ) a2
Pi χi ) (8)
By rewriting Equation (16) using Equation (5), then
∂ πi a1
C. Game Equilibrium = (1 − χi )Dci − πi (17)
∂ Pi a2 Pi
The equilibrium of the above-described game is solved
using a backward induction methodology. Thus, the followers’ ∂ 2 πi −(1 − χi )a1 a1 ∂ πi a1
2
= Dc i − + πi (18)
(service providers) sub-game is solved first to obtain their ∂ Pi a2 Pi a2 Pi ∂ Pi a2 Pi2
response Pi to the service consumers. The leader’s (cloud
By simplifying Equation (18) and substituting Equation (12),
platform) sub-game is then computed considering all the
we obtain:
possible reactions of its followers to maximize its payoff [25]. ∂ 2 πi Dc a1
Every service provider SPi determines its optimal decision Pi∗ , = i (1 − )(1 − χi ) (19)
∂ Pi2 Pi a2
while considering the CP’s optimal decision χi∗ as an input
parameter. The players’ best responses are discussed in the Since Dci and Pi are always positives, then
following.
∂ 2 πi a1 a1
< 0 ⇒ (1 − ) < 0 ⇒ >1 (20)
Theorem 1. The best responses in the two-sided game are as ∂ Pi2 a2 a2
follows:
Similarly, to obtain the optimal χi∗ , we derive Equation (8)
1) The best response of the service provider SPi is given with respect to χi as given by Equation (21):
by:
a1 fc
Pi∗ = (9) a a β 1
a
(a1 − a2 )(1 − χi ) ∂π 1 a
4 1− a1 a4 φ fs (k2 k1 ) a2 a32 aφ2 −1
a1 a1
= (k1 k2α ) a2 ( +1)Pi 2 χi 2 −( )Pi χi =0
if: > 0 and >1 ∂ χi a2 a2
a1 −a2 a2 (21)
2) The best response of the cloud platform with respect to a4 −φ
a2 +1 φ k2 k 1
β a1 +a3
a2 −1
a service i is given by: χi = fs ( )( 1α ) a2 Pi (22)
a4 + a2 k1 k2
a4 −φ a1 +a3
+1 φ
a2 a2 −1 By substituting Equation (12) in Equation (22), we get:
χi (1 − χi ) = fs × ( )
a4 + a2
β (10)
k2 k1 a1 a1 fc a1a+a3 −1 a4 −φ
+1 a1 +a3
φ
β
k2 k 1 a1 fc a1a+a3 −1
×( ) 2 ×( ) 2 χi
a2
(1− χi ) a2 −1
)( 1α ) a2 (
= fs ( ) 2
α
k1 k2 a1 − a2 a4 + a2 k1 k2 a1 − a2
if: a4 + a 2 − φ < 0 (23)
To verify the type of χi∗ ’s optimality, we compute a second
Proof. Consider the service payoff given by Equation (6), the derivative test by deriving Equation (21) as given by Equation
optimal price Pi∗ is defined by ∂ πi /∂ Pi = 0 as follows: (24):

1 ∂ πi 1 − χi a1 a2 −a1 a4
× = − =0 (11) ∂ 2π 1 a (a + a )
4 4 2 a2 a −1
πi ∂ Pi Pi (1 − χi ) − fc (a2 )Pi 2
= (k1 k2α ) a2 ( )Pi χi 2
∂ χi a22
⇒ 1 (24)
a1 f c β a
Pi∗ = (12) fs φ (φ − a2 )(k2 k1 ) a2 a32 aφ2 −2
(a1 − a2 )(1 − χi ) −( )Pi χi <0
a22
Since Pi∗ is always positive, then
By substituting Equation (22) in Equation (24), we obtain:
a1
>0 (13)
a1 − a2 a4 + a2 − φ < 0 (25)
To verify the type of Pi∗ ’s optimality, i.e maximum or mini-
mum, we compute a second derivative test by deriving Equa-
tion (5):
IV. S IMULATIONS AND E MPIRICAL A NALYSIS
∂ πi ∂ Dci In this section, we evaluate the performance of the proposed
= (1 − χi )Dci + (Pi (1 − χi ) − fc ) (14)
∂ Pi ∂ Pi two-sided game solution in comparison with the fifty-fifty and
By deriving Equation 3, then the pay-as-to-go approaches in terms of total surpluses of
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 7

Simulaon Inputs and its consumption of EC2 instances are represented by SPi
α
 Pi K1 Ps fs fc (service data provider) and Dsi respectively. The mean of the
supply function Dsi consists of 32 EC2 instances. However,
enterprise customers have varying business types and hence
Simulaon Scenario vary in terms of the amount of needed cloud resources. To
model this variation in our simulations, the customers’ demand
Two-sided market Fi
y- y Pay-as-to-go
scenario scenario scenario on EC2 instances is normally distributed around the mean with
a standard deviation of 10. This means that the co-domain of
the supply function Dsi ranges from 1 to 53 EC2 instances.
The real dataset [26] registers the log file of computational big
Total Cloud Total data Total data Total Cloud

Total data Total data Total Cloud Total data Total data
 data jobs executed by tremendous enterprise customers over
payo  providers consumers payo providers consumers

payo providers consumers

payo demand payo demand payo demand
similar instances of EC2. This dataset helps us extract reliable
Simula on Outputs
ranges of consumers’ demands Dci as well as the externalities
Figure 4: Simulation overview α and β as described in what follows. The computational
power of each instance, extracted from the same dataset [26], is
normally distributed with a mean of 0.38 job per second and a
involved parties (i.e., payoffs of the cloud platform, service standard deviation of 0.1. The average computational power is
data providers and service data consumers). Specifically, we represented in the proposed model by the externality factor α,
aim to: 1) verify the effectiveness of the proposed game vis- which means that α ranges from 0.1 to 0.7. According to the
à-vis the current cloud computing business model (i.e., the assumption presented in [9], the cross group externalities fac-
pay-as-to-go model); 2) study the equilibrium of the two- tor should be bounded by 0 and 1, i.e., 0 < αβ < 1. Hence, the
sided game in presence of the fifty-fifty choice (i.e., egalitarian externality factor β would range from 0 to 1/α. A consumer’s
choice in ultimatum games) which is the typical solution demand Dci on each enterprise ranges from 0.1×1 to 0.7×53,
for such games; and 3) investigate the impact of the model which is 0.1 to 37 requests per second. The service price,
parameters on the performance of our solution. Figure 4 shows Pi , is estimated through observing the prices of 150 business
an overview of our simulation setting in terms of inputs, intelligence computing services including big data and IoT
scenarios, and results. The simulation inputs and scenarios are services located in the the AWS marketplace [28]. According
described in Sections IV-A and IV-B respectively, while the to the observed prices, Pi is normally distributed with a mean
simulation results are discussed in detail in Sections IV-C, of 1.7 USD/hour and a standard deviation of 0.5 USD. This
IV-D and IV-E. means that the service prices range from 0.2 USD/hour to
3.2 USD/hour. The parameter φ represents the greediness of
the cloud platform with respect to the service providers. The
A. Simulation Setup subsidizing factor 0 < φ < 1 represents the rational behavior
In this section, we conduct a simulation analysis grounded (subsidizing behavior) of the cloud, while 1 < φ represents the
on statistical observations from BMR [1] and real data from greedy behavior of the cloud platform. The price elasticities
[26]. According to [1], in 2019, Amazon Web services (AWS) are set up between 0.1 − 0.35 (i.e., γ), which are similar to the
received 30 billion USD in revenue with net income around 10 sensitivity of mobile/telecommunication services price shown
billion USD from 1 million active customers running monthly in the literature [29]. We assume that k2 = 1 in our simulation.
70 million hours of their enterprises on custom instances of By substituting the expected values of α, Dsi , Dci , γ, and Pi
Elastic Compute Cloud (EC2). So, 1) an enterprise customer into Equation (1) and considering the assumption (k2 = 1),
spends on average up to 30, 000 USD per year in monthly we find that the multiplier k1 ranges from 0.1 to 0.99. It
renting 70 hours of cloud resources; and 2) the marginal oper- is worth mentioning that consumers demand (Dci ) and cloud
ating costs for the cloud platform is 66% of revenues (amazon resources (i.e., computing and storage resources) Dsi supplied
received 10, 000 USD as net payoff from each consumer). By to CSi are only estimated under simulation setup to extract
entering these numbers in the Amazon calculator [27], we can a suitable range for the multiplier k1 , but they are not given
conclude that the customer rents on average 70 hours monthly as simulation inputs. The simulation calculates the expected
of 32 instances of Amazon EC2 where each instance includes consumer demand and the optimal supply of cloud resources
16 VMs, 30 GB of Memory, and 1000 GB of hard disk storage as explained in Section IV-B. The values of all associated
at rate 36 USD/hour. The price rate (36 USD/hour) is denoted parameters are summarized in Table II.
by Ps , which will be used later to calculate the cloud and data
providers payoffs in the pay-as-to-go model as explained in B. Simulation Scenarios
Section IV-B. The cloud provider (amazon) entails 66% of We consider a group of 300 data service providers in the
instances price (36 USD/hour) as operating costs, which is cloud under three scenarios: 1) proposed two-sided game; 2)
23.7 USD/hour. The operating costs are denoted in our model fifty-fifty scenario which follows our model except that the
by fs . In fact, 40% of revenues as a profit and 60% as an cloud platform and data provider agree to share the revenue
operation cost are common in business. Thus, we assume the equality; and 3) pay-as-to-go scenario, which is the current
marginal cost of data consumers ( fc ) entailed by data providers business model adopted by the main cloud providers such as
has the same distribution as ( fs ). The enterprise customer Amazon and Google.
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 8

System Parameters Values Algorithm 2 Fifty-fifty scenario

Pi 0.2 - 3.2 USD per hour Input: α, β ,γ, k1 , fs , fc
φ 0-5 Output: CloudPayoff, ProvidersPayoff, ConsumersDemand
α 0.1 - 0.7 1: for each Data service provider SPi do
β 0 - 1/α 2: χi ← 0.5
γ 0 - 0.35 3: φ ←1
Ps 36 USD per hour 4: SPi declares Pi through maximizing Equation (5)
k1 0.1 − 0.9 5: The cloud calculates Dsi through maximizing Equation
(7)
Table II: Simulation parameters values 6: Equation (1) is used to determine ConsumersDemand
7: Equation (5) is used to determine ProvidersPayoff
8: Equation (7) is used to determine CloudPayoff
1) Two-sided scenario: The two-sided model, explained in 9: end for
details in Section III-A, is described in Algorithm 1. Given
a data service price Pi , the cloud platform determines the
optimal portion of revenue χi and the required amount of cloud the pay-as-to-go model since the cloud platform and data
resources that maximize its payoff. consumers do not directly interact nor do they exhibit mutual
cross group externalities between each other. The pay-as-to-go
Algorithm 1 Two-sided scenario
model is described in Algorithm 3. Given a price rate of Ps
Input: α, β , φ ,γ, k1 , fs , fc USD/hour, the data provider determines the optimal amount
Output: CloudPayoff, ProvidersPayoff, ConsumersDemand of rented cloud computing infrastructure through maximizing
1: for each Data service provider SPi do Equation (26). The optimal amount of rented cloud computing
2: SPi declares its price Pi infrastructure D∗si is given in Equation (28) through substituting
3: χi is calculated through maximizing Equation (5) Equation (1) with Equation (26) and computing its derivatives
4: The cloud calculates Dsi through maximizing Equation with respect to Dsi .
(7)
5: Equation (1) is used to determine ConsumersDemand
6: Equation (5) is used to determine ProvidersPayoff
7: Equation (7) is used to determine CloudPayoff ĂƚĂƐĞƌǀŝĐĞ ZĞŶƚŝŶŐƉƌŝĐĞ;WƐͿ

8: end for ĂƚĂ

ŽŶƐƵŵĞƌƐ ĂƚĂWƌŽǀŝĚĞƌ ůŽƵĚWƌŽǀŝĚĞƌ

2) Fifty-fifty scenario: The egalitarian (fifty-fifty) scenario

follows the two-sided market model in terms of consumer
WƌŝĐĞŽĨĚĂƚĂƐĞƌǀŝĐĞ;WŝͿ ŽŵƉƵƚŝŶŐŝŶĨƌĂƐƚƌƵĐƚƵƌĞ;ƐͿ
demand and supply function, thus considering the externalities Figure 5: Pay-as-to-go model
among the involved parties (i.e., Equations (1) and (2)). Thus,
the utilities of the cloud platform and data providers are
formalized using the same payoff equations used in the two-
πi = (Pi − fc )Dci − Ps Dsi (26)
sided model (i.e., Equations (7) and (5) ). However, under
this scenario, the cloud platform requests 50% of the revenue
and hence the subsidizing factor φ is reset by 1. The fifty- π = (Ps − fs )Dsi (27)
fifty scenario is described in Algorithm 2. Specifically, given
a shared portion χi = 0.5 and a subsidizing factor φ = 1, the Ps 1
data provider determines its optimal service price Pi through D∗si = ( −γ ) α−1 (28)
αk1 (Pi − f c)Pi
maximizing its payoff given in Equation (5). Thereafter, the
cloud platform calculates the optimal computing infrastructure
Dsi by maximizing its payoff given in Equation (7). Algorithm 3 Pay-as-to-go scenario
3) Pay-as-to-go model: Figure 5 depicts the pay-as-to-
Input: α, β ,γ, Ps , k1 , fs , fc
go model. As shown in the figure, the data provider rents
Output: CloudPayoff, ProvidersPayoff, ConsumersDemand
the cloud computing infrastructure (Dsi ) for a price of Ps
1: for each Data service provider SPi do
USD/hour. Thereafter, the data provider delivers its own data
services to the data consumers for a price of Pi USD/hour. 2: The cloud declares Ps
The data provider and cloud payoffs under the pay-as-to-go 3: SPi declares Pi
model are given in Equations (26) and (27) respectively. The 4: SPi calculates Dsi using Equation (28)
consumer’s demand function under the pay-as-to-go model 5: Equation (1) is used to determine ConsumersDemand
has the same characteristics as under the two-sided market 6: Equation (26) is used to determine ProvidersPayoff
model. This means that the consumer’s demand is formalized 7: Equation (27) is used to determine CloudPayoff
8: end for
as given in Equation (1). However, the provided cloud com-
puting infrastructure (Dsi ) is not given as a function under
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 9

C. Sensitivity analysis of externalities efficient outcomes when the externalities become stronger (i.e
We first investigate in Figures 6, 7, and 8 the impact of αβ ∈ [0.3−0.7]) where the cloud platform receives less payoff
the externality parameters (αβ ) on the payoffs of the cloud under certain subsidizing factor such as (φ = 1.5) or (φ = 2).
platform, data providers, and data consumers respectively. We On the other hand, the fifty-fifty choice and the two-sided
run the simulation with different ranges of the subsidizing model show similar outcomes in terms of data providers’
factor (φ ), i.e., 0.5, 1.0, 1.5, 2.0, and 5.0. Those values of payoff and consumers demand as shown in Figures 7 and 8
externality and subsidizing factor are given as inputs to the respectively. However, as mentioned in Section I, the involved
simulation program to adjust the strategies of the involved parties do not follow the egalitarian choice despite its good
players. Specifically, the cloud platform adjusts the amount of outcomes in some cases, mainly because of their greediness
provided infrastructure (Dsi ) and the demanded portion (χi ), and because of high subsidies in real cloud markets.
while the data provider (SPi ) calculates the impact of variation
in (Dsi ) and (χi ) on the expected demand of data consumers
(Dci ) and accordingly adjusts its price (Pi ).
In Figure 6 , we study the impact of varying the externality
parameters (αβ ) and subsidizing factor (φ ) on the cloud’s
payoff. As shown in the figure, the cloud platform obtains in
general higher payoff when it follows the two-sided model,
rather than the pay-as-to-go model. For example, under the
two-sided model, the cloud platform receives 1200, 500 and
400 USD as payoff when (αβ ) = 0.2 and φ = 5.0, 2.0 and 1.5
respectively. On the other hand, under the pay-as-to-go model
and under the same externality and subsidizing parameters, the
cloud platform receives less payoff of (200) USD. Similarly,
data providers receive higher payoff and data consumers’
demand is increased under the two-sided model compared to
the pay-as-you-go model as shown respectively in Figures 7
and 8 .

Figure 7: Data providers payoff over externalities αβ

As can be observed from these figures, the two-sided market

model shows less efficient outcomes as the externalities be-
come stronger. Specifically, the cloud platform, data providers
and consumers’ surpluses gradually decrease as the values of
the externality parameters increase. For example, in Figure
7, the data providers receive a payoff of 800 USD at an
externality parameters value of (αβ = 0.2) and subsidizing
factor of φ = 5.0, which is higher than that received with
externality parameter value of (αβ = 0.3) and subsidizing
factor of φ = 5.0. This decrease continues until the two-sided
market model reaches almost the same efficiency of the pay-
to-go model under the strong externality values of [0.55 -
0.7]. Similar behavior is observed in terms of cloud payoff
and data consumers’ demand as depicted in Figures 6 and 8
respectively. In fact, the cloud’s payoff and data consumers’
demand are higher under weak externality values, i.e., [0−0.5]
Figure 6: Cloud payoff over externalities αβ rather than in strong externality values [0.5 − 0.7] . The reason
behind such a behavior is that the cloud platform needs to
In addition, we study the impact of the fifty-fity (egalitarian) provide more computing and storage resources under strong
choice on the two-sided market model, where the cloud externalities to attract smaller number of data consumers,
platform and data providers share fifty percent of revenues which adds more costs and then leads to less payoff.
considering the sitting of our two-sided model. As shown in Figure 9 shows the number of computing infrastructure
Figure 6, the fifty-fifty choice shows more efficient outcomes units provided by the cloud platform over varying externality
than the two-sided model in terms of cloud payoff under values. As shown in the figure, at a subsidizing factor of
week externalities (αβ ∈ [0.1 − 0.3]) and certain values of φ = 2.0 and externality values of (αβ = 0.2) and (αβ =
subsidizing factor φ . However, the cloud platform can receive 0.53), the cloud platform provides 4800 units of computing
higher payoff if it chooses higher subsidizing factor φ such infrastructure. However, in Figure 8, at a subsidizing factor
as φ = 5. Nevertheless, the egalitarian choice shows less of (φ = 2.0) and externality values of (αβ = 0.2), the cloud
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 10

φ
on the size of its contribution. This leads to having χi = χi ,
meaning that the provided computing infrastructure is linearly
probational with respect to the market share χi . In this case,
neither the cloud platform nor data providers act greedily, and
they do not subsidize each other at the same time.

Figure 8: Data consumers demand over externalities αβ

platform attracts (3.3 × 106 ) data consumers, while it attracts

a smaller number of data consumers, i.e., (0.5 × 106 ) with
externality values of (αβ = 0.53). In other words, the cloud
platform attracts (3.3 × 106 ) of data consumers by providing
Figure 9: Cloud Infrastructure over externalities αβ
4800 units of computing infrastructure at externality values
of (αβ = 0.2), while it attracts (0.5 × 106 ) data consumers
In Figures 6, 7 and 8, we consider the impact of the sub-
by providing the same number of computing infrastructure
sidizing factor φ in the sensitivity analysis of the externalitiy
units with externality values of (αβ = 0.53). Consequently, in
parameters. As can be observed from those figures, the cloud
such a case, the cloud platform asks for a higher portion χi
platform and data consumers receive higher payoff than the
of revenues to maximize its payoff, which negatively affects
data providers as the subsidizing factor increases. For example,
the payoff of the data providers. In Figure 10, we study the
in Figure 6, the cloud’s payoff is higher under a subsidizing
impact of the externality parameters on the portion of revenues
factor of φ = 2.0 than it is under a subsidizing factor of
asked by the cloud platform. As shown in the figure, the cloud
φ = 1.0. Similarly, we can observe in Figure 8 that data
platform asks for higher portions as the values of externality
consumers’ demand is higher under a subsidizing factor of
parameters are increased. For example, at a subsidizing factor
φ = 2.0 than it is under a a subsidizing factor of φ = 0.5.
of φ = 1.5 and externality values of (αβ = 0.3), data providers
Unlike the cases of cloud and data consumers, data providers’
share 40% of their revenues, while they share 68% of their
payoff is less under φ = 5.0 than it is when φ = 1.5, as depicted
revenues with externalitiy values of (αβ = 0.6) and a constant
in Figure 7.
subsidizing factor of φ = 1.5.
To better clarify the impact of the subsidizing factor on the
payoffs of involved parties in our model, we run the simulation
D. Sensitivity analysis of subsidizing factor and greedy be- over a continuous, reasonable and wider range of subsidizing
havior of involved parties factor values. Specifically, we describe the payoff of the
φ
The exponential function χi captures the rational/greedy cloud, data providers and data consumers as a function of the
behavior of the cloud platform and data providers. The subsi- subsidizing factor φ in Figures 11, 12 and 13 respectively. The
dizing factor φ implicitly describes the reactions of the cloud results shown in these figures confirm the insights extracted
platform to the sharing portions offered by data providers. from Figures 6, 7 and 8. Furthermore, we notice in Figures
Theoretically, the cloud platform subsidizes the data providers 12 and 13 that the payoff of the data providers and consumers
by imposing a subsidizing factor φ that is less than 1. This increases when the value of φ is between 0 to 1. Then slightly
φ
leads to having χi > χi since the base χi is defined to be decreases /stabilizes as φ becomes greater than 1. On the
between 0 and 1, meaning that larger amounts of computing other hand, the cloud’s payoff, as shown in Figure 11, largely
infrastructure units Dsi should be provided. On the other hand, increases when φ becomes greater than 1.
data providers offer higher portions χi when the cloud platform This behavior can be practically interpreted with the results
acts greedily by imposing a subsidizing factor φ that is greater shown in Figure 10. In this figure, we notice that data
φ
than 1. This leads to having χi < χi , meaning that smaller providers react to the greedy behavior of the cloud platform
amounts of computing infrastructure units Dsi need to be via increasing the shared revenue, which negatively affects
provided. When the subsidizing factor is equal to 1, the data the payoff of data providers as shown in Figures 7 and 12,
providers offer a market share to the cloud platform based but positively affects the cloud’s payoff as shown in Figures
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 11

Figure 10: Shared revenue among the cloud and data providers Figure 12: Data providers payoff over the subsidizing factor
(χi ) over externalities αβ φ

Figure 13: Consumer demand over the subsidizing factor φ

Figure 11: Cloud payoff over the subsidizing factor φ

6 and 11.

However, the behavior of the data providers (reacting to

the greedy behavior of the cloud platform via increasing the
shared portions) sustains a higher level of consumers’ demand
as shown in Figures 11 and 8. Similarly, as shown in Figure 10,
the cloud platform acts to the low offered portion by imposing
a subsidizing factor φ that is less than 1. This negatively
impacts the cloud’s payoff but positively affects the cloud
platforms’ payoff. In summary, we conclude that our game
includes a recovery mechanism that helps sustain efficient
payoff outcomes for all the parties, in case any of the involved
parties decides to act greedily toward the others.
Figure 14: Cloud payoff over demand elasticity γ
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 12

E. Sensitivity analysis of consumer demands elasticity (γ) and

the multiplier (k1 )

We now move to analyzing the impact of consumers’

demand elasticity on the surpluses of all involved parties
(Figures 14 - 16). A high negative elasticity value positively
impacts the surplus of all parties under the two-sided game
model through yielding higher payoffs for the cloud platform
and data providers. The reason is that the exponential function
−γ
Pi increases when γ increases from 0 to 0.34 as long as the
base Pi is less than 1. This makes data providers able to charge
their consumers higher prices without significantly leading to
a decrease in the demands.

Figure 17: Cloud payoff over multiplier k1

Figure 15: Data providers payoff over demand elasticity γ

Figure 18: Data providers payoff over multiplier k1

Figure 16: Consumers demand over demand elasticity γ

Figure 19: Consumers demand over multiplier k1

JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 13

We also study the impact of the multiplier k1 on the [14] R. Pal and P. Hui, “Economic models for cloud service markets: Pricing
surplus of all involved parties. As shown in Figures 17 - 19, and capacity planning,” Theoretical Computer Science, vol. 496, pp. 113
– 124, 2013.
the surpluses of the cloud platform, service data providers [15] H. Xu and B. Li, “A general and practical datacenter selection framework
and consumers increase as the value of the multiplier k1 for cloud services,” in 2012 IEEE Fifth International Conference on
increases. The reason is that increasing the multiplier k1 leads Cloud Computing, 2012, pp. 9–16.
[16] J. Ding, R. Yu, Y. Zhang, S. Gjessing, and D. H. K. Tsang, “Service
to increasing consumers’ demands, which positively affects provider competition and cooperation in cloud-based software defined
the total revenues. wireless networks,” IEEE Communications Magazine, vol. 53, no. 11,
pp. 134–140, 2015.
[17] P. Samimi and A. Patel, “Review of pricing models for & cloud
V. C ONCLUSION computing,” in 2011 IEEE Symposium on Computers & Informatics,
2011, pp. 634–639.
In this paper, we proposed a game theoretical model based [18] O. A. Wahab, A. Mourad, H. Otrok, and T. Taleb, “Federated machine
on the two-sided market theory to monetize big data services learning: Survey, multi-level classification, desirable criteria and future
over the cloud. In particular, we studied the impact of cross directions in communication and networking systems,” IEEE Communi-
cations Surveys & Tutorials, 2021.
group externalities between the amount of provided cloud [19] V. D. Valerio, V. Cardellini, and F. L. Presti, “Optimal pricing and service
resources and number of service consumers. The objective is provisioning strategies in cloud systems: A stackelberg game approach,”
to come up with a new vision in which the cloud can play in IEEE CLOUD. IEEE Computer Society, 2013, pp. 115–122.
[20] N. C. Luong, P. Wang, D. Niyato, Y. Wen, and Z. Han, “Resource
a primordial role in introducing big data service providers management in cloud networking using economic analysis and pricing
and data consumers to each other, which results in higher models: A survey,” IEEE Communications Surveys Tutorials, vol. 19,
benefits for all the involved players. To achieve this goal, we no. 2, pp. 954–1001, 2017.
[21] N. C. Luong, D. T. Hoang, P. Wang, D. Niyato, D. I. Kim, and
designed a game theoretical model in which the cloud platform Z. Han, “Data collection and wireless communication in internet of
and big data service providers engage in a closed loop of things (IoT) using economic analysis and pricing models: A survey,”
dependencies that makes them interested in satisfying each IEEE Communications Surveys Tutorials, vol. 18, no. 4, pp. 2546–2590,
2016.
other interest, instead of following an aggressive competition [22] Y. Zhang, Z. Xiong, D. Niyato, P. Wang, H. V. Poor, and D. I. Kim,
strategy. Empirical results showed that our model outperforms “A game-theoretic analysis for complementary and substitutable IoT
the state-of-the-art cloud business models, i.e., the fifty-fifty services delivery with externalities,” IEEE Transactions on Communi-
cations, vol. 68, no. 1, pp. 615–629, 2020.
and pay-as-you-go models in terms of total surpluses earned [23] A. S. Bataineh, R. Mizouni, M. E. Barachi, and J. Bentahar, “Monetiz-
by the different parties. ing personal data: A two-sided market approach,” Procedia Computer
Science, vol. 83, pp. 472 – 479, 2016.
[24] L. Mashayekhy, M. M. Nejad, and D. Grosu, “A two-sided market
R EFERENCES mechanism for trading big data computing commodities,” 2014 IEEE
[1] “Dmr amazon statistical report 2018,” https://expandedramblings. International Conference on Big Data (Big Data), pp. 153–158, 2014.
com/index.php/downloads/dmr-amazon-web-services-report/, accessed: [25] O. A. Wahab, J. Bentahar, H. Otrok, and A. Mourad, “A stackelberg
2019-01-31. game for distributed formation of business-driven services communities,”
[2] Y. Jiao, P. Wang, S. Feng, and D. Niyato, “Profit maximization mecha- Expert Systems with Applications, vol. 45, pp. 359–372, 2016.
nism and data management for data analytics services,” IEEE Internet [26] Google, “Google clustred data,” https://github.com/google/cluster-data,
of Things Journal, vol. 5, no. 3, pp. 2001–2014, 2018. [Online; accessed 19-July-2019].
[3] D. Niyato, D. T. Hoang, N. C. Luong, P. Wang, D. I. Kim, and Z. Han, [27] Amazon, “Simple monthly calculator,” https://calculator.s3.amazonaws.
“Smart data pricing models for the internet of things: a bundling strategy com/index.html, [Online; accessed 19-July-2019].
approach,” IEEE Network, vol. 30, no. 2, pp. 18–25, 2016. [28] ——, “IoT and big data services in Amazon market places,”
[4] A. Jin, W. Song, P. Wang, D. Niyato, and P. Ju, “Auction mechanisms https://aws.amazon.com/marketplace/search?page=1&category=
toward efficient resource sharing for cloudlets in mobile cloud com- 96c2cd16-fe69-4b18-99cc-e016c61e820c, [Online; accessed 19-
puting,” IEEE Transactions on Services Computing, vol. 9, no. 6, pp. Nov-2019].
895–909, 2016. [29] P. J. Danaher, “Optimal pricing of new subscription services: Analysis
[5] A. S. Bataineh, R. Mizouni, J. Bentahar, and M. El Barachi, “Toward of a market experiment,” Marketing Science, vol. 21, no. 2, pp. 119–138,
monetizing personal data: A two-sided market analysis,” Future Gener- 2002.
ation Computer Systems, vol. 111, pp. 435–459, 2020.
[6] J. Chakareski, “Cost and profit driven cloud-P2P interaction,” Peer-to-
Peer Networking and Applications, vol. 8, no. 2, pp. 244–259, Mar 2015.
[7] Z. Xiong, S. Feng, D. Niyato, P. Wang, and Y. Zhang, “Competition Ahmed Saleh Bataineh received the bachelor’s
and cooperation analysis for data sponsored market: A network effects degree in computer engineering from Jordan Uni-
model,” CoRR, vol. abs/1711.01054, 2017. versity of science and technology, Jordan, in 2010,
[8] W. Güth, R. Schmittberger, and B. Schwarze, “An experimental analysis and the master’s degree in electrical and computer
of ultimatum bargaining,” Journal of Economic Behavior and Organi- engineering from Concordia University, Canada, in
zation, vol. 3, no. 4, pp. 367 – 388, 1982. 2014. He is currently pursuing the Ph.D. degree
[9] J. Rochet and J. Tirole, “Platform competition in two-sided markets,” with the Concordia Institute for Information Systems
Journal of the European Economic Association, vol. 1, no. 4, pp. 990– Engineering (CIISE) under the supervision of Prof.
1029, 2003. Jamal Bentahar. His past research activities include
[10] A. Greenberg, J. Hamilton, D. A. Maltz, and P. Patel, “The cost of a multi-agent systems, verification and model checking. His current research
cloud: Research problems in data center networks,” SIGCOMM Comput. interests include economics, optimization problems and game theory applied
Commun. Rev., vol. 39, no. 1, pp. 68–73, Dec. 2008. to IoT services and big data.
[11] N. Zhang and H. Hämmäinen, “Cost efficiency of SDN in LTE-based
mobile networks: Case Finland,” in 2015 International Conference and
Workshops on Networked Systems (NetSys), 2015, pp. 1–5. Jamal Bentahar received the Ph.D. degree in
[12] S. Rebai, M. Hadji, and D. Zeghlache, “Improving profit through cloud computer science and software engineering from
federation,” in 2015 12th Annual IEEE Consumer Communications and Laval University, Canada, in 2005. He is a Professor
Networking Conference (CCNC), 2015, pp. 732–739. with Concordia Institute for Information Systems
[13] Y. Lin and H. Shen, “Autotune: game-based adaptive bitrate streaming Engineering, Concordia University, Canada. From
in P2P-assisted cloud-based VoD systems,” in 2015 IEEE International 2005 to 2006, he was a Postdoctoral Fellow with
Conference on Peer-to-Peer Computing (P2P), 2015, pp. 1–10. Laval University, and then NSERC Postdoctoral
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 14

Fellow at Simon Fraser University, Canada. He is an

NSERC Co-Chair for Discovery Grant for Computer
Science (2016-2018). His research interests include the areas of computational
logics, model checking, multi-agent systems, service computing, game theory,
and software engineering.

Rabeb Mizouni is an associate professor in Elec-

trical and Computer Engineering at Khalifa Univer-
sity. She got her PhD and her MSc in Electrical and
Computer Engineering from Concordia University,
Montreal, Canada in 2007 and 2002 respectively.
Currently, she is interested in the deployment of
context aware mobile applications, software product
line and cloud computing.

Omar Abdel Wahab is an Assistant Profes-

sor at the Department of Computer Science and
Engineering, Université du Québec en Outaouais,
Canada. He holds a Ph.D. in Information and
Systems Engineering from Concordia University,
Montreal, Canada. He received his M.Sc. in com-
puter science in 2013 from the Lebanese American
University (LAU), Lebanon. From 2017 to 2018,
he was a postdoctoral fellow at the École de Technologie Supérieure (ÉTS),
Canada, where he worked on an industrial research project in collaboration
with Rogers and Ericsson. The main topics of his current research activities
are in the areas of cybersecurity, data analytics, Internet of Things (IoT) and
game theory.

Gaith Rjoub received the B.S. degree in infor-

mation technology engineering from Arab Open
University, Amman, Jordan, in 2008 and M.Eng.
degree in Quality Systems Engineering from Con-
cordia Institute for Information Systems Engineer-
ing (CIISE), Canada in 2014. He is currently
pursuing his Ph.D. degree in Information Systems
Engineering at concordia University. His research interests include cloud
computing, machine learning, artificial intelligence and big data analytics.


May El Barachi is a next generation network-

ing expert holding Ph.D. and master’s degrees in
Computer Engineering from Concordia University
(Canada). She has 12 years’ experience in the
field with a strong focus on smart and resource
efficient systems. During those years, she acquired
academic experience as associate professor and
has worked as researcher with Ericsson Research
Canada. She co-founded and directed a cutting-
edge research lab aiming at investigating sensory, cloud, and mobile technolo-
gies. Presently, she works as Associate Professor at University of Wollongong
Dubai and is involved in several research collaborations with international
universities and local industrial partners.