Quality-Assured Secured Load Sharing in Mobile Cloud ... · quality-assured secured load sharing...

Quality-Assured Secured Load Sharing in MobileCloud Networking Environment

Snigdha Das, Manas Khatua, Student Member, IEEE, Sudip Misra , Senior Member, IEEE, and

Mohammad S. Obaidat, Fellow, IEEE

Abstract—In mobile cloud networks (MCNs), a mobile user is connected with a cloud server through a network gateway, which is

responsible for providing the required quality-of-service (QoS) to the users. If a user increases its service demand, the connecting

gateway may fail to provide the requested QoS due to the overloaded demand, while the other gateways remain underloaded. Due to

the increase in load in one gateway, the sharing of load among all the gateways is one of the prospective solutions for providing

QoS-guaranteed services to the mobile users. Additionally, if a user misbehaves, the situation becomes more challenging. In this

paper, we address the problem of QoS-guaranteed secured service provisioning in MCNs. We design a utility maximization problem for

quality-assured secured load sharing (QuaLShare) in MCN, and determine its optimal solution using auction theory. In QuaLShare, the

overloaded gateway detects the misbehaving gateways, and, then, prevents them from participating in the auction process.

Theoretically, we characterize both the problem and the solution approaches in an MCN environment. Finally, we investigate the

existence of Nash Equilibrium of the proposed scheme. We extend the solution for the case of multiple users, followed by theoretical

analysis. Numerical analysis establishes the correctness of the proposed algorithms.

Index Terms—Mobile cloud networking, auction theory, Nash equilibrium, bandwidth distribution, load sharing

Ç

1 INTRODUCTION

NOW-A-DAYS mobile devices are inseparable componentsof our lives. Rapid reduction of hardware cost and the

availability of numerous user-friendly applications such ashealth monitoring, gaming, entertainment, social network-ing, and trading applicationsmake these devicesmore popu-lar to the users. Despite the increase of usage of mobiledevices, exploitation of their full functionalities is restrictedby the limited energy supply, low bandwidth, low computa-tional capacity, and seamless mobility. For overcoming theselimitations, mobile devices are integrated with cloud com-puting in mobile cloud computing (MCC) systems. Mobilecloud networking (MCN), which one of the foundationalaspects of MCC, is responsible for designing cost-efficientand secure service delivery across data communication net-works. In MCN, a mobile user requests for expensive serv-ices to the cloud server though their respective interfacinggateways. The cloud server processes the service-requests,and, finally, provides the services to the users.

1.1 Motivation

On-demand service delivery is one of the basic characteris-tics of MCN. For providing real-time services over the

wireless networks, dynamic allocation of resources such asbandwidth is essential. It may happen that the cloud serviceprovider (CSP) has adequate bandwidth for providing therequested services to the mobile users, but the users are notable to harness it due to improper bandwidth management.Additionally, the cloud server has to assure a certain levelof QoS in terms of service delay, reliability, jitter, andresponse time for encouraging the users to take servicesfrom it. Let us consider a cloud service provider, which pro-vides different service demands to mobile users using asimple architecture, as shown in Fig. 1. When a mobiledevice changes its location from one place to another, theinterfacing gateway also changes for making it feasible tooffer uninterrupted service with adequate QoS from thecloud. After the change of location of a user, the currentinterfacing gateway may experience additional service load.We assume that the service request of all the users are inde-pendent of one another. Again, the connecting gatewaymay feel excessive service load when one or more mobileusers increase their service demands. For maintaining theadditional service request, proper bandwidth managementis one of the prospective solutions. Recently, this problemwas addressed by Misra et al. [1] by proposing an auction-based bandwidth distribution approach. The authors definethe term “bandwidth distribution” as the proportionalassignment of bandwidth to all the gateways, even if only afew gateways change their demands. Therefore, their pro-posed solution involves all the gateways in the distributionprocess, which we consider as additional overhead to theusers with differential demands. We further observe thatthe overloaded gateway may be surrounded by under-loaded gateways. In this case, a localized solution that onlyinvolves few local gateways can be computationally

� S. Das, M. Khatua, and S. Misra are with the School of Information Tech-nology, Indian Institute of Technology, Kharagpur, India.E-mail: {snigdhadas, manas.khatua, smisra}@sit.iitkgp.ernet.in.

� M.S. Obaidat is with the Department of Computer and InformationScience, Fordham University, NY. E-mail: [email protected].

Manuscript received 12 Nov. 2014; revised 27 June 2015; accepted 10 July2015. Date of publication 16 July 2015; date of current version 6 Mar. 2019.Recommended for acceptance by D. Kliazovich.For information on obtaining reprints of this article, please send e-mail to:[email protected], and reference the Digital Object Identifier below.Digital Object Identifier no. 10.1109/TCC.2015.2457416

102 IEEE TRANSACTIONS ON CLOUD COMPUTING, VOL. 7, NO. 1, JANUARY-MARCH 2019

2168-7161� 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See ht _tp://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on April 04,2020 at 18:59:04 UTC from IEEE Xplore. Restrictions apply.

https://orcid.org/0000-0002-2467-6414

https://orcid.org/0000-0002-2467-6414

https://orcid.org/0000-0002-2467-6414

https://orcid.org/0000-0002-2467-6414

https://orcid.org/0000-0002-2467-6414

mailto:

mailto:

inexpensive with respect to the global solution that involvesall the gateways of the network. These observations moti-vated us for designing a localized solution scheme, namelyquality-assured load sharing. Thus, the mobile users are totallytransparent about this load sharing process. During theprocessing time, the system will not accept any additionalrequest. The system tracks those requests and processes inthe next run. Additionally, optimal load sharing in a local-ized scheme cannot be guaranteed in the presence of a mis-behaving gateway.

Definition 1 (Misbehaving Gateway). A gateway is termed asamisbehaving gatewaywhen it provides wrong information dur-ing the load sharing process for receiving additional privilege.

As a consequence, the other gateways loose in the deci-sion process. Nevertheless, the actual overloaded gatewaymay fail to maintain the required QoS to the mobile usersconnected to it. Therefore, a truthful load sharing scheme isrequired for avoiding such inefficiency of CSP.

1.2 Contributions

In this paper, we address the problem of QoS violation dueto increase in load by a single user or multiple users as theCSP is responsible for providing QoS-guaranteed serviceprovisioning in an MCN environment. Traditionally, loadmanagement is primarily achieved through load balancing.Our works step ahead by proposing a load sharing schemefor load management. At this juncture, it is pertinent to clar-ify that load sharing differs from the traditional concept ofload balancing in that, while the former concerns sharingthe additional demand with local gateways, the latter con-cerns equalizing the service demand among all the gate-ways. An architectural view of the problem is shown inFig. 1. The dotted square in Fig. 1 represents the boundaryof a local region, whereas all the gateways are in the globalregion. We assume that the CSP is authorized for the distri-bution of initial loads when global solution is required, andtracks the current QoS index level for all the connected gate-ways. As in the work by Misra et al. [1], we also considerQoS-guarantees in terms of service delay. In the loadsharing process, our approach intelligently identifies the

misbehaving gateways, and, then, avoids them from involv-ing in the further execution process of the proposed scheme.We present load sharing as a utility maximization problemfor all the gateways, in which the utility is formed using therevenue and payoff functions. The revenue function of a gate-way is defined by two components—transmission delay-specific and service delay-specific revenue functions. Thepayoff function illustrates the pricing strategy between theoverloaded gateway and its neighboring ones. We use anauction theory-based scheme to solve the optimization prob-lem. Each neighbor of the overloaded gateway shares theexpected QoS index, only if the mobile user in considerationis within the vicinity of it. The QoS index is representativeof the interest of the gateway to take part in the auction-based load sharing process. The overloaded gateway meas-ures the estimated QoS index based on its own knowledgeand the information received from the CSP. Finally, theoverloaded gateway verifies the estimated QoS index withthe pre-specified expected value for including only thetruthful gateways in the auction process. We, further, estab-lish the existence of Nash Equilibrium (NE) in the proposedscheme. We theoretically analyze the upper and lower pric-ing limits, and prove that the algorithm reaches optimality.We extend the solution for multiple users’ load incrementin the presence of multiple misbehaving gateways. We sum-marize the main contributions of this paper as follows:

� We, theoretically, prove the requirement of loadsharing in case of increase in demand by a user inMCN environment.

� We design a utility maximization problem for quality-assured secured load sharing (QuaLShare) in MCNwith overloaded demand of a gateway.

� We analyze the optimality criteria and the existenceofNashEquilibrium for theproposed scheme,QuaLShare.

� We extend the solution for the multiple users’demand increment problem, followed by its theoreti-cal analysis, for checking the optimality and NashEquilibrium.

1.3 Paper Organization

The rest of the paper is organized as follows. In Section 2, webriefly present few relevant related works reported in theexisting literature. In Section 3, we theoretically prove thenecessity of the load sharing process along with the networkmodel description. We present the utility maximizationproblem formulation followed by an auction-based solutionapproach for load sharing with single user in Section 4. InSection 5, we analyze the problem for the case of multipleusers. We present the experimental results in Section 6.Finally, we conclude the paper in Section 7 with discussionsabout how this work can be extended in the future.

2 RELATED WORK

Although in MCC mobile devices integrate with cloud forproviding flexible, affordable, and capable service provi-sioning to the mobile users, the nascent integration facesmajor challenges due to finite power supply, low connectiv-ity, presence of misbehaving users, and inefficient protocoldesign with respect to energy consumption, QoS provision-ing, and resource allocation. Dinh et al. [7] and Fernando

Fig. 1. Basic architecture of mobile cloud networks.

DAS ET AL.: QUALITY-ASSURED SECURED LOAD SHARING IN MOBILE CLOUD NETWORKING ENVIRONMENT 103


et al. [8] discussed about the subtleties behind enablingMCC. They highlighted several open issues such as secu-rity, resource management, network management, andquality of service provisioning. Sanaei et al. [9] describedthe present challenges and opportunities introduced by het-erogeneity in MCC. Abolfazli et al. [10] indicated the exist-ing issues on quality assurance in cloud based mobiledevices. An energy-time efficient heterogeneous resourceframework for hybrid mobile cloud computing has beenstudied by Sanaei et al. in [11]. A few recent works such as[12], [13], [14], and [15] emphasized on the existing securityissues in cloud and mobile cloud environment. Such baremigration builds the motivational platform of our presentwork on secure load sharing in MCC.

Several works exist on load sharing and load balancingin other networks such as mobile computer network,homogeneous and heterogeneous system, and cognitiveradio network. Othman and Hailes [16] discussed a powerconservation strategy for mobile computers using loadsharing. They reduced CPU power consumption by mov-ing the jobs from a mobile host to a fixed host. Eager et al.[17] discussed a simple but adaptive load sharing policyfor redistributing workload among users in homogeneoussystem. Another adaptive load sharing scheme for hetero-geneous system was proposed by Mirchandaney et al.[18]. A game-theoretic framework was designed by Grosuand Chronopoulos [6] for obtaining a distributed, lowcomplexity, and optimal load balancing scheme. Wanget al. [19] studied a resource allocation scheme with loadbalancing for maximizing the total data rate of all users incognitive radio networks (CRNs). Wang et al. [20] pro-posed a novel auction-based approach for spectrum shar-ing in CRNs, in which the primary user (PU) allocatesspectrum to the secondary users (SUs) based on thedemand of the SUs, without violating its own perfor-mance. A stochastic differential game based dynamicbandwidth allocation problem was investigated by Zhuet al. [21] for both the SUs and the service providers (SPs).Guaranteed bandwidth allocation, which includes QoSrequirement of the network components, was also pro-posed by Elias et al. [22]. Kun et al. [23] proposed a band-width allocation scheme for allocating equal amount ofbandwidth to all users with the same type of service

request. Zhang et al. [24] briefly summarized the differentauction approaches for resource allocation in wireless sys-tems. Although the existing works are similar to our pres-ent work, the former works do not consider mobility aswell as security issue in their problem.

There exists very few works, specifically on load balanc-ing in cloud networks. Nuaimi et al. [25] highlighted theexisting challenges of load balancing and task scheduling incloud computing. They compared the existing algorithmswith respect to overall network load and time series.Randles et al. [4] studied three distributed approaches,namely Honeybee Foraging Behavior, Biased Random Sam-pling, andActive Clustering, for comparing different distrib-uted load balancing schemes. Doyle et al. [2] proposed ascheme for scheduling incoming traffic to the nearest serverto reach energy efficiency. They used Voronoi partitions forselecting low carbon emitted data center. A stochastic modelof a cloud computing cluster was investigated by Maguluriet al. [5]. Zaman and Grosu [3] discussed an auction-basedmechanism for dynamically allocating the virtual machines(VM). Papagianni et al. [26] proposed a unified resource allo-cation framework for cloud networks. Amamou et al. [27]discussed a service level agreement-aware dynamic band-width allocator (DBA), in which bandwidth is allocatedamong the VMs based on the application requirements. Themapping of cloud server andmobile users inMCCwas stud-ied by Das et al. [28]. A varying QoS requirement problemfor different data incentive applications in cloud computingsystems was studied by Lin et al. [29]. For distributing het-erogeneous resources in mobile cloud systems, Yang et al.[30] proposed a combinatorial auction scheme. Wang et al.[31] introduced a double multi-attribute auction basedresource allocation mechanism in cloud environment. How-ever, the fundamental differing characteristic of the work pre-sented in this paper from the existing ones is that weconsider resource sharing in the presence of misbehavingusers. Recently, Misra et al. [1] studied the bandwidth shift-ing and redistribution problems in a typical MCC environ-ment. They studied proportional distribution of the availablebandwidth among all the gateways in MCC. Therefore, thetotal cost for distribution is more as all the network gatewaysparticipate in the distribution process. A comparative studyof the current state of the work is shown in Table 1.

TABLE 1Different Resource Allocation Schemes and Respective Solutions with Parameter Metrics

Different schemes Resource Type Delay Metric Misbehavior Mobility DistributedAlgorithm

Local/GlobalApproach

AQUM: Mirsa et al. [1] Bandwidth @ � @ � GlobalStratus: Doyle et al. [2] Data Center @ � � @ GlobalZaman and Grosu [3] Virtual Machine � @a � @ Global

Honeybee Server � � � @ Global

Randles et al. [4]Biased RandomSampling

Job @ � � @ Global

Active Clustering System � � � @ Global

Maguluri et al. [5] CPU, Memory,Storage

@ � � @ Global

Grosu and Chronopoulos [6] Computer @ @b � @ GlobalQuaLShare (Our scheme) Bandwidth @ @ @ @ Local

aUser’s misbehavior is considered in terms of truthfulness.bUser’s selfishness is considered, whereas we consider the gateway’s misbehaving characteristics.



Many security related works such as [15], [32], [33], [34],[35], and [13] also exist in a cloud computing environment.An anti-cheating bidding scheme was proposed by Hu et al.[12] for investigating the bidding strategy of bidders andfinally securing the system from malicious bidders. Chenet al. [33] introduced three auction-based mechanisms forcheat-proof distributive spectrum allocation in cognitiveradio based systems. However, none of these schemesaddresses the problem of providing QoS-guaranteed loadsharing in the presence of misbehaving users in MCN. Inthis paper, we propose a truthful auction-based QoSguaranteed load sharing scheme in MCN.

3 MOBILE CLOUD NETWORK MODEL

Consider a simple mobile cloud environment with one CSPand I single channel gateways G ¼ fG1; G2; . . . ; GIg con-nected with the CSP through wireless channel. We furtherconsider that each gateway Gi has Ki number of mobileusers connected with it through a mobile network. Hence,

Ni ¼ fNi1; Ni2; . . . ; NiKig, and, thus, N ¼ S I

i¼1Ni. At this

juncture, we should mention that, throughout the paper, thebold-faced characters represent the set of corresponding ele-ments. Let us consider that the total allocated bandwidthvector for the gateways G is B ¼ fB1; B2; . . . ; BIg. Btot

denotes the total available bandwidth of the CSP. If a mobileuser requests any service from the CSP, the service is pro-vided through the connecting gateway. We compute theservice delay di of gateway Gi as follows [1]:

di ¼ Btot

PjNKij

k¼1 Tik

Eið1� aiÞBi; (1)

where Tik, aa ¼ fa1;a2; . . . ;aIg, and E ¼ fE1; E2; . . . ; EIgrepresent the ideal transmission delay required for access-ing a service by the mobile node Nik, if the entire Btot is allo-cated to gateway Gi, the protocol overhead and the spectralefficiency of each wireless channel, respectively. In Table 2,we list few notations used in this paper.

3.1 Necessity of Load Sharing

In MCN, the load of a gateway Gi increases when either aconnected user increases its demand or a mobile userchanges its connecting gateway from Gj to Gi due to loca-tion change. Therefore, the service delay of the gateway Gi

is affected. For maintaining the QoS in terms of servicedelay, the gatewayGi calculates the total transmission delayfor all the connected mobile users. In Theorem 1, we provethat the service delay differs with the heterogeneous servicedemand.

Theorem 1. Let us consider that a mobile user Nil modifies itsservice demand. The service delay of the connecting gatewayGi before the demand change is unequal to that after thedemand change, because of the variation in total transmissiondelay of Gi.

Proof. We consider the case in which only one mobile user,say Nil, of the gateway Gi changes its service demand.Therefore, the transmission delay for the mobile user

varies from Til to bTil. We represent the service delay com-putation before and after the service demand change by

diðt1Þ and diðt2Þ, respectively. Using Equation (1), wecompute diðt1Þ as follows:

diðt1Þ ¼Btot½

PjNijk¼1;k 6¼l Tik þ Til�

Eið1� aiÞBi: (2)

Similarly, we compute diðt2Þ as follows:

diðt2Þ ¼Btot½

PjNijk¼1;k6¼l Tik þ bTil�

Eið1� aiÞBi: (3)

As the values of Til and bTil are different, it can beinferred that,

diðt1Þ 6¼ diðt2Þ: (4)

This concludes the proof. tuDepending upon the transmission delay variation, the

following scenarios occur:

TABLE 2Summary of Notations

Notation Description

Gi Gateway iI Total number of gatewaysNiki Set of users connected to a gateway Gi

Bi Allocated bandwidth to a gateway Gi

Btot Total available bandwidth in CSPdi Service delay associated with a gateway Gi

Tik Transmission delay required for accessing a serviceby a mobile user Nik before load incrementbTikTransmission delay required for accessing a serviceby a mobile user Nik after load increment

Ei Spectral efficiency of a channel associated with Gi

ai Protocol overhead corresponding to a gateway Gi

Ui Utility of gateway Gi

Uiniti

Utility of gateway Gi before load incrementri Revenue per unit service delay received by Gi

qi Revenue per unit transmission delay received by Gi

#i Payoff accumulated at Gi due to load sharingdui Threshold value of service delay corresponding to Gi

dminu

Minimum delay thresholddmaxu Maximum delay threshold

dei Expected service delay associated with a gateway Gi�di Expected QoS permissible index of gateway Gi

�dei Estimated QoS permissible index of gateway Gi

�dci Current QoS permissible index of gateway Gi

dik Service delay occurred for user Nik

Dp Change in bid for single user load incrementH Number of iterationp pricing value for single user load incrementp upper pricing limit for single user load incrementp lower pricing limit for single user load incrementDP Change in bid for multiple users load incrementP pricing value for multiple users load incrementP upper pricing limit for multiple users load incrementP lower pricing limit for multiple users load incrementn rate parameterg Fractional constantGj� Winner Gateway for S-QuaLShareG� Winner Gateway set for M-QuaLShareLil Position of the userNil



Case I: We have diðt1Þ � diðt2Þ, if Til � bTil. It means thatQoS is not violated in terms of service delay.

Case II: We have diðt1Þ < diðt2Þ, if Til < bTil. In this case,the decision of QoS is taken by the value of service delaythreshold dui of the gateway Gi as follows:

Case IIa: The QoS provisioning by the gateway Gi is notviolated if diðt2Þ � dui .

Case IIb: The QoS provisioning by the gateway Gi is vio-lated if diðt2Þ > dui .

From Theorem 1, we observe that the QoS provisioningby a gateway fails when the modified service delayexceeds the service delay threshold. This necessitates therequirement of ensuring load sharing. Although the band-width distribution approach proposed by Misra et al. [1] isone of the useful remedies for the above problem, itinvolves all the gateways in decision making, and is thus,computationally expensive, as explained in Section 1.Unlike bandwidth distribution, the load sharing approachresults in low overhead, as the solution approach does notinvolve all the gateways until and unless it is required.

4 QUALITY-ASSURED SECURED LOAD SHARING

FOR SINGLE USER LOAD INCREMENT

In this section, we consider the case of single user loadincrement, whereas the consideration of multiple userload increment is discussed in Section 5. We formulatethe utility function of the gateway for designing an opti-mal solution of the load sharing problem under singleuser load increment. Thereafter, we present an auctiontheory-based load sharing scheme, namely Quality-assuredSecured Load Sharing (QuaLShare), followed by the theoret-ical analysis of the QuaLShare approach. Henceforth,throughout the paper, we use the prefixes ‘S’ and ‘M’along with the name QuaLShare for differentiating theproposed solution under the single user and multipleusers load increment, respectively.

4.1 Utility Function

We design a utility function for computing the overall pay-off of a gateway. The utility function depends on both thetransmission and service delays. The transmission delayspecific revenue of a gateway is calculated for measuringthe amount of service which is provided to the mobile users.The service delay specific revenue is computed for measur-ing the quality of service which is provided to the connect-ing mobile users. In addition to that, the utility functionconsiders the payoff a gateway gains or loses due to loadsharing. Considering the above three factors, the utilityfunction of a gateway can be expressed as follows:

U ¼ RT ðTÞ þ RQðdÞ þ ##; (5)

where d ¼ fd1; d2; . . . ; dIg denotes the service delay vectorand T ¼ fT1; T2; . . . ; TIg represents the transmission delay

vector of the interfacing gateway. RT ðTÞ and RQðdÞ definethe transmission delay-specific revenue function and theservice delay-specific revenue functions, respectively, and #denotes the amount of payoff a gateway receives from theload sharing solution. We model the three utility compo-

nents RT ðTÞ, RQðdÞ, and ## as follows.

1) Transmission delay specific revenue function modeling:The increase in transmission delay increases the rev-enue of a gateway. So, we formulate the transmissiondelay specific revenue function as follows:

RT ðTÞ ¼ qT; (6)

where the vector q ¼ fq1; q2; . . . ; qIg denotes the rev-enue per unit transmission delay.

2) Service delay specific revenue function modeling: The ser-vice delay of a gateway is inversely proportional tothe revenue of the gateway. Hence, we formulate theservice delay-specific revenue function as follows:

RQðdÞ ¼ ðdu � dÞr; (7)

where the vector r ¼ fr1; r2; . . . ; rIg defines the reve-nue per unit service delay.

3) Load Sharing specific payoff modeling: The vector## ¼ f#1; #2; . . . ; #Ig defines the payoff which is accu-mulated during the load sharing process. If #i equalszero, the connecting gateway only serves its own ser-vice request demand. The positive #i value indicatesthat the gateway is overloaded, and it presentlyshares excess load with its neighboring gateways. Onthe contrary, the negative #i denotes that the gatewayserves the demands of its neighboring gateways.

Hence, the utility of a gateway Gi is defined as:

Ui ¼ qiXjNiki

j

k¼1;k6¼l

Tik þ bTil

24

35þ riðdu � diÞ � #i: (8)

Finally, the utility maximization problem is expressed asfollows:

maximize Ui

subject to di � dui ; i 2 ð1; IÞ: (9)

4.2 S-QuaLShare Algorithm

Auction theory is a well-known procedure for purchasingand selling resources by offering prices according tobidding value. Therefore, we use an auction theory-basedapproach for designing the S-QuaLShare algorithm in MCNenvironment. In the S-QuaLShare algorithm, the overloadedgateway participates as a seller cum auctioneer, and theneighbors of the overloaded gateway act as buyers. We usethe auction process for selecting such a truthful buyer, whomaximizes the utility of the overloaded gateway amongall the neighbors. We describe the basic steps of the S-QuaL-Share algorithm as follows.

1) Broadcast Pre-requisites: We consider that, each gate-way Gi knows the location information of all theusers who are currently connected with it. The gate-way Gi computes the transmission delay accordingto a user’s demand. Initially, the overloaded gatewayGi broadcasts the following information of user Nil –

transmission delay bTil and location Lil.2) Expected value computation of QoS permissible index:

Each neighboring gateway Gj; j 2 NeðGiÞ computes



the expected delay dej followed by the expected QoS

permissible index �dj based on the transmission andservice delays of all mobile users connected with it.

The expressions of dej and�dj are shown as follows:

dej ¼Btot½

PjNKjj

k¼1 Tjk þ bTil�Ejð1� ajÞBj

(10)

�dj ¼dejduj

: (11)

Finally, the gateway Gj shares �dj to the overloadedgateway Gi if the user Nil is within the proximityregion of it.

3) Runtime estimation of QoS permissible index: The over-loaded gateway Gi estimates the QoS permissible

index �dej for a participating gateway Gj, depending

on the present QoS permissible index �dcj, minimum

delay threshold dminu , and the service delay dil that

occurred forNil. So,

�dej ¼ �dcj þdildminu

: (12)

In this computation, the gateway Gi receives �dcj and

dminu related to the overloaded gateway from the CSP.

4) Misbehavior detection: The overloaded gateway Gi

compares the expected and estimated QoS permissi-ble index for identifying misbehaving gateways. Thegateway Gi accepts the interested gateways, as par-ticipants in the auction process, whose value of theabove difference lies within �. Remaining gatewaysare excluded from the process.

5) Determine bid value: The overloaded gateway submitsa bid value p, which represents the minimum pricefor sharing the load at the initial stage. In every itera-tion, the bid balue is incremented by some positivequantity, Dp, as follows:

pðH þ 1Þ ¼ pðHÞ þ Dp: (13)

Gi increments its bid value until its current value ofutility exceeds the initial value.

6) Bid selection: The participating gateways accept thegiven bid value if and only if the expected value offuture payoff is less than the current payoff value.

7) Load allocation: Gi allocates the user Nil to the gate-way G�

j who has accepted the bid. If multiple gate-

ways accept the bid in the same iteration cycle, Gi

assigns the user to the gateway G�j with minimum

�dej as lesser QoS index denotes better service. Oth-

erwise, we conclude that load sharing is not suffi-cient for providing services with adequate QoSguarantee. Henceforth, the CSP invokes the band-width distribution algorithm [1] for redistributingthe bandwidth such that the modified distributionneutralizes the effects of QoS violation due to loadincrement.

8) Payoff calculation: The winning gateway Gj� gets

the benefit value (pðHÞ �Hn bTil) from the over-loaded gateway Gi. Each loser gateway Gj6¼j� getsno benefit. Therefore, #i and #�

j are represented

as follows:

#i ¼ pðHÞ �Hn bTil (14)

#�j ¼ �pðHÞ þHn bTil: (15)

The pseudo-code of the S-QuaLShare scheme ispresented in Algorithm 1.

Algorithm 1. S-QuaLShare Algorithm

Inputs:Nilð bTil; LilÞOutput: Gj�1: Invoke Misbehavior Detection algorithm for single user

share as shown in Algorithm 22: if UiðH þ 1Þ � Uinit

i < 0 then3: if user not assign then4: Bandwidth Distribution required.5: end if6: else7: if ðmaxð�dj; �dejÞ � 1Þ then8: Gi shares the bid pwith the participant9: if any Gj� accepts the price then10: Assign the user to Gj�11: else12: ifmultiple gateways accept the price then13: Assign the user to Gj� with minimum �dej14: Gi pays #i to the assignee gateway15: else16: Revise the bid price pðH þ 1Þ ¼ pðHÞ þ Dp17: Goto Step 218: end if19: end if20: end if21: end if

Algorithm 2.Misbehavior Detection for Single User Share

Inputs:Nilð bTil; LilÞOutput: Gj�1: Gi broadcasts Nilð bTil; LilÞ2: Each gateway Gj; j 2 NeðGiÞ calculates �dj3: Gj; j 2 NeðGiÞ shares �dj if within proximity4: Gateway Gi calculates �dej5: if j�dej � �djj > � then6: Exclude misbehaving gateways from auction7: end if

4.3 Pricing Limit

In this section, we determine the upper and lower pricinglimits in the S-QuaLShare algorithm, as shown in Theo-rem 2. We further conclude from the theorem that anygateway does not accept the bid value in the auction pro-cess if the price is less than the lower pricing limit p. Sim-ilarly, if the bid price is more than the upper pricing limitp, the overloaded gateway does not execute the sharingprocess.



Theorem 2. The upper and lower pricing limits are �pi ¼ ðqi þHnÞbTil � qiTil and pi ¼ ½maxð�dj � �dcjÞ�dmaxu ri, respectively.

Proof. The overloaded gateway Gi shares the load of amobile user only if it does not lose from sharing. It implies

that Ui � Uiniti � 0, where Ui ¼ qi

PjNikij

k¼1;k6¼l Tik þ qi bTil þriðdui � diÞ � pi þHn bTil and Uinit

i ¼ qiPjNiki

jk¼1;k6¼l Tik þ qiTilþ

riðdui � diÞ. So,

Ui � Uiniti � 0 ) ðqi þHnÞ bTil � qiTil � p: (16)

From this inequality, we determine the upper bound ofthe pricing limit as follows:

�pi ¼ ðqi þHnÞ bTil � qiTil: (17)

Similarly, a gateway accepts allocation through theauction process if its utility is not negatively effected bythe new sharing. The service delay of the gateway isincreased due to the user sharing process. For includingall the gateways in the auction process, we considerthe maximum service delay increment equal to

½maxð�dj � �dcjÞ�dmaxu , where �dcj and dmax

u denote the current

QoS permissible index of the gateway Gj and the maxi-mum delay threshold among the gateways, respectively.Now, we compute the lower price limit as follows:

pi¼ ½maxð�dj � �dcjÞ�dmax

u ri: (18)

This concludes the proof. tu

4.4 Nash Equilibrium

We investigate the existence of Nash Equilibrium of theS-QuaLShare algorithm, which is stated by the followingtheorem using the rate parameter n.

Theorem 3. There exists Nash Equilibrium in the S-QuaLSharealgorithm if and only if n bTilð�pi � p

iÞn�1 ¼ ðDpÞn.

Proof. A neighboring gateway accepts the bid value onlyif the final payoff is higher than that in the previousiteration of the auction process. Let us assume thatthe bidding price is increased uniformly by Dpamount in each iteration H. Using the first andsecond derivatives of the final payoff function #i withrespect to H, we determine the maximization condi-tion, as follows:

Hn�1 ¼ Dp

n bTil

: (19)

If the maximum payoff value exists within the pric-ing range (shown in Theorem 2), the utility of all thegateways increases. Otherwise, the final utility of theoverloaded gateway or the sharing gateway reduces.Therefore, we conclude that there exists Nash Equilib-rium at the maximum payoff point, which dependson the rate parameter n. We, finally, measure thevalue of n using heuristic analysis. In other words, forthe existence of Nash Equilibrium, the upper pricinglimit pi must be greater than or equal to the final

bidding value (piþ DpH). Hence, we obtain the neces-

sary condition for the existence of Nash Equilibrium,as follows:

n bTilð�pi � piÞn�1 ¼ ðDpÞn: (20)


5 QUALITY-ASSURED SECURED LOAD SHARING

FOR MULTIPLE USERS LOAD INCREMENT

In this section, we extend the S-QuaLShare solution for mul-tiple users load increment. We modify the utility function,and propose an auction theory-based load sharing scheme,namely Quality-assured Secured Load Sharing for MultipleUsers Load Increment (M-QuaLShare).

5.1 Utility Function

We modify the utility function for computing total payoff ofall the gateways due to multiple users’ demand increment.Similar to the case of single user demand increment, theutility of the overloaded gateway, say Gi, is computeddepending on the transmission delay, service delay, andpayoff value, as described in Equation (8). The transmissionand service delay-specific revenue functions are shown inEquations (6) and (7), respectively. Finally, we construct autility maximization problem using the utility functions ofthe overloaded and the underloaded gateways. The utilitymaximization problem for all the gateways is expressedas follows:

maximize U

subject to d � du:(21)

5.2 M-QuaLShare Algorithm Design

We adopt the merely identical auction scheme for sharingthe load of multiple users among the neighboring gateways.The overloaded gateway Gi acts as an auctioneer cum bid-der. The neighbors of Gi are treated as the consumer, andthe loads are represented as the selling objects. The auctionprocess picks up the reliable consumers among all theneighbors of Gi. The elemental steps of the M-QuaLSharealgorithm are illustrated as follows:

1) Broadcast pre-requisite: We assume that each gatewayGi is aware of its own proximity region as wellas the positions of all the connected mobile users.During the time of connectivity establishment, eachgateway measures the transmission delay accordingto the user’s demand. The overloaded gateway Gi

starts the auction process by broadcasting the follow-

ing details—N ¼ fNil1ð bTil1 ; Lil1Þ; Nil2ð bTil2 ; Lil2Þ; . . . ;Nilxð bTilx ; LilxÞg.

2) Expected value computation of QoS permissible index:Each neighbor Gj; j 2 NeðGiÞ measures the expecteddelay vector de

j ¼ fdejl1 ; dejl2 ; . . . ; dejlxg and the expected

QoS permissible index vector �dj ¼ f�djl1 ; �djl2 ; . . . ; �djlxgfor all the users using Equations (10) and (11), respec-tively. If any user is not in the proximity region of



Gj; j 2 NeðGiÞ, the gateway Gj replaces the expectedQoS permissible index for that user by xð> 1Þ.Finally, each Gj shares �dj with the gateway Gi, if atleast one user is within the proximity region of it.

3) Runtime estimation of QOS permissible index: The gate-way Gi estimates the QoS permissible index vectorfor all the participated gateway-user pair usingEquation (12). Hence, we have,

�dej ¼

�de1l1�de1l2 �de1lx

..

. ... . .

. ...

�deIl1�deIl2 �deIlx

2664

3775: (22)

If the expected QoS permissible index of any gate-way-user pair is greater than unity, Gi updates theestimate of QoS permissible index for that pair byxð> 1Þ value.

4) Misbehavior detection: The overloaded gateway Gi

compares the estimated QoS permissible indexwith the expected QoS permissible index for eachinterested gateway-user pair. Finally, the gatewayGi selects the neighboring gateways as partici-pants, who have the value of QoS index differenceless than �.

5) Determine bid value: At the starting phase, the over-loaded gateway Gi submits the bid value P, whichdenotes the minimum price per unit transmissiontime for sharing the load. During the auction pro-cess, it increments the bid value by a positive quan-tity DP, as follows:

PðH þ 1Þ ¼ PðHÞ þ DP: (23)

The overloaded gateway Gi attracts the participantsby increasing the bid value until the current utilityreaches the initial utility value.

6) Bid selection: Any gateway Gj accepts the submittedbid value if the expected total payoff in the next iter-ation is lesser than the present total payoff.

7) Load allocation: The gateway Gi, initially, sorts theusers, based on the required transmission time.From the list of gateways that accepts the submitted

bid, a gateway G�j with minimum value of �dejlk for the

maximum transmission time bTilk in the list Ti is

selected for temporarily reserving the user. Themotivation behind such type of selection is to over-come the scenario where the total available resourcesof the neighbor gateways are sufficient to providethe services, but the individual ones are not. Thetemporary allocation process for the gateway G�

j is

continued until the users’ transmission time listbecomes empty. Likewise, the gateway Gi furthercontinues the temporary reservation process until ittemporarily reserve all the users, or the list of partici-pating gateways becomes empty. Finally, the gate-way Gi permanently assigns the users to therespective winner gateways G� only if all the usersare temporarily reserved. Otherwise, the process ter-minates unsuccessfully stating that the load sharing

process is not sufficient to share the load among theneighbors due to the insufficiency of the resources.

8) Payoff calculation: The overloaded gateway pays theamount ðPðHÞ � gHnÞP bTilk to the winning gate-

ways G�, whereP bTilk represents the total transmis-

sion time corresponding to each winning gateway,and g is a constant value (0 < g < 1). The failedgateways do not get any benefit from the auctionprocess. Hence, we determine #i and #�

j as follows:

#i ¼ ðPðHÞ � gHnÞbTi (24)

#�j ¼ �ðPðHÞ � gHnÞ

X bTilk : (25)

The pseudo-code of the M-QuaLShare scheme ispresented in Algorithm 3.

Algorithm 3.M-QuaLShare Algorithm

Inputs: N ¼ fNil1ð bTil1 ; Lil1Þ; . . . ; Nilxð bTilx ; LilxÞgOutput: G�

1: Invoke Misbehavior detection algorithm for multiple usersshare, as shown in Algorithm 4

2: if UiðH þ 1Þ � Uiniti < 0 then

3: if all users not assigned then4: Bandwidth Distribution required.5: else6: if all users temporarily assigned then7: Assign all users and pay accordingly8: end if9: end if10: else11: Gi shares the bid pwith the participants12: ifmultiple gateways accept the price then13: Sort Ti of the user in descending order14: Call Temporary Allocation (Algorithm 5) for the

gateway Gj with minimum �dejlk15: Repeat from Step 14 until all users temporarily assign

or empty participants list16: else17: if any one Gj accepts the price then18: Sort Ti of the user in descending order19: Call Temporary Allocation (Algorithm 5)20: else21: Revise the bid price PðH þ 1Þ ¼ PðHÞ þ DP22: Goto Step 223: end if24: end if25: end if

Algorithm 4. Misbehavior Detection for Multiple UsersShare


1: Gi broadcastsN2: Each participant Gj; j 2 NeðGiÞ calculates �dj

3: Gj; j 2 NeðGiÞ shares �dj if any Nilk is within range4: Gateway Gi calculates �d

ej

5: if j�dejlk � �djlk j > � then6: Exclude misbehaving gateway from auction process7: end if



Algorithm 5. Temporary Allocation


1: if ðmaxð�djlk ; �dejlkÞ � 1Þ then2: Temporarily reserve the user for G�

j

3: Reserve price #�j

4: Exclude bTilk from Ti

5: end if6: Move to next transmission delay value in list Ti

7: Repeat from Step 1 until the list Ti is moved to the end

5.3 Pricing Limit

In this section, we determine the upper and lowerpricing limits in the M-QuaLShare algorithm, as shown inTheorem 4. We further conclude from the theorem that anygateway does not accept the multiple load share auctionprocess, if the price is less than P. On the other hand, if the

price is more than P, the overloaded gateway does not orga-nize the multiple load sharing process.

Theorem 4. The upper and lower pricing limit per unit transmis-sion are represented as follows:

�Pi ¼ ðqi þ gHnÞ � qi

Pxk¼1 TilkPxk¼1

bTilk

Pi ¼rid

maxuPx

k¼1bTilk

Xxk¼1

½maxð�djlk � �dcjÞ�:

Proof. The overloaded gateway Gi shares the load of mobileusers if and only if the gateway gets benefit from sharing.

This implies that Ui � Uiniti � 0, where Ui ¼ qi

Ph 6¼l Tih þ

qiPx

k¼1bTilk þ riðdui � diÞ � ðP � gHnÞPx

k¼1bTilk and

Uiniti ¼ qi

Ph 6¼l Tih þ qi

Pxk¼1 Tilk þ riðdui � diÞ. Hence,

we have,

Ui � Uiniti � 0

) 1Pxk¼1

bTilk

�qiXxk¼1

bTilk þ gHnXxk¼1

bTilk � qiXxk¼1

Tilk

�� P:

From this inequality, we determine the upper bound ofthe pricing limit as follows:

Pi ¼ ðqi þ gHnÞ � qi

Pxk¼1 TilkPxk¼1

bTilk

: (26)

Likewise, a gateway takes part in the auction process

if its utility is increased by new sharing. The gateway

serves for the additional users due to load sharing.

Therefore, the service delay of the gateway is increased.

For incorporating all the gateways in the auction process,

we consider that the average maximum service delay

increment equalsdmaxuPx

k¼1bTilk

Pxk¼1½maxð�djlk � �dcjÞ�, where �dcj,

dmaxu , and

Pxk¼1

bTilk denote the current QoS permissible

index of the gateway Gj, the maximum delay threshold

among the gateways, and total transmission time of the

load improved users, respectively. Now, we computes

the lower price limit as follows:

Pi ¼rid

maxuPx

k¼1bTilk

Xxk¼1

½maxð�djlk � �dcjÞ�: (27)


5.4 Nash Equilibrium

We investigate the existence of Nash Equilibrium of theM-QuaLShare algorithm, which is stated by the followingtheorem using the rate parameter n.

Theorem 5. There exists Nash Equilibrium in the M-QuaLSharealgorithm, if and only if ngðPi � PiÞn�1 ¼ ðDPÞn.

Proof. The neighboring gateways participate in the loadsharing process only if the final payoff increases withrespect to that in the previous iteration cycle. We con-sider that the bidding price is uniformly increased by Dpamount in each iteration H. Using the first and secondderivatives of the final payoff function #i with respect toH, we determine the maximization condition as follows:

Hn�1 ¼ DPng

: (28)

If the maximum payoff value exists within the pricingrange (shown in Theorem 4), the utilities of all thegateways are increased. Otherwise, the final utility ofthe overloaded gateway (or the sharing gateways) isdecreased. Therefore, we conclude that there exists NashEquilibrium at the maximum payoff point, whichdepends on the rate parameter n. Finally, we derive thevalue of n using heuristic analysis. In other words, the

upper pricing limit Pi must be greater than or equal tothe final bidding value Pi þ DPH for the existence ofNash Equilibrium. Hence, we obtain the necessary condi-tion for the existence of Nash Equilibrium as follows:

ngðPi � PiÞn�1 ¼ ðDPÞn: (29)


6 NUMERICAL RESULTS

In this section, we illustrate numerical results of the pro-posed quality-assured secured load sharing algorithms inan MCN environment. At the outset, we describe an mobilecloud network scenario followed by the parameter settings.We prove the necessity of load sharing followed by band-width allocation to the gateways. We also show the opt-imality and Nash Equilibrium of both the load sharingalgorithms for a single and multiple users’ load incrementin the presence of a misbehaving gateway. We executedeach experiment 30 times, and the ensemble average valuesare plotted with 95 percent confidence interval. We use theAQUM algorithm proposed by Misra et al. [1] as a bench-mark for comparing the quality of service and bandwidthallocation of the gateways in the load sharing scenario.



6.1 Parameter Settings

Let us consider an MCN environment, as shown in Fig. 1,with one CSP and 10 gateways (G1; G2; . . . ; G10). In MCN,the CSP dynamically allocates bandwidth to the gatewaysaccording to their demands for providing resource expen-sive services to the mobile users through the gateways.Each gateway Gi has four connected mobile users(Ni1; Ni2; Ni3; Ni4). We consider the revenue per unit trans-mission qi ¼ 30; 8i 2 I, revenue per unit service delayri ¼ 1; 8i 2 I, and total available bandwidth Btot ¼ 100Mbps. Throughout the experiments, we initially considerthat the ideal transmission time for each mobile user equals

0:1 sec and the fixed SNR equals 10 dB. BER = f10�5; 10�4;

5� 10�5; 10�5; 5� 10�5; 10�5; 10�4; 5� 10�5; 10�5; 5� 10�5gand aa ¼ f0:02; 0:008; 0:008; 0:008; 0:02; 0:02; 0:008; 0:008;0:008; 0:02g represent the target BER and protocol overheadof all the gateways. The Ei of all the gateways are computedbased on the values of the target BER and the SNR using theEquation (3) in [1].

In all the experiments, we consider that the gateway G1

suffers from excess service load. Therefore, the gateway G1

executes the load sharing process with the existing fourneighboring gateways G2; G3; G4 and G5. The delay thresh-olds of all the gateways are {4; 5; 5; 5; 5} sec. We furtherassume that, in the initial stage, the bandwidth is equallydistributed among the gateways. Each gateway is allocated9:5Mbps bandwidth. Initially, the payoff of all the gatewaysis zero, as no one shares other’s load.

The notations LS, SDth, ULI, and SD, adopted in thelegends of the figures, represent the cases of load sharing,service delay threshold, user load increment, and servicedelay, respectively.

6.2 Single User Load Increment

In all the experiments of single user load increment, weconsider that the mobile user N11 connected with thegateway G1 increases its demand. Consequently, theideal transmission time for the modified demandincreases from 0:1 to 0:2 sec. We further consider thatthe mobile user N11 is also within the vicinity of all theneighbors of G1, so that all the neighbors participate inthe auction process.

6.2.1 Single User Load Sharing

Fig. 2 establishes the necessity of load sharing due to singleuser load increment. Whenever the user N11 increases itsdemand, the ideal transmission time for providing a serviceincreases. Consequently, the service delay of the connectinggateway G1 increases, which ultimately crosses the thresh-old value. The service delay of the winner gateway G2

remains undisturbed after the load increment of N11.Finally, the service delay of G2 increases after executing thesingle user load sharing process. Furthermore, it is evidentthat load sharing is sufficient for assuring service delayguarantee in G1 as well as the winner gateway G2, while theothers remain unchanged.

6.2.2 Bandwidth Distribution

We compared the final bandwidth distribution amongthe gateways for the proposed single user load sharingalgorithm and the existing bandwidth distribution algo-rithm, AQUM. We plotted the results in Fig. 3. In thisexperiment, we use the default settings for load sharing.For the AQUM algorithm, we adopt the similar data set-tings except that the additional shifting parameter is set10:0 for all the gateways. In this figure, we observe that thebandwidth distribution remains undisturbed for the singleuser load sharing algorithm, whereas the bandwidth dis-tribution of all the 10 gateways completely differ in thecase of the AQUM algorithm. The CSP allocates morebandwidth to the gateway G1, as it gets additional servicerequest from its down-link user. Hence, we conclude thatthe gateways try to resolve the additional service requestamong themselves in the single user load sharing process,so that additional bandwidth is not required. Thus, ourproposed algorithm does not suffer from any additionaloverhead due to bandwidth management.

We further compared the service delay of all the gate-ways for both the algorithms. Fig. 4 shows that only for thewinner gateway G2, the service delay increases in the singleuser load sharing process, while the others remain unal-tered. Using the AQUM algorithm, the service delay of allthe gateways change due to the varying bandwidth distri-bution. Hence, our proposed algorithm is similar to AQUM,and maintains the service delay of all the gateways within

Fig. 2. Load sharing for single user load increment with overloaded gate-way G1 and winner gatewayG2.

Fig. 3. Bandwidth allocation comparison between AQUM andS-QuaLShare.



the service delay threshold value. In the S-QuaLShare algo-rithm, only the neighbors of the overloaded gateway partici-pate. Hence, the gateways except the neighbors of theoverloaded one remain isolated from the load sharing pro-cess and provide uninterrupted services to their associatedmobile users, while the overloaded gateway involves in theload sharing process. Additionally, the proposed algorithmincludes only the neighbors of the overloaded gateway withsufficient bandwidth. Thus, only a limited part of the MCCsystem is involved in the S-QuaLShare algorithm, whereasAQUM includes the entire MCC system. Hence, we con-clude that using the proposed solution, the performance ofS-QuaLShare is better than AQUM, due to the lower band-width management overhead and limited participation ofthe gateways.

6.2.3 Optimality

We test the optimality criteria of the single user load sharingalgorithm. We performed this test with the default parame-ter settings. Fig. 5 shows that the payoff initially increaseswith the increase of the bid value. At the optimal point, thepayoff value is maximum. The payoff decreases with theincrease of the bid value beyond the optimal point. There-fore, the participating neighbor gateway accepts the bid at

the optimal point to get the maximum benefit. We executethe single user load sharing process 30 times and plot theresults with 95 percent degree of confidence. It is pertinentto mention that the length of the confidence interval is vari-able in nature due to the random selection of bid incrementin each iteration. Further, the interval length is increasing innature with the increase of iteration index because of thecumulative effect of Dp on the computation of p as shown inEquation (13). Finally, we observe that the algorithmreaches the optimal value on an average at iteration index 4according to our experimental setup.

6.2.4 Nash Equilibrium

Wemeasured the payoff valuewith varying n by keeping thevalues of the other parameters, as shown in Section 6.1. Weobserve in Fig. 6 that the payoff maximizes at iteration index4 for n ¼ 1:5. The maximum payoff indicates the existence ofNash Equilibrium in the proposed single user load sharingalgorithm. We further investigate that the algorithm isunable to reach the maximum point for n ¼ 1:00 and 1:25, asthe bid value reaches the maximum limit and the algorithmterminates fast for n ¼ 1:75without getting optimal payoff.

6.3 Multiple Users Load Increment

In all the multiple users load increment experiments, weconsider that the mobile users N11; N12; and N13 connectedwith the gatewayG1 increase their demand, so that the idealtransmission times for their demand increase from 0:1 sec to0:2; 0:3; and 0:2 sec, respectively. The mobile users N11; N12;andN13 are within the vicinity of all the neighbors of G1.

6.3.1 Multiple Users Load Sharing

Fig. 7 establishes the necessity of load sharing due to multi-ple users load increment. When the users N11; N12; and N13

increase their demands, the ideal transmission time of ser-vice demands increases. This affects the service delay of theconnecting gateway G1 and the value exceeds the thresholdlimit. In this experiment, multiple users load sharingensures service delay guarantees in the overloaded gatewayG1, as well as the winner gateways G3; G2; and G5. The ser-vice delay of the others remain the same during the sharingprocess.Fig. 5. Optimality of S-QuaLShare algorithm.

Fig. 4. Service delay comparison between AQUM and S-QuaLShare. Fig. 6. Nash equilibrium of S-QuaLShare algorithm.



6.3.2 Bandwidth Distribution

We analyze the gateways’ final bandwidth distribution forboth the proposed multiple users load sharing and theAQUM algorithms. In AQUM, we use 10 as the value ofthe shifting parameter and the remaining all the valuesare the same as the default settings of the multiple usersload sharing process. Fig. 8 shows that the amount of allo-cated bandwidth remains the same for all the gateways inthe multiple user load sharing process, whereas that valuediffers for all the gateways used in the AQUM algorithm.As three users of gateway G1 requests additional service,the CSP distributes the bandwidth to G1 with a greaterproportion. Therefore, we conclude that only a few gate-ways involve to resolve the incremental service request ofa particular gateway in the multiple users load sharingprocess, while keeping the bandwidth amount the same.Hence, M-QuaLShare does not suffer from the overheaddue to bandwidth management.

We further compared the service delay of all the gatewaysfor both the algorithms. We plotted the received outcomes inFig. 9. From this figure, we observe that the service delayincreases only for the winner gateways in the multiple usersload sharing process. The service delay varies for all the

gateways using AQUM, as the bandwidth is completelyredistributed among all the gateways, which are connectedto the CSP. Additionally we observed that, similar toAQUM, M-QuaLShare also maintains the service delaywithin the service delay threshold. Besides the no bandwidthmanagement overhead, the M-QuaLShare algorithm onlyconsiders the nearby neighbor gateways of the overloadedone. Therefore, the proposed algorithm involves minimalnumber of gateways for sharing the loads, whereas AQUMdeals with the entire set of gateways present in the system.Hence, we conclude that the performance of M-QuaLShareis better than AQUM in the localized scenarios, where theoverloaded gateway has sufficient underloaded neighbors.

6.3.3 Optimality

Fig. 10 shows the optimality of the proposed multiple userload sharing algorithm. In this experiment, we computedthe payoff and bid values for each iteration cycle. The opti-mal point is a point at which the payoff is maximum andbeyond this point the users do not receive any additionalbenefit. With the default parameter settings, we observe thatat the beginning of the experiment the payoff increases withthe increase of the bid value, the payoff value is maximumat the optimal point, and finally, the value decreases. We

Fig. 8. Bandwidth allocation comparison between AQUM and M-QuaLShare.

Fig. 9. Service delay comparison between AQUM and M-QuaLShare.Fig. 7. Load sharing for multiple user load increment with overloadedgatewayG1 and winner gatewaysG3, G2 and G5.

Fig. 10. Optimality of M-QuaLShare algorithm.



perform the multiple users load sharing process 30 times,and on an average, we reach the optimality at iteration 10.

6.3.4 Nash Equilibrium

We computed the payoff value with varying n in the defaultdata setup. Fig. 11 depicts that the payoff initially increasesfor all values of n. For n ¼ 1:5, the payoff maximizes at itera-tion index 10. Additionally, the plotted result shows that thealgorithm concludes within a fewer iterations for n ¼ 1:6and 1:7, before reaching the optimal value. On the otherhand, for n ¼ 1:4, the algorithm does not terminate, as thebidding value crosses the maximum limit. Hence, we con-clude that the maximum payoff indicates the existenceof Nash Equilibrium in the proposed multiple users loadsharing algorithm.

7 CONCLUSION

In this paper, we identified and addressed the problem ofload sharing for distributing the loads of the overloadedgateway in an MCN environment. The load sharing processdiffers from the traditional load balancing process [2] inthat while the former concerns sharing the additionaldemand with local gateways, the latter concerns equalizingthe service demand among all the gateways. We furtherconsider the load sharing problem in the presence of misbe-having gateways, where they provide inadequate informa-tion for getting the access of the mobile users who haveincreased their demand. Thereafter, the winner misbehav-ing gateways share those users with their neighbors to gainadditional benefit. We have proposed an auction-basedQoS-guaranteed secure load sharing algorithm for maximiz-ing the utility of the overloaded gateway in both single userand multiple users load increment scenarios. We haveverified the requirement of load sharing through numericalanalysis. We have also investigated the existence of NashEquilibrium and the optimality criteria of the proposedalgorithms.

Even though the proposed algorithms are able to distin-guish the misbehaving gateways from all the participantsusing the information partially provided by the CSP, inreal environments, this information sharing is not alwaysfeasible. Therefore, a self-computing scheme is required.

In this work, we have considered single hop connectionbetween the gateway and the mobile user. In the future,we also envision to work on an architecture having multi-hop connectivity between the gateway and the mobileuser. On the other hand, we assume that the servicerequest of all the mobile users are independent of oneanother. Hence, the discussion regarding the intra-depen-dency and inter-dependency of the users’ services need tobe discussed in the future.

REFERENCES

[1] S. Misra, S. Das, M. Khatua, and M. S. Obaidat, “QoS-guaranteedbandwidth shifting and redistribution in mobile cloud environ-ment,” IEEE Trans. Cloud Comput, vol. 2, no. 2, pp. 181–193, Apr.-Jun. 2014.

[2] J. Doyle, R. Shorten, and D. OMahony, “Stratus: Load balancingthe cloud for carbon emissions control,” IEEE Trans. Cloud Com-put., vol. 1, no. 1, pp. 116–128, Jan.-Jun. 2013.

[3] S. Zaman and D. Grosu, “A combinatorial Auction-based mecha-nism for dynamic vm provisioning and allocation in clouds,”IEEE Trans. Cloud Comput., vol. 1, no. 2, pp. 129–141, Jul.-Dec.2013.

[4] M. Randles, D. Lamb, and A. Taleb-Bendiab, “A comparativestudy into distributed load balancing algorithms for cloudcomputing,” in Proc. IEEE 24th Int. Conf. Adv. Inf. Netw. Appl.Workshops, 2010, pp. 551–556.

[5] S. T. Maguluri, R. Srikant, and L. Ying, “Stochastic models of loadbalancing and scheduling in cloud computing clusters,” in Proc.IEEE INFOCOM, 2012, pp. 702–710.

[6] D. Grosu and A. T. Chronopoulos, “Noncooperative load balanc-ing in distributed systems,” J. Parallel Distrib. Comput., vol. 65,pp. 1022–1034, 2005.

[7] H. T. Dinh, C. Lee, D. Niyato, and P. Wang, “A survey of mobilecloud computing: Architecture, applications, and approaches,”Wireless Commun. Mobile Comput., vol. 13, no. 18, pp. 1587–1611,Dec. 2013.

[8] N. Fernando, S. W. Loke, and W. Rahayu, “Mobile cloud comput-ing: A survey,” Future Generation Comput. Syst., vol. 29, no. 1,pp. 84–106, Jan. 2013.

[9] Z. Sanaei, S. Abolfazli, A. Gani, and R. Buyya, “Heterogeneity inmobile cloud computing: Taxonomy and open challenges,” IEEECommun. Surveys Tuts., vol. 16, no. 1, pp. 369–392, Jan.-Mar. 2014.

[10] S. Abolfazli, Z. Sanaei, E. Ahmed, A. Gani, and R. Buyya, “Cloud-based augmentation for mobile devices: Motivation, taxonomies,and open challenges,” IEEE Commun. Surveys Tuts., vol. 16, no. 1,pp. 337–368, Jan.-Mar. 2014.

[11] Z. Sanaei, S. Abolfazli, A. Gani, and M. Chen, “HMCC: A hybridmobile cloud computing framework,” in Proc. IEEE Int. Conf.Mobile Cloud Comput., Serv., Eng., 2015, pp. 157–162.

[12] H. Hu, Z. Li, and H. Hu, “An Anti-cheating bidding approach forresource allocation in cloud computing environments,” J. Comput.Inf. Syst., vol. 8, no. 4, pp. 1641–1654, Feb. 2012.

[13] C. Rong, S. T. Nguyen, and M. G. Jaatun, “Beyond lightning: Asurvey on security challenges in cloud computing,” Comput. Elect.Eng., vol. 39, no. 1, pp. 47–54, Jan. 2013.

[14] S. Abolfazli, Z. Sanaei, A. Gani, F. Xia, and L. T. Yang, “Richmobile applications: Genesis, taxonomy, and open issues,”J. Netw. Comput. Appl., vol. 40, pp. 345–362, 2014.

[15] A. N. Khan, M. M. Kiah, S. U. Khan, and S. A. Madani, “Towardssecure mobile cloud computing: A survey,” Future GenerationComput. Syst., vol. 29, no. 5, pp. 1278–1299, Jul. 2013.

[16] M. Othman and S. Hailes, “Power conservation strategy formobile computers using load sharing,” ACM SIGMOBILE MobileComput. Commun. Rev., vol. 2, no. 1, pp. 44–51, Jan. 1998.

[17] D. L. Eager, E. D. Lazowska, and J. Zahorjan, “Adaptive load shar-ing in homogeneous distributed systems,” IEEE Trans. Softw. Eng.,vol. 12, no. 5, pp. 662–675, May. 1986.

[18] R. Mirchandaney, D. Towsley, and J. A. Stankovic, “Adaptive loadsharing in heterogeneous distributed systems,” J. Parallel Distrib.Comput., vol. 9, pp. 331–346, 1990.

[19] H. Wang, J. Ren, and T. Li, “Resource allocation with load balanc-ing for cognitive radio networks,” in Proc. IEEE Global Telecom-mun. Conf., Dec. 2010, pp. 1–5.

Fig. 11. Nash Equilibrium of M-QuaLShare algorithm.



[20] X. Wang, Z. Li, P. Xu, Y. Xu, X. Gao, and H. Chen, “Spectrumsharing in cognitive radio networks-an auction-based approach,”IEEE Trans. Syst., Man, Cybern., Part B: Cybern., vol. 40, no. 3,pp. 587–596., Jun. 2010.

[21] K. Zhu, D. Niyato, and P. Wang, “Dynamic bandwidth allocationunder uncertainty in cognitive radio networks,” in Proc. IEEEGlobal Telecommun. Conf., 2011, pp. 1–5.

[22] J. Elias, F. Martignon, and A. Capone, “An efficient dynamicbandwidth allocation algorithm for quality of service networks,”in Proc. Autonomic Netw., Sep. 2006, pp. 132–145.

[23] Z. Kun, D. Niyato, and P. Wang, “Optimal bandwidth allocationwith dynamic service selection in heterogeneous wirelessnetworks,” in Proc. IEEE Global Telecommun. Conf., 2010, pp. 1–5.

[24] Y. Zhang, C. Lee, D. Niyato, and P. Wang, “Auction approachesfor resource allocation in wireless systems: A survey,” IEEECommun. Surveys Tuts., vol. 15, no. 3, pp. 1020–1041, Jul.-Sep.2013.

[25] K. A. Nuaimi, N. Mohamed, M. A. Nuaimi, and J. A. Jaroodi, “Asurvey of load balancing in cloud computing: Challenges andalgorithms,” in Proc. IEEE 2nd Symp. Netw. Cloud Comput. Appl.,2012, pp. 137–142.

[26] C. Papagianni, A. Leivadeas, S. Papavassiliou, V. Maglaris, C. C.Pastor, and A. Monje, “On the optimal allocation of virtual resour-ces in cloud computing networks,” IEEE Trans. Comput., vol. 62,no. 6, pp. 1060–1071, Jun. 2013.

[27] A. Amamou, M. Bourguiba, K. Haddadou, and G. Pujolle, “Adynamic bandwidth allocator for virtual machines in a cloud envi-ronment,” in Proc. IEEE Consum. Commun. Netw. Conf., Jan. 2012,pp. 99–104.

[28] S. Das, S. Misra, M. Khatua, and J. J. P. C. Rodrigues, “Mapping ofsensor nodes with servers in a mobile Health-cloud environ-ment,” in Proc. IEEE 15th Int. Conf. e-Health Netw., Appl. Serv.,Lisbon, Portugal, Oct. 2013, pp. 481–485.

[29] J. W. Lin, C. H. Chen, and J. M. Chang, “QoS-aware data replica-tion for data intensive applications in cloud computing systems,”IEEE Trans. Cloud Comput., vol. 1, no. 1, pp. 101–115, Nov. 2013.

[30] Z. Yang, N. Dusit, and W. Ping, “An auction mechanism forresource allocation in mobile cloud computing systems,” in Proc.8th Int. Conf. Wireless Algorithms, Syst., Appl., Berlin, Heidelberg,2013, pp. 76–87.

[31] X. Wang, X. Wang, C. li Wang, K. Li, and M. Huang, “Resourceallocation in cloud environment: A model based on double Multi-attribute auction mechanism,” in Proc. IEEE Int. Conf. Cloud Com-put. Technol. Sci., 2014, pp. 599–604.

[32] M. Zhou, R. Zhang, W. Xie, W. Qian, and A. Zhou, “Security andprivacy in cloud computing: A survey,” in Proc. 6th Int. Conf.Semantics, Knowl. Grids, 2010, pp. 105–112.

[33] Y. Chen, Y. Wu, B. Wang, and K. J. R. Liu, “Spectrum auctiongames for multimedia streaming over cognitive radio networks,”IEEE Trans. Commun., vol. 58, no. 8, pp. 2381–2390, Aug. 2010.

[34] S. Subashini and V. Kavitha, “A survey on security issues in ser-vice delivery models of cloud computing,” J. Netw. Comput. Appl.,vol. 34, no. 1, pp. 1–11, Jan. 2011.

[35] A. Verma and S. Kaushal, “Cloud computing security issues andchallenges: A survey,” in Proc. 1st Int. Conf. Adv. Comput. Com-mun., Kochi, India, Jul. 2011, pp. 445–454.

Snigdha Das received the BTech degree ininformation technology from the West BengalUniversity of Technology, India in 2010 and sheis currently working toward the MS degree fromthe School of Information Technology, IndianInstitute of Technology Kharagpur, India, and isworking as a senior project officer in Develop-ment of Feasibility Assessment Model forAdaption of Underground Coal gasification Tech-nology in North-Eastern Region of India fundedby DeitY, Government of India. She was an asso-

ciate software engineer in CSC for two years. Her current research inter-ests include mobile computing and cloud computing.

Manas Khatua received the BTech degree incomputer science and engineering from the Uni-versity of Kalyani, India, in 2003, the MTechdegree in information technology from BengalEngineering and Science University, India, in2007, and is currently working toward the PhDdegree from the School of Information Technol-ogy, Indian Institute of Technology Kharagpur,India. Presently, he is a research assistant of theSingapore University of Technology and Design,Singapore. Prior to this, he was associated with

Tata Consultancy Services, Kolkata, for two and half years. He was alecturer of the Bankura Unnayani Institute of Engineering, India, formore than two years. His research interests include the performanceevaluation and resource optimization in Wireless LANs. He also workson sensor networks, network security, and mobile cloud computing. Heis one of the recipients of NIXI Fellowship 2014. He is a student memberof the IEEE.

Dr. Sudip Misra received the PhD degree incomputer science from Carleton University,in Ottawa, Canada. He is an associate professorin the School of Information Technology at theIndian Institute of Technology Kharagpur. Prior tothis, he was associated with Cornell University(USA), Yale University (USA), Nortel Networks(Canada), and the Government of Ontario(Canada). He has received eight research paperawards in different conferences. He was awardedthe IEEE ComSoc Asia Pacific Outstanding

Young Researcher Award at IEEE GLOBECOM 2012. He was alsoawarded the Canadian Governments prestigious NSERC Post DoctoralFellowship and the Humboldt Research Fellowship in Germany. He is asenior member of the IEEE.

Mohammad S. Obaidat (F’05) received the PhDdegree in computer engineering from Ohio StateUniversity. He is an internationally well-knownacademic/ researcher/ scientist. He currently thechair and full professor of computer & informationscience at Fordham University. Among his previ-ous positions are a chair of the Computer Sci-ence Department and the director of GraduateProgram at Monmouth University, a dean ofengineering at Prince Sultan University, and anadvisor to the president of Philadelphia Univer-

sity. He served as an advisor to the president of Philadelphia Universityand as a president, senior VP, VP membership, and VP conferences ofSCS. He has received extensive research funding and has publishedmore than 38 books and 600 refereed technical articles. He is an editor-in-chief of three scholarly journals and an editor of many other interna-tional journals. He received numerous awards. He is a fellow of the IEEEand SCS.

" For more information on this or any other computing topic,please visit our Digital Library at www.computer.org/publications/dlib.



Quality-Assured Secured Load Sharing in Mobile Cloud ... · quality-assured secured load sharing...

Documents

Transcript of Quality-Assured Secured Load Sharing in Mobile Cloud ... · quality-assured secured load sharing...