Providing service assurance in mobile opportunistic networksxinwenfu/paper/Journals/14_CN_74... ·...

Computer Networks 74 (2014) 114–140

Contents lists available at ScienceDirect

Computer Networks

journal homepage: www.elsevier .com/locate /comnet

Providing service assurance in mobile opportunistic networks

http://dx.doi.org/10.1016/j.comnet.2014.07.0181389-1286/� 2014 Elsevier B.V. All rights reserved.

⇑ Corresponding author. Tel.: +1 978 944 6802.E-mail addresses: [email protected] (B. Kaushik), honggang@

storm.cis.fordham.edu (H. Zhang), [email protected] (X. Yang),[email protected] (X. Fu), [email protected] (B. Liu), [email protected](J. Wang).

Bhanu Kaushik a,⇑, Honggang Zhang b, Xinyu Yang c, Xinwen Fu a, Benyuan Liu a, Jie Wang a

a Department of Computer Science, University of Massachusetts, Lowell, MA 01854, USAb Department of Computer and Information Science, Fordham University, Bronx, NY 10458, USAc Department of Computer Science and Technology, Xi’an Jiaotong University, PR China

a r t i c l e i n f o

Article history:Received 16 December 2013Received in revised form 1 July 2014Accepted 2 July 2014Available online 26 September 2014

Keywords:AuctionAd-hoc networkCollaborative data sharingMobile data offloadingService assuranceWireless network

a b s t r a c t

Concomitant to the growing popularity of Internet enabled mobile devices such as smart-phones, tablets, PDAs, portable media players etc., however, are the concerns about avail-ability of Internet access points for these devices. Network traffic originating from thesedevices has marked a multi-fold increase over the past few years. This increase in networktraffic has in turn resulted in increased content availability and maintainability cost to thecellular service providers. Major challenges faced by mobile Internet users are limitedhardware resources (3G or LTE) or exhaustion of carrier enforced data plans. Thus, usersoften either overpay for service availability such as (3G or LTE) or suffer incapability ofaccessing Internet services. In this paper, we argue that the ad-hoc network comprisingof spatio-temporally co-existing users, identified as ‘‘Familiar Strangers’’ could beharnessed for viral dissemination of mobile application data. We present the architectureof a collaborative data sharing framework which comprises of two services, namelySmartParcel and CollabAssure. ‘‘SmartParcel’’ allows users to share mobile application datawith their spatio-temporally co-existing peers. This service allows a user to receive appli-cation specific data from its neighboring nodes instead of cloud infrastructure. This reducesthe network overhead generated by a large number of devices and thus, reducing contentmaintainability and dissemination costs at the server side. This system, however, suffersfrom the ‘‘tragedy of commons’’ problem, to overcome which we present the design ofCollabAssure’’. ‘‘CollabAssure’’ is an auction based, ad-hoc market model, assuring serviceby giving users incentives to ‘‘sublet’’ their surplus data plans to the users without Internetaccess. CollabAssure service allows users to participate in ad hoc auctions with their spatio-temporally co-existing neighbors as buyers or sellers of the data bandwidth. This in turnreduces under utilization overhead of the overpaying mobile data users. Users with limitedor no data plans also benefit from CollabAssure service by receiving services from ad-hocpeers. We perform trace based simulations by varying data refresh rates, allowed serverconnections, probability of user participation and analyze the data availability at differenttimes of the day. Our simulation results advocate a higher Internet data service is ensuredby using our proposed data sharing framework.

� 2014 Elsevier B.V. All rights reserved.

1. Introduction

The availability of a huge number of diverse applica-tions is one of the major reasons why consumers favormobile devices [1]. There are about 6.2 billion users around

http://crossmark.crossref.org/dialog/?doi=10.1016/j.comnet.2014.07.018&domain=pdf

http://dx.doi.org/10.1016/j.comnet.2014.07.018

mailto:[email protected]







http://dx.doi.org/10.1016/j.comnet.2014.07.018

http://www.sciencedirect.com/science/journal/13891286

http://www.elsevier.com/locate/comnet

B. Kaushik et al. / Computer Networks 74 (2014) 114–140 115

the globe [2,3] and the proliferation of mobile Internetusage among consumers has cast the spotlight on the datatraffic that originates from these devices. There has been arapid growth in network traffic, i.e., 100 petabytes/monthin 2007 to 700 petabytes/month in 2012 [3], mainly dueto the use of mobile applications on the Internet.

The demand for higher data rates in wireless networksto support these applications has triggered the designand development of new data-minded cellular standards,such as WiMAX (802.16e), 3GPPs, HSPA, and LTE standards.In addition, Wi-Fi mesh networks are being developed toprovide nomadic high-rate data services in a more distrib-uted manner [4]. The emergence of these new mobile andwireless networks provides major opportunities to expandtraditional mobile Internet-based applications. However,the best way to increase the system capacity of a wirelesslink is by moving the transmitter and receiver closertogether, which obtains the dual benefits of higher-qualitylinks and greater spatial reuse. The rise of the mobile userbase has been accompanied by service disruption; thus, ina network with nomadic users, the deployment of furtherinfrastructure, typically in the form of micro-cells, hotspots, distributed antennae, or relays, is crucial for provid-ing a better service. Thus, satisfying the growth in userneeds is a burden for cellular service providers.

To meet the growing needs of the Internet user base,cellular network providers offer wide coverage, therebyenabling ubiquitous wireless data access services. How-ever, this service is expensive and cellular service provid-ers only allow users to subscribe to a fixed data limit,which they often exhaust before the end of billing cycleor underutilize due to the overall available limits at theend of the billing cycle. Alternatively, users may choose‘‘pay-per-use’’ schemes, where they pay significantlyhigher prices for the services used. In these scenarios, userseither experience overpayment for higher service availabil-ity or are rendered incapable of accessing Internet serviceson occasions. Due to these issues and constraints, the over-all mobile data user base can be roughly divided into twogroups at any given moment in time: those with Internet

Fig. 1. Diversity of mobile Internet users.

access capabilities and those with no Internet accesscapability, as shown in Fig. 1. Note that some devices areconnected to the Internet via 3G/LTE services, whereasothers may be connected via Wi-Fi. The other majorchallenges faced by mobile Internet users include theunavailability of hardware (3G or LTE), unavailability ofaccess points, service outages, and network and serveroverloads. These challenges result in the unavailability ofapplication data to users and high service maintenancecosts for both the service providers and hosting servers.

These issues have been addressed by researchers inindustry and academia. For example, a previous study [5]proposed the architecture of a macro-femto strategy formacrocell offloading. In this femtocell environment, thedata traffic flows over the air interface to the femtocell(which is connected to the user’s broadband connection),and then over the Internet to the operator’s core networkand/or to other Internet destinations [6]. In this setting,when a subscriber enters the coverage area of the femto-cell, the users equipment associates with it automaticallyand traffic that previously flowed between the macrocelland the users equipment now flows through the femtocelland the subscriber’s broadband connection. The majordrawback of this strategy is that data offloading occurs pri-marily indoors (homes or offices) and it cannot be scaled.

In an alternative strategy, Wi-Fi is used mainly for off-loading the overburdened cellular network traffic. Indeed,due to degradation of cellular services in overloaded areas,an increasing number of users are already using Wi-Fi toaccess Internet services to obtain a better experience. Fromthe service provider’s perspective, Wi-Fi is attractivebecause it allows data traffic to be shifted from expensivelicensed bands to free unlicensed bands (2.4 GHz and5 GHz). Studies have shown that expanding networksusing Wi-Fi is significantly less expensive than a networkrollout. The architecture of a large-scale data offloadingscheme was presented in [7] and [8]. In addition, [9] and[10] addressed this issue by using the Wi-Fi proximity ofusers to provide data only to Influential Nodes (nodes withthe highest out-degree in a social graph) or by scanningWi-Fi routers, which then transfer data to other nodes inclose proximity to achieve maximal coverage in a socialgraph. Although promising, these approaches require sub-stantial changes to state-of-the-art software and hardwaretechnologies, and they do not consider the heterogeneity ofapplication data. Hence, there is a critical need for an infra-structure for data sharing and content availability, and spa-tio-temporal coexisting neighbors appear to be the perfectchoice.

If we consider a typical scenario, such as ‘‘reading thenewspaper on the subway,’’ many users tend to read news-papers on their mobile devices while they commute to theirfinal destinations. However, the users without 3G or LTEhardware and data plans are unable to achieve this taskin a subway setting because there is a lack of access points.In addition, the 3G/LTE data users have to buy a high databandwidth in order to access multimedia data from thenews servers, which they do not fully utilize by the end ofthe monthly billing cycle. In this scenario, our proposedframework allows these users to exchange their application(news) data with each other via ad hoc interfaces. This

116 B. Kaushik et al. / Computer Networks 74 (2014) 114–140

inherently reduces the number of requests sent to the dataserver and a better service is provided to the mobileapplication users. However, in this type of ad hoc networksetting, the users might act selfishly and fail to participatein data sharing. In this event, the framework also allowsusers to participate in ad hoc auctions, where the buyerspay for the data service provided by the sellers.

In this study, we present the design and implementa-tion of a collaborative data sharing framework to addressthe issue of cellular data offloading and service assurance,which harnesses the spatio-temporal network of coexis-ting mobile devices, i.e., ‘‘familiar strangers’’ [11]. Thesedecentralized mobile ad hoc networks are characterizedby their completely autonomous, dynamic, self-organized,and ubiquitous nature. We present the architecture of acollaborative data sharing framework that comprises twoservices: SmartParcel and CollabAssure. ‘‘SmartParcel’’allows users to share mobile application data with theirspatio-temporal co-existing peers. This service allows auser to receive application-specific data from neighboringnodes instead of a cloud infrastructure. This reduces thenetwork overheads generated by a large number ofdevices, thereby reducing the content maintenance anddissemination costs on the server side. However, this sys-tem could be affected by the ‘‘tragedy of the commons’’problem. Therefore, ‘‘CollabAssure’’ is an auction-based, adhoc market model, which ensures the provision of a serviceby giving users incentives to ‘‘sublet’’ their surplus dataplans to users without Internet access. The CollabAssureservice allows users to participate in ad hoc auctions withtheir spatio-temporal co-existing neighbors as buyers orsellers of the data bandwidth. This also reduces theunder-utilization overheads of the overpaying mobile datausers. Users with limited or no data plans also benefit fromthe CollabAssure service by receiving services from their adhoc peers.

Together, ‘‘SmartParcel’’ [12] and CollabAssure addressthe issue of the increasing volumes of online digitalcontent that originate from Internet-based applicationsrunning on various mobile devices. This framework is a‘‘one-for-all’’ multiple incentive system for applicationdevelopers, Internet service providers, and application dataproviders (e.g., cloud services), which provides additionalbenefits to consumers. It should be attractive to applica-tion developers and consumers because it offers greaterdata availability and minimizes cellular data usage, espe-cially for devices running on limited data plans, or whenInternet access points are unavailable. In terms of energyconsumption, the reliance on cheaper network technolo-gies (Bluetooth and Wi-Fi) compared with 3G and LTE ser-vices makes the framework more adaptable for consumers[13]. Excluding consumer incentives, lower network trafficand higher service availability can be achieved at muchlower costs, with fewer data requests to applicationservers.

Our proposed framework is designed to attract users,application developers, and Internet service providers(ISPs) by facilitating data sharing and increased serviceavailability. The increased availability of application datato users means that the framework satisfies the needs ofapplication developers and data hosting servers. Together,

the SmartParcel and CollabAssure services reduce the net-work overheads generated by the mobile devices. Smart-Parcel is a best-effort service where a node sends a datarequest to the server only when the application data areunavailable from neighboring nodes. By contrast, a requestis made on-demand with CollabAssure. Users only act asdata sellers when they have an abundance of data to subletafter meeting their own requirements. This reduces theoverheads for the ISP by decreasing resource utilization,especially during peak periods.

We performed trace-based simulation experimentsusing various data refresh rates, allowed server connec-tions, and probabilities of user participation, and we ana-lyzed the data availability at different times of day. Oursimulation results demonstrate that a better Internet dataservice is ensured by using our proposed data sharingframework. We found that higher data availability wasachievable using SmartParcel when increasing numbers ofdevices were connected to the server. Specifically, whenwe allowed one, 20, and 50 devices to connect and receivedata from the source, the data availability levels achievedwere 40%, 85%, and 98%, respectively. Our results also indi-cate that human social activity levels play a vital role indata dissemination. With higher human activity levels dur-ing 8:00 am to 8:00 pm, the data availability was 60–85%and it increased with the number of allowed server con-nections. Due to the lower human social activity levels inthe night and early morning hours, we found that the dataavailability only reached up to 12% during these hours. Wealso considered the unwillingness of users to participate indata sharing. The key factor that determined the unwilling-ness of users to participate was a concern about energyconsumption. We found that a higher user participationrate resulted in greater data availability. In particular, with10%, 20%, and 90% user participation rates, the data avail-ability was 65%, 70%, and 85%, respectively, when only 10devices were allowed to receive data from the server. ForCollabAssure, we found that the data refresh rates, buyerparticipation probability, and seller participation probabil-ity significantly affected the overall market utility, whichwas the total amount of bandwidth sold in our study. Evenwith limited participation by users in both the buyer andseller roles, the market utility was still improved signifi-cantly. The user activity was very high during the daytimewhen more trades were observed, whereas the number oftrades was relatively low during the early morning and lateat night.

In summary, the main contributions of our study are asfollows.

� We harness the ‘‘familiar strangers’’ network formed byspatio-temporal coexisting mobile users to facilitatecollaborative data sharing among these nomadic mobileusers.� We present and evaluate the design of a ‘‘best effort ser-

vice’’ called SmartParcel for data offloading and contentdissemination between mobile devices.� We also present the design of an ad hoc auction strategy

that allows users to sublet mobile bandwidth. Toprevent the tragedy of the commons problem fromaffecting the SmartParcel approach, we present the


architecture of a service assurance framework calledCollabAssure. Our proposed CollabAssure frameworkallows users to sublet their cellular data plans.� We evaluated the performance of both proposed ser-

vices, SmartParcel and CollabAssure, using MIT realitymining data and our results demonstrated that betterservice availability, data dissemination, and bandwidthutilization were obtained with the proposed system.

The remainder of this paper is organized as follows. Sec-tion 2 provides an overview of related work. The design andanalysis of SmartParcel are presented in Section 3, and thedesign and analysis of CollabAssure are presented in Sec-tion 4. Finally, we provide our conclusions in Section 5.

2. Related work

The competition in the emerging mobile operating sys-tems (OSs) race, the number and diversity of availableapplications, user behavior [14], usage patterns [15], andsecurity concerns [16–20] have made mobile device-basedresearch a major focus in academia and industry in the lastcouple of years. Android [21] is one of the most popularand fully customizable software stacks for mobile devicesand, because it is open source, it has attracted much atten-tion from researchers. The Android mobile software frame-work is supported by Google and it comprises an OS,system utilities, and middleware in the form of a virtualmachine. The Android platform includes about 130 appli-cation-level permissions, which allow applications to usedifferent access controls and hardware resources. Thesepermissions are set by the application developer and theyare requested at installation, and then silently enforcedwhen the application is executed. The SmartParcel and Col-labAssure frameworks are both implemented in theAndroid framework, although their modular architectureallows them to be extended to other mobile platforms.

Mobile devices continue to increase in popularity andpervasiveness. The capabilities of these devices continueto grow with each new generation. In particular, devicesare now more commonly equipped with short-range radiocommunication capabilities. These short-range radios,such as Bluetooth, are ideal for mobile devices because oftheir relatively low power requirements. Bluetooth-enabled mobile devices are particularly suitable for newpeer-to-peer communication applications because theyoffer low-power short-range data transfers [22]. In addi-tion, because these devices are increasingly pervasive,almost anyone with a Bluetooth device in their pocketcan be a potential participant in the data forwardingprocess. To enable data transfer in these scenarios withBluetooth, the neighboring devices first need to discovereach other, i.e., learn about each other’s presence.

2.1. Cellular data offloading

Cellular data offloading, also referred to as mobile trafficoffloading, is the use of complementary network commu-nication technologies to deliver mobile data traffic thatwas originally planned for transmission over cellular

networks. In the original delay-tolerant approach, delaysare usually caused by intermittent connectivity. For exam-ple, motivated by the fact that the coverage of WLAN hot-spots may be very limited and that mobile users might notalways be able to connect to the Internet via them, a previ-ous study [23] explored opportunistic web access viaWLAN hotspots for mobile phone users. Thus, we proposethe offloading of cellular data and information via freeopportunistic communications with the goal of reducingmobile data traffic. Some cellular data offloading tech-niques route data traffic via alternative means such asWi-Fi or femtocell networks. At present, the consensusopinion is that data offloading is a cost-effective [24] andenergy-prudent method, which benefits operators andmobile users [9].

2.1.1. Femtocell for indoor environmentsThe femtocell technique is a good solution for indoor

voice and data services in cellular networks. Femtocellsoperate on the same licensed spectrum as the macrocellsof cellular networks [25]. The major advantages of usingfemtocells include the lack of a requirement for specialhardware support, although it is necessary to install a basestation inside an indoor complex, such as home or office,which then connects to the Internet [5]. Femtocells offera low power alternative and their small size makes themsuitable for deployment in an indoor setting [26]. How-ever, the major drawback of this strategy is that data off-loading occurs primarily indoors (homes or offices) and itcannot be scaled to a bigger setting, thus it is not suitablefor a true ad hoc setting.

2.1.2. Offloading traffic to Wi-Fi networksThe increasing number of devices and their high

reliance on cellular data have focused the attention ofdevelopers on cellular data offloading schemes. This hasbeen addressed by the rapid deployment and trouble-freeoperation of large-scale Wi-Fi networks solutions. Thus,Alvarion [8] and Cisco [7] ensure a superior user experi-ence and satisfaction with a high quality Wi-Fi service byallowing carriers to optimize 3G and LTE network services,and by switching from an untrusted Wi-Fi network to atrusted and integral part of a carrier’s network. In this con-text, Han et al. [9] considered the issue of informationdelivery based on target-set selection in a social context,while Lee et al. [10] allowed data dissemination by Wi-Fiin the case of data bottlenecks or service delays. In anotherapproach [27], Wi-Fi was used for offloading the overbur-dened cellular network traffic. Due to the degradation ofcellular services in overloaded areas, an increasing numberof users are already using Wi-Fi to access Internet servicesto obtain a better experience. However, from the serviceprovider’s perspective, Wi-Fi is attractive because it allowsdata traffic to be shifted from expensive licensed bands tofree unlicensed bands (2.4 GHz and 5 GHz). Studies haveshown that expanding networks using Wi-Fi is signifi-cantly less expensive than a network rollout. Many studieshave highlighted the use of Wi-Fi-based cellular data off-loading, which can be roughly classified into the followingapproaches: (a) bypass mode, (b) managed approach, and(c) integral mode. In the bypass mode, the user’s device


makes a transparent transition from a core cellular net-work whenever a Wi-Fi network is available, therebyallowing minimal use of cellular services. The ease ofdeploying this approach makes it very attractive, but thecellular network loses control of the cellular device in thissetting because Wi-Fi is not an integral part of the cellularnetwork, and this makes it impossible to deliver the sub-scribed content. To overcome this problem, the cellularoperator can deploy a session-based mechanism, whichfacilitates a managed approach [10]. The managedapproach uses intelligent gateway-based session manage-ment, which also allows the cellular provider to keep trackof its users. This does not enforce the complete integrationof Wi-Fi and cellular networks. When the cellular networkand Wi-Fi networks are fully integrated with each other,this approach is called the integral mode [28]. In this set-ting, the Wi-Fi network is connected indirectly to the cellu-lar core network through an external IP network, such asthe Internet, and service connectivity is provided by roam-ing between the two networks [29,4].

However, these approaches require major modificationsto state-of-the-art hardware and software technologies,thereby leading to huge infrastructure changes and highadoption costs. In addition, adopting these technologieswould not address the issue of the heterogeneous applica-tion data transferred over cellular networks and they donot provide any incentives to users, application developers,etc. By contrast, we propose a multi-incentive solution,which ensures high mobile application data availability, abetter service for users who are incapable of accessingthe Internet, and, above all, a monetary return for userswho overpay for their carrier-enforced data plans.

2.2. Delay-tolerant networks

In recent years, delay-tolerant networks (DTNs) havebeen the subject of extensive research efforts, such asspace communication, networking in sparsely populatedareas [30], and vehicular ad hoc networks [31]. Unlike tra-ditional tethered networks like the Internet, a DTN is asparse mobile network where the connections betweenthe nodes in the network change over time, which meansthat communications constantly experience high delaysand disconnections [32]. Because an end-to-end pathmight never exist in DTNs, effective communication inDTNs requires the cooperation of all the nodes during rout-ing and forwarding, where the intermediate nodes on acommunication path are expected to store, carry, and for-ward the packets in an opportunistic manner, which is alsoknown as opportunistic data forwarding. However, DTNscomprise numerous resource-constrained nodes in mostcases, i.e., limited storage. Therefore, if they are carriedfor a certain amount of time without an available down-stream node, the packets must be dropped by the carryingnode, which leads to highly unreliable forwarding in DTNs.Therefore, efficient packet forwarding in DTNs is an espe-cially challenging issue, but a number of DTN packet for-warding schemes have been proposed recently toimprove the reliability [33].

Several concepts related to opportunistic networks arederived from studies of DTNs by the Internet Research Task

Force, which have led to the specification of a DTN architec-ture [34–36]. The DTN architecture comprises a network ofindependent Internets, each of which is characterized byinternal Internet-like connectivity, but there are only occa-sional communication opportunities among them, whichare sometimes scheduled in time, whereas others arecompletely random. The independent Internets locatedapart from each other constitute so-called DTN regionsand a system of DTN gateways is in charge of providinginterconnections among them. Therefore, in DTNs, thepoints of possible disconnections are known and isolatedat gateways. Each Internet relies on its own protocol stack,which is most suitable for the particular infrastructure,communication means, and technologies available in thatparticular Internet’s region. The protocols used in the dif-ferent DTN regions are likely to differ from each other.However, at the DTN nodes, a novel overlay protocol isadded on top of the traditional transport layers to manageend-to-end data transfers among the DTN regions.

2.3. Familiar strangers

According to Milgram (1972), ‘‘familiar strangers’’ [37]are individuals who regularly observe each other and exhi-bit common patterns during their daily activities [38], suchas ‘‘taking the train or subway,’’ ‘‘having lunch in cafeteria,’’or ‘‘attending a seminar,’’ but they do not interact with eachother [39,40,38]. Understanding the mechanisms thatdrive daily face-to-face encounters is still limited becausethe field lacks large-scale datasets that describe individualbehaviors and their collective interactions. Access to largedatasets of human activities and interactions has been lim-ited by the difficulty and cost of collecting such informa-tion. However, the increasing availability of digital tracesof human actions is now enabling the widespread repre-sentation and the analysis of massive volumes of informa-tion on human behavior. Thus, by analyzing variouspreferences and constraints on individual behavior, andspatio-temporal and collective patterns, [41] identifiedthe regularity of daily life activities, such as morning/even-ing peaks in transportation, the degree of crowding inshopping malls and supermarkets at the weekends, andin restaurants during dining periods, etc. Therefore, wecould exploit the spatio-temporal co-existence of thesefamiliar strangers to promote covert data sharing betweentheir devices. Fig. 2 shows the nodes arranged in a familiarstranger network setup and the interactions between thenodes in the proposed systems.

2.4. Auctions

In economic theory, an auction may also refer to anymechanism or set of trading rules used for commodityexchange between participants. The modus operandi ofthe auction is used as the basis for classifying the auctionmechanism. In a typical ‘‘first price, sealed bid auction,’’bidders submit simultaneous ‘‘sealed bids’’ to the seller.This terminology is derived from the original format ofthese auctions, where bids were written down and pro-vided in sealed envelopes to the seller, who would thenopen them all together. The highest bidder wins the object

Fig. 2. Data offloading architecture in a familiar stranger network.


and pays the value that he bid [42]. We use a reversedsealed bid auction where sellers submit their respectivebids to a potential buyer and the buyer selects the sellerby evaluating the minimum of all the bids received [43].A successful auction system must be financially efficient[44], but it should also provide an efficient bidding processand allow rapid execution. The bids express the prefer-ences of users for various outcomes.

First price, sealed bid auction mechanisms are oftenused to sell services. The use of an auction mechanismhas several advantages compared with a fixed price. Fromthe buyer’s point of view, using a fixed price would providethe advantage of removing uncertainty about the price ofthe service, but a fixed price would not incentivize the ade-quate use of scarce resources. By contrast, with an auctionmechanism, there is uncertainty about the price the buyeris going to pay, but the buyer never pays a price higherthan what they are willing to pay for the service, i.e., buy-ing the service from the cellular network provider. More-over, our reimbursement proposal reassures the sellerbecause if it provides the service, the money paid for theservice will be given back to them.

In any auction, the following properties are desirable.(a) Auction Correctness: Assuming that all bidders act hon-estly, the correct winner is identified by the auction process.(b) Bid Confidentiality: The bid amounts are not revealed toany bidders. (c) Auction Fairness: After submission, the bidscannot be changed or repudiated. Meeting these require-ments to internalize the externalities of the pricing of directand external costs of resource usage in grid, cloud, sharednetwork resources, and services has been the focus ofresearchers for a couple of decades [29,45,46]. In previousstudies, examples can be found where auctions have beenused to allocate computation and communication resourcesin the grid or G-Commerce [47,48]. For example, a previousstudy [49] proposed a market-based approach for resourceallocation and utilization in the cloud. Using these types ofmodels ensures the efficient usage of shared resources byconsumers. Auctions are also used widely in networkresource sharing, particularly in congested networks[50–53], where resources are identified as those that can be

used by more than one person, but increasing usage degradestheir quality or only a limited number of users have access tothe resources [54]. In these networks, it is compulsory to offerincentives to selfish nodes to forward the traffic from theirpeers to obtain better performance [54]. The use of a sec-ond-price, auction-based bandwidth pricing mechanismwas proposed in [55] to address congestion-related problemsin wireless networks. In addition, [56] extended thisapproach to the design of a pricing mechanism for thedownlink transmission power in Code division multipleaccess (CDMA)-based wireless networks. Furthermore, byfocusing on a second-price auction-based structure, [57]targeted the problem of decentralized cooperative partnerselection during the uplink of cellular or ad hoc networks.CollabAssure targeted the pricing of network resources andfocused on achieving two different goals, i.e., obtaining themaximum revenue for the network and managing theefficient allocation of resources. Hence, targeting overallnetwork revenue maximization and the utilization of net-work resources are the main goals of this approach, whereadditional monetary incentives are given to the user.

2.5. Android architecture

The Android [21] mobile software framework is sup-ported by Google, and it comprises an OS, system utilities,and middleware in the form of a virtual machine. TheAndroid platform includes about 130 application level per-missions that allow applications to use different accesscontrols and hardware resources. These permissions areset by application developers and they are requested dur-ing installation, and they are then silently enforced whenthe application is executed. Most of the Internet-basedapplications that run on mobile devices use a client–servermodel for data delivery, which involves high networkusage and traffic rates [58]. SmartParcel offers a new setof permissions that allow applications to use alternativedata sources in their social proximity, thereby ensuringlower service costs. We modified the Android permissionmodel to include a set of five new permissions using differ-ent combinations of available network technologies. The

Table 1Device resources used with different permissions.

Group name BT WiFi NFC Disk-IO

SMP_ALLp p p p

SMP_BTp � � p

SMP_WIFI � p � p

SMP_NFC � � p p

SMP_BT_WIFIp p � p

CLB_ALLp p � p

CLB_BTp � � p

CLB_WIFI � p � p


names and resources used in each permission are shown inTable 1. We integrate SmartParcel and CollabAssure servicesin the Android framework in the ‘‘System Server’’ module,which is launched by Zygote,1 as shown in Figs. 7 and 22.While performing the boot operation, Zygote forks these ser-vices as a system service, which ensures that the SmartParceland CollabAssure services acquire system level privileges andthey are independent of the application context.

2.6. Dataset

In this study, we used the MIT Reality Mining dataset[14], which contains over 500,000 hours of data collectedduring 9 months by 100 unique devices. The datasetincludes information about call logs, Bluetooth devices inclose proximity, cell tower IDs, application usage, andphone status (such as charging and idle). We focused onlyon the Bluetooth interactions between users in the dataset.Fig. 3 shows clearly that the user activity level is higherduring the day (8:00 am to 8:00 pm) and lower duringthe night and the early morning hours.

The distribution of the number of devices encounteredper scan is shown in Fig. 4. Thus, a minimum of two anda maximum of 65 devices were recorded per encounter,with an average of four devices per encounter and a stan-dard deviation of 8.67. Fig. 5 shows the frequency ofdevices active per day over the trace period. The deviceactivity level in the dataset comprised a minimum of fourand a maximum of 901 active devices per day, where themean and standard deviation were 243 and 133,respectively.

3. SmartParcel: harnessing public anxiety and play

3.1. SmartParcel service overview

In this section, we describe the architecture and theimplementation of an OS-independent framework for adata sharing service, ‘‘SmartParcel’’, which is used for col-laborative sharing among mobile devices. SmartParcelrelies on the concepts of the selective and opportunisticdata delivery model. The SmartParcel service frameworkallows different applications to cache data in the devicememory, which can then be shared opportunistically with

1 Zygote: This is the parent of all the Dalvik VMs in the system and itoften loads the classes used by applications into its heap.

other devices in the spatio-temporal proximity. We arguethat these spatio temporal co-existing users, who are iden-tified as ‘‘familiar strangers’’ [37], can play important rolesin the viral dissemination of data [38]. We implementedSmartParcel for Android but its modular architectureensures that it can also be extended to other mobile soft-ware platforms. We modified the Android SDK to offernew permissions to include SmartParcel and to offer serviceAPIs that will encourage application developers to useSmartParcel. These permissions are highlighted in Table 1.

SmartParcel is a ‘‘one-for-all’’ multiple incentive systemfor application developers, ISPs, and application data pro-viders (e.g., cloud services), which provides additional ben-efits for consumers. It offers application developers andconsumers better data availability and reduced cellulardata usage, especially for devices that run on limited dataplans, or when access points are unavailable. In terms ofenergy consumption, its reliance on cheaper network tech-nologies (Bluetooth, Wi-Fi) compared with 3G and LTE ser-vices makes SmartParcel more adaptable for consumers[13]. Excluding consumer incentives, lower network trafficand higher service availability can be achieved at muchlower costs with fewer data requests to applicationservers.

A typical user case for SmartParcel is ‘‘reading a newspa-per on the subway.’’ The phrase ‘‘Can I borrow your newspa-per?’’ is heard often on the subway. Thus, the concept ofborrowing can still be exploited even in digital age wherenewspapers have been replaced by newspaper applica-tions. When a SmartParcel node (e.g., a smartphone withan active data plan) Alice accesses a newspaper application,it registers the SmartParcel service on Alice’s device. Assum-ing that there are no SmartParcel nodes in close proximity,the data are requested from the server and cached bySmartParcel, which periodically sends a broadcast messageover different network interfaces. When another SmartPar-cel node, Bob, which is using a portable media player (withno active data plan) comes into Alice’s proximity, a one-to-one connection is set up to receive newspaper data fromAlice after receiving the broadcast message. Thus, bothAlice and Bob can serve as hosts for any other users thatmay arrive on the subway at a later time. The SmartParcelnodes broadcast the latest timestamps of the data cachedfor all applications that are registered by the service. ASmartParcel node can request and serve data for multipleapplications, which are selected from the broadcast mes-sage as a single chunk.

3.2. SmartParcel architecture

Fig. 6 shows the proposed architecture of SmartParcel,which has six different components, as discussed below.

3.2.1. Service discovery managerThis unit identifies the candidates available for data

transfer by broadcasting a ‘‘SYN’’ message periodically toneighboring devices. The ‘‘SYN’’ packet contains metadataabout the applications registered by SmartParcel. Themetadata are organized as a key value pair, i.e., (‘‘Applica-tionId, TimeStamp’’). On the receiver end, a device runningthe SmartParcel service receives multiple SYN packets from

Fig. 3. Hourly variation in device encounters.

Fig. 4. Distribution of device encounters.

Fig. 5. Distribution of active devices per day.


Fig. 6. SmartParcel service architecture.


neighboring devices and based on the metadata informa-tion it sets up a one-to-one connection to allow data trans-fer with the device that has the latest data.

3.2.2. Data transfer managerAfter a connection has been established by the Service

Discovery Manager, this unit starts the data transfer. Thisunit is designed to manage concurrent connections withmultiple devices. To reduce the network overhead, wedesigned SmartParcel’s data transfer manager to send datafor multiple applications as a single chunk, i.e., we usedevice-to-device data sharing rather than application-to-application data sharing.

3.2.3. Service cache managerTo preserve the heterogeneity of the application-

specific data, we use a service cache to store the applica-tion-specific data. To improve performance, two types ofcache are implemented, as follows.

� Dynamic Cache: An in-memory cache for storing theapplication’s metadata information, which is storedinside the device RAM. It is implemented as a HashMap with (Application Id, Timestamp) as the key-valuepairs.� Static Cache: A static cache for storing the actual appli-

cation-specific data, which is maintained as an SQLitedatabase. We use the schema for the database (‘‘Applica-tion Id (as string), Data (as blob), Time Stamp’’) whereApplicationId and TimeStamp are used together as a pri-mary key. By default, we empty this cache at the end ofday. We also provide flexibility so the developer canassign ‘‘Time-to-live’’ and ‘‘Reset-Time’’ to the applica-tion data.

3.2.4. Network interface managerThis internal service is responsible for managing the

network connections and it helps the Service DiscoveryManager to identify available devices on different networkinterfaces that are supported by the device (e.g., 3G, LTE,Wi-Fi, and BlueTooth).

3.2.5. Service APIsA set of service APIs is also implemented to allow appli-

cation developers to interact with the SmartParcel service.These mainly comprise APIs to subscribe or unsubscribeto services, update application data, settings, share statis-tics, etc.

3.2.6. Central control managerAs suggested by its name, the job of the Control Man-

ager is to manage the control of all the components ofthe SmartParcel service. All of the components of theSmartParcel service work under the same instance of theCentral Control Manager to ensure synchronous operation.

3.3. Data security and privacy

SmartParcel only allows application-based data cachingand sharing. The system offers complete flexibility forapplication developers to enforce data encryption at the

application level. In particular, SmartParcel targets applica-tions that are not utilizing user-specific and thus sensitivedata, i.e., public data applications such as news, blogs, andhealth updates, thereby avoiding inherent privacy issues.

3.4. SmartParcel analysis

Using the MIT reality mining data set, we formulated aset of experiments to analyze the performance of SmartPar-cel using different parameters and assumptions, as follows.

� Data refresh rate (DRR): The frequency with which thedata are refreshed.� Allowed server connections (ASC): The number of

devices allowed to obtain data from the server eachday. This parameter can be interpreted in two differentways: (a) The maximum number of concurrent connec-tions that a data server can serve, or (b) The maximumnumber of nodes in the familiar stranger network withInternet access.� User participation probability (UPP): The probability of

a user not acting selfishly, i.e., not limiting their partic-ipation by only receiving data and not participating indata dissemination.

We calculated the data availability ratio (DAR) as theratio of data availability for the DRR interval, i.e., at theend of each interval,

DAR ¼ No: of Devices with DataTotal No: of Active Devices

:

First, we investigated the effect of a user’s social activitylevel on data dissemination during the day. It is importantto note that the results are presented as the 5th, 50th, and95th percentiles of the observed results. With UPP as 100%,i.e., every user participated, we set DRR to one refresh perday and considered the case when ASC = 1 with 30 devicesper day. We observed a steep rise in DAR during the peakhours of the day, i.e., 8:00 am to 8:00 pm, as shown inFig. 8.

With only one active device, DAR approached about 40%by the end of the day, but when ASC = 30, it approachedabout 95%, as shown in Fig. 9. Figs. 8 and 9 show clearly

Fig. 7. SmartParcel service design and placement within the androidframework.


that the data availability was affected greatly by the num-ber of devices that were allowed to connect and receivedata from the server. To investigate this further, we variedASC and analyzed the data availability at the end of eachday. The results shown in Fig. 10 demonstrate that therewere large increases in data availability when one, 20,and 50 devices were allowed to connect and receive fromthe server, i.e., 43%, 85%, and 98%, respectively, and itapproached 100% when 65 devices were allowed.

Given that application data are refreshed periodically,we investigated two scenarios that allowed two and threerefresh intervals per day. Thus, we divided the day intotwo and three intervals of 12 hours and 8 hours, respec-tively. We did not consider user participation, which wasset to 100%. The results shown in Figs. 11 and 12 andFigs. 13–15 demonstrate the direct dependency of data dis-semination on the human social activity level. A lower DARwas expected due to the lower human activity levels duringthe night. The results shown in Fig. 11 (12:00 am to 11:59am) and Fig. 13 (12:00 am to 07:59 am) agree with thisexpectation. Data availability levels of 35% and 12% wereobserved during the intervals of 12:00 am to 11:59 amand 12:00 am to 07:59 am, respectively. The level increasedduring the day, thus high DAR values were observed, i.e.,85% for the interval of 12:00 pm to 11:59 pm and 60% forthe intervals of 08:00 am to 3:59 pm and 04:00 pm to11:59 pm.

Thus, we showed that human social activity levels playan important role in data dissemination. Next, we used theUPP to consider selfishness and the unwillingness of usersto participate in sharing. Fig. 16 shows clearly that DARincreased with ASC and UPP. The data points in Fig. 16 rep-resent the median of DAR at the end of each day, aggre-gated over 1000 simulations runs. We can see that for afixed ASC, DAR increased with UPP. Specifically, whenASC = 10, we obtained DAR values of 65%, 72%, 80%, and84% with UPP levels of 10%, 20%, 50%, and 90%, respec-tively. When UPP = 90%, the data availability was 98%when only 40 devices were allowed (ASC) to receive datafrom the server. With lower values of UPP, i.e., 10%, 20%,and 50%, DAR approached 70%, 85%, and 93%, respectively.

4. CollabAssure: incentive-based data sharing

In the previous section, we introduced the design andimplementation of a collaborative data sharing system,‘‘SmartParcel,’’ which addresses the issue of cellular data off-loading and service assurance by harnessing a spatio-tem-poral network of coexisting mobile devices, i.e., ‘‘familiarstrangers’’ [11]. We argue that by providing a mobile OSapproach, SmartParcel can easily be integrated into currentmobile platforms. Although it is promising, SmartParcel onits own would be vulnerable to the ‘‘tragedy of the commons’’problem [59], i.e., in a typical scenario, all of the users wouldwait for a generous user to share data. The major reason whya user would be unwilling to participate is the lack of anincentive to provide data to other nodes in the proximity.

In this section we propose, CollabAssure, which is anauction-based, ad hoc market model that ensures a serviceis provided to users with no Internet access capability. The

CollabAssure framework provides service assurance viaopportunistic ad hoc networks formed by spatio-temporalco-existing mobile users. The system allows users to ‘‘sub-let’’ their surplus data plans to users without Internetaccess. If a user requests a service in an opportunistic net-work, CollabAssure allows users to participate in an auction,where the nodes that can offer services submit their bids tothe nodes that request services, thereby resulting in theformation of an ad hoc market where buyers and sellersrequest and offer services, respectively.

The system comprises mobile devices such as smart-phones and tablets. The current policies of the cellular ser-vice providers enforce data plan requirements forsmartphones in the USA, whereas they are not mandatoryfor tablet users. Thus, tablet users are often rendered inca-pable of using Internet services when Wi-Fi hot-spots areunavailable. In addition, smartphone users experience theunavailability of Internet services during service outagesor because of the exhaustion of their carrier-enforcedmonthly data limits. Therefore, these polices force certaingroups of users to increase their monthly data limits atan extra cost. These users are not able to exhaust the com-plete allocated share of the data services, which are typi-cally renewed by the end of the billing cycle. Eventually,these users pay for services that they do not use or use onlypartially. In the CollabAssure service, the typical settingsallow users to ‘‘sublet’’ their data plans to other userswho require data services. This revenue model provides a‘‘one-for-all’’ solution to all the users where the users withabundant data services are compensated for their overpay-ment and those in need of data services are provide withservices. The CollabAssure system inherently increases theavailability of the services offered by various Internet-based mobile applications when users are deprived dueto the unavailability of access points. All of these factorsare benefits of adopting CollabAssure for mobile applicationdevelopers, OS vendors, and users.

Fig. 17 shows an overview of the operation of CollabAs-sure, and the straw man of CollabAssure is illustrated inFig. 18. In a typical user case scenario for CollabAssure atperiod 0, Alice and Carol enable the CollabAssure mode

Fig. 8. Allowed server connections (ASC) = 1.

Fig. 9. Allowed server connections (ASC) = 30.

Fig. 10. Effect of allowed server connections (ASC) on the data availability ratio (DAR).


on their devices and set a data threshold for their ownuse. We assume that Alice and Carol both subscribe to a2 GB monthly data plan, where 1 GB is reserved for theirown use and the remaining 1 GB can be sublet to otherusers. Alice and Carol also have total freedom to modify

these limits during their complete monthly cycle. Inaddition, suppose that another user, Bob, owns aWi-Fi-enabled tablet device and Bob has enabled theCollabAssure service on his device. On a typical subway,Alice and Carol can both enjoy their data services whereas

Fig. 11. Variation in the data availability ratio (DAR) with the data refresh rate (DRR) when DRR = 2 and refresh interval = 12:00 am to 11:59 am.

Fig. 12. Variation in the data availability ratio (DAR) with the data refresh rate (DRR) when DRR = 2 and refresh interval = 12:00 pm to 11:59 pm.

Fig. 13. Variation in the data availability ratio (DAR) with the data refresh rate (DRR) when DRR = 2 and refresh interval = 12:00 am to 07:59 am.


Bob cannot due to the unavailability of Wi-Fi accesspoints. This setup is highlighted in Fig. 18. In this setting,Bob can access the Internet by agreeing to pay either Aliceor Carol for procuring their Internet services. In period 1,Bob requests a service from nodes within its proximity.When the CollabAssure services on Alice’s and Carol’s

devices receive this request, each generates a bid that iscommunicated to Bob’s device. Next, Bob selects the userwith the lowest bid and requests data in period 2. Weassume that Alice submitted the lowest bid and the tradewas set up between Alice’s and Bob’s devices. After thecontract is set, Bob and Alice have Internet access, thus

Fig. 14. Variation in the data availability ratio (DAR) with the data refresh rate (DRR) when DRR = 2 and refresh interval = 08:00 am to 03:59 pm.

Fig. 15. Variation in the data availability ratio (DAR) with the data refresh rate (DRR) when DRR = 2 and refresh interval = 04:00 pm to 11:59 pm.

Fig. 16. Variation in the data availability ratio (DAR) with the user participation probability (UPP) and allowed server connections (ASC).


the benefits to both users are ensured. Because Carol’s bidwas higher, she does not participate in the transactionand will not earn any utility from this trade. In period3, after the completion of the service, both the seller

(Alice) and the buyer (Bob) report their trade agreementto their respective vendors, [60] and [61]. Bob is chargedfor procuring the services on a monthly basis and Aliceis compensated for providing the data services.

Fig. 17. Overview of the operation of CollabAssure.

Fig. 19. Bipartite graph of sellers and buyers in a full mesh topology.


4.1. Ad Hoc market approach

This section describes the auction-based ad hoc marketapproach, where all users participate continuously to buy

Fig. 18. CollabAssure: straw man an

or sell mobile data services. We use an auction-based strat-egy, which is similar to that described in [62]. At anyinstance of time, for a given set of N users arranged in a‘‘familiar stranger’’ network setup, we assume that everyuser strives to maximize their own utility and that theycan communicate with every other user in a full-meshtopology, as shown in Fig. 19. In the remainder of thisstudy, we use the following terminology.

d typical user case scenario.

��

��

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

�

��

�

��

��

�

��

��

�

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

�

��

�

��

�

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

�

��

�

��

�

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

�

��

�

��

�

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

�

��

�

��

�

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

�

��

�

��

�

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

�

��

�

��

�

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

��

��

��

��

��

�

��

�

��

��

�

��

��

�

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

�

��

��

�

��

��

��

��

��

��

�

��

��

�

��

��

��

��

Fig. 20. Variation in the satisfaction function with usage.

Fig. 21. CollabAssure service architecture.


� Total data count ðDiÞ: Total number of data units avail-able for auction by seller i.� Bid ðbiÞ: The bidding valuation expressed by seller i for

each unit of data.� Satisfaction count ðxijÞ: Amount of data units that buyer i

consumes to download content from a seller j.� Grant count ðyijÞ: Amount of data units granted from

seller i to buyer j, i.e., the amount that user i requiresto satisfy user j’s needs.� Usage vector ðxiÞ : xi ¼ xijf8j 2 ½1;N�g� Served vector ðyiÞ : yi ¼ yijf8j 2 ½1;N�g

According to the definitions of the Satisfaction count andGrant count, buyer i can satisfy his needs through seller jonly if seller j grants the corresponding amount of dataunits, i.e., yji ¼ xij. Now, we define the usage and servedmatrices as X ¼ fxi : ð8iji 2 ½1;N�Þg and Y ¼ fyi : ð8iji 2½1;N�Þg, respectively.

In the current scenario, we assume that a mobile datauser Si offers data services at a marginal cost bi per unitof data. Suppose that another user Bj exists with no avail-able Internet connection (or one paying a much higher ser-vice cost) and that they request a service. We allow user Si

to ‘‘sell/sublet’’ their data plan to any other user with lim-ited data connectivity but they are unaware of the utility ofthe user Bj. We also assume that a single demand from userBj is drawn from a continuous distribution H, with densityh and a finite mean. We reiterate that CollabAssure targetsthe maximum utilization of the user’s data service plan.

In the event of lost connectivity, an unmet demand is lost,thereby resulting in the margin being lost (to the seller),but there is no additional penalty to the buyer. In this case,the CollabAssure framework treats this as ‘‘no-trade’’ andthe buyer will not be charged for the trade. This loss isincurred by the seller. In a general commodity market, thismight not be a fair approach, but it can be treated as a costof operation in the cellular data transaction model, such asCollabAssure, where the transactions are usually very small.

Each user i sets a threshold on their data limit for per-sonal use, which we call the total monthly data limit Ti,and ti

o is the threshold for the users own data needs. Hence,the total data available for auction by user i, Di, is evaluatedas Di ¼ Ti � ti

o. Assuming that ti are the data used by user iat any point in time, Fig. 20 shows the behavior of the util-ity functions based on their usage over the month.Fig. 20(a) shows the scenario when the usage ti > ti

o andwhen ti 6 ti

o in Fig. 20(b), where both ti and tio 6 Ti. Hence,

for any user i, the total amount of data used in auctionscannot exceed the monthly threshold, as follows.

XN

j¼1

yij 6 Di ð1Þ

In addition, the total amount of data used personally byuser i and that used in auction cannot exceed the monthlydata limit, which is represented as follows.

XN

j¼1

yij þ tio 6 Ti ð2Þ

To design an effective bidding strategy, it is intuitive toexpect a higher bid when a user has very little data left andthey are participating in an auction during the daytime. Bycontrast, a lower bid is expected during an early morningtrade when the data availability is relatively high. Tomodel this behavior, we set the bid bi for each transactionas a function of the available data limit ðdiÞ and currenttime of day ðkÞ, as follows.

bi ¼ bðk; diÞ ð3Þ

For CollabAssure, we select a simple function, bðk; diÞ¼

ffiffiffiffiffiffiffiffiffiffiffiffik� dip

jk; di 2 ½0;1�. We could select a more complexbid generation function, but we prefer to keep this verysimple in this case. Any transaction would consume theuser’s battery power, which is required during the day,

Fig. 22. Integration of CollabAssure in the android framework


thus we select higher values of k during different hours inthe day, and lower values in the early morning and latenight. In addition, the amount of data left for the userdecreases with successful trades, thus we use di (ratio ofdata available) as a factor during bid generation. For anyseller i, di is computed as di = Di / Ti.

A trade ckðxij; bi, occurs when user i consumes xij dataunits to download content from user j at bi unit price.We also define C ¼ fck : ð8kjk 2 ½1;N�Þg as the set of alltrades, where N is the set of all natural numbers. We con-sider a general case where user i obtains different utilityvalues for each trade with user j. For any users i and j inthe market, the utility can be characterized as follows.

� A constant utility is earned by a user who utilizes theirown resources. Let nijðxijÞ be the utility earned by user iwhen they use their own data.2

2 Theconsiste

nijðxijÞ ¼0 when i – jc when i ¼ j

�ð4Þ

� The utility earned by using the resources of a nearbyuser. Let !ijðxijÞ be the utility of user i satisfying the dataneeds of user j. A user is charged for the data units theydo not use, hence the function !ið�Þ is a monotonicallyincreasing utility function, which is associated witheach trade.� The utility earned by subletting resources to another

user. Let WijðyijÞ be the utility incurred by the user afterproviding resources to another user j.

Hence, when the conditions in (1) and (2) are met, thecharacterized utility function for any user i can be writtenas follows.

Piðxi; yjÞ ¼XN

j¼1

ð!ijðxijÞ þWijðyijÞ þ nijðxijÞÞ ð5Þ

We use a very simple model to characterize this ad hocmarket, where the perceived utility of Si can be character-ized by the function Pið�Þ. The proposed model incurs ahigher utility and revenue for any successful trade, ratherthan no trade.

4.2. Architecture

DTNs [30] target the interoperability between andamong challenged networks to deliver data units from asender to a receiver in the presence of opportunistic con-nectivity using different transport protocols [63]. CollabAs-sure harnesses the concepts of node discovery andopportunistic data delivery from the DTN architecture toallow mobile devices to participate in ad hoc markets forcollaborative data sharing. We propose the CollabAssuredesign for Android, but its modular design could easily beintegrated into other available mobile OSs. Various compo-nents of the CollabAssure framework are shown in Fig. 21and its integration in the Android framework is shown inFig. 22. This framework has the following components.

behavior of nij is independent of j. We use this notation to maintainncy with other utility functions.

1. Network interface manager: We classify different net-work technologies as: (a) WAN facing or high cost inter-faces (3G, LTE, GPRS, etc.), and (b) ad hoc facing or lowcost (WiFi and Bluetooth). Android devices offer multiplenetwork technologies, which allow them to manage allof the connections used by the Network Interface Man-ager. The Central Control Manager is implemented as aninternal service, which is responsible for managing thenetwork connections on all interfaces.

2. Auction manager: The Auction Manager determinesthe ideal bidding price per unit data at a given instancein time when requested by the Central Control Man-ager. A large number of factors3 govern the cost of thedata download operation, e.g., the current battery level,available data limit, days until the next billing cycle,and the time of the day. In this study, we use the currentdata limit and the time of the day to determine the cur-rent bid.

3. Cache: The Central Control Manager maintains a datacache to store the data associated with each data down-load. This cache is maintained as an SQLite database,which is available in the Android framework. We usethe following schema: <AppID, TimeStamp, Data>. Thiscache helps the user to maximize their utility by usingcached data from multiple devices (buyers) at the sametime.

4. Central control manager: This component manages thecontrol flow for all the other components of the Collab-Assure service. All of the other components are operatedat the same instance by the Central Control Manager toensure synchronous operation. The actions performedare: (i) using the Network Interface Manager to partic-ipate in the ad hoc market; (ii) coordinating with theAuction Manager during bid generation; (iii) in eventof a trade, triggering data downloading and updatingthe cache; and (iv) after completion, reporting the tradeand cost to a central entity.

5. Download manager: The primary task of this unit is toinitiate a data download when requested by the CentralControl Manager. The Central Control Manager calls this

3 Modeling all of these factors and the user usage behavior is anotherresearch question. Any combination of parameters could be used to decidethe optimal bidding strategies.


internal service with the Application ID and this unittriggers the data download via the application.4 Thesedata are then returned to the Central Control Manager.

6. Security manager: To ensure bid confidentiality, Collab-Assure employs a public key encryption scheme for eachdevice pair involved in a trade. Because each device isregistered with its respective vendor’s market, the pub-lic keys of the participating devices are stored in thevendor’s database, which can be used for authenticationand later in the bidding process. When a seller sends itsbid to a buyer, it is encrypted with the buyer’s public keyand the encrypted bid is sent to the buyer. The buyerthen decrypts it with its own private key and obtainsthe bid. Public key algorithms are known to be compu-tationally expensive, so after the buyer identifies apotential seller, a much faster symmetric algorithm isused to encrypt and decrypt subsequent messagesexchanged between the buyer and seller. Specifically,the buyer generates a shared secret key (session key),encrypts it with the seller’s public key, and then sendsthe encrypted session key to the seller. The seller thenuses its private key to decrypt the session key.

7. Pricing and reporting unit: Pricing is a difficult task indecentralized ad hoc networks, thus we overcome thisby using a two-way reporting strategy where both ofthe devices that participate in the trade report to a cen-tral entity. Every device maintains a static log of thetransactions, which may be reported later. This reducesthe additional reporting overheads for the buyers andcompensates for the unavailability of the Internet tothe buyer devices. Both the buyer and the seller mustreport the same trade to obtain a successful payment.The central entity could be a third party server but, toavoid privacy issues, we propose an e-commerce modelwhere each device reports to the OS vendors, i.e., digitalmarkets [60,61]. Mobile OS vendors uniquely identifyeach device, which makes this approach easy to inte-grate and maintain.

4.3. Android and CollabAssure

To integrate CollabAssure in the Android framework andfor use in Android application development, we modifiedthe Android permission model to include the permissionnamed COLLABASSURE, which uses the underlying net-work technologies, i.e., Bluetooth, Wi-Fi, and NFC, for adhoc network connections. To use this service, Androidapplication developers could use the permissions byincluding the tag ‘‘android.permission.COLLABAS-SURE’’ in the AndroidManifest.xml. The service is integratedin the ‘‘System Server’’ module, which is launched byZygote, as shown in Fig. 22. Zygote is a process that startsat boot time and it is the parent of all the Dalvik VMs in thesystem. Zygote loads and initializes the classes in its heapthat are expected to be used very often by applications.During the boot operation, Zygote forks the CollabAssureservice as a system service. This ensures that the CollabAs-sure service obtains the system level privileges and it

4 For details, see section: Android and CollabAssure.

remains independent of the application context. We mod-ified the Android SDK to provide a wrapper around theAndroid activity as CollabActivity, which provides a rawinterface so Android application developers can includean additional static update function. This allows the Collab-Assure service to invoke the update method in the back-ground for each registered application as and whenrequired. We also encourage application developers touse CollabAssure APIs to update the cache when a user uti-lizes the application, so the cache is always up-to-date.

4.4. Data security and privacy

Because CollabAssure only allows application-baseddata sharing, the system offers complete flexibility forapplication developers to enforce data encryption at theapplication level. Typically, the CollabAssure system targetsapplications that do not deal with user-sensitive data, suchas banking, emails, and social networks. Therefore, devel-opers of these applications should not use the CollabAssureservice, whereas news, blogs, RSS feeds, etc. will benefitfrom the CollabAssure approach. Application developerscan also deploy two-phase data labeling, i.e., public andprivate. Public data can be made available for sharingthrough CollabAssure, whereas private and user-specificdata are not. For example, a banking application shouldmark the nearby ATM locations as public data, which canbe shared through CollabAssure, whereas the remaindershould be labeled as private data that cannot be shared.

4.5. CollabAssure analysis

We used the MIT Reality Mining dataset discussed inSection 2.6 to perform a trace-based evaluation ofCollabAssure.

4.5.1. Effect of the data refresh rateFirst, we analyzed the effect of DRR based on the number

of trades during the day. Figs. 3–5 show that the user activitylevel was higher between 8:00 am and 8:00 pm. There was asimilar trend in the number of trades. Figs. 23(a)–(c) high-light the number of successful trades at different times dur-ing the day when DRR was 1 hour, 4 hours, and 12 hours,respectively. Fig. 23(a) and (c) show that there were similartrends in the numbers of trades when DRR = 1 hour and12 hours. In both cases, when DRR = 1 hour and 12 hours,the maximum number of trades was 800 and the medianof number of trades ranged between 0 and 80 trades. WhenDRR = 4 hours, the maximum number of trades was 1000and the median varied from 0 to 150 trades per hour, asshown in Fig. 23(b). In our next experiment, we focused onthe effects of the user participation probabilities on theaverage earned utility. The market comprised buyers andsellers so we analyze the effects of their participationindividually.

4.5.2. Effect of the Buyer Participation Probability (BPP)We fixed the seller participation probability (SPP) at

100% and measured the variation in the ratio of successfultrades relative to the total possible trades. We alsoanalyzed the volume of data sold in each transaction on

0

200

400

600

800

Hour of Day

Tot

al N

umbe

r of

Suc

cess

ful T

rade

s

1 3 5 7 9 11 13 15 17 19 21 23

0

200

400

600

800

1000

Hour of Day

Tot

al N

umbe

r of

Suc

cess

ful T

rade

s

0

200

400

600

800

Hour of Day

Tot

al N

umbe

r of

Suc

cess

ful T

rade

s

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

hours

hours

hour

Fig. 23. Variation in the number of trades with the data refresh rate(DRR).


an hourly basis when DRR was 1 hour and 12 hours. Theresults obtained when BPP was 10%, 30%, and 70% areshown in Fig. 24. We found that the social activity levelsof mobile users had the expected effect on the ratio of suc-cessful trades. A higher success ratio was obtained duringworking hours, i.e., 9:00 am to 9:00 pm, whereas a low

success ratio was measured during the early morningand late night. Only a small median shift, i.e., 0.1–0.2,occurred when DRR was 1 hour or 12 hours, whereas theBPP had a significant effect on the number of successfultrades compared with the effect of the DRR, as shown inFig. 24. When we fixed DRR to 1 hour, there was a signifi-cant increase in the number of trades when BPP increasedfrom 10% to 30% and 70%. A similar trend was observedwhen we fixed DRR at 12 hours, as shown in Fig. 24. Inthe hours with the most activity, i.e., 10:00 am to 4:00pm, the median values of the successful trade ratios were0.15–0.2 and 0.4–0.6 when BPP = 10 % and 30%, respec-tively. The highest trade success ratio was obtained whenBPP = 70%, i.e., 0.8–0.9.

Using the same experimental setup, the variations inthe median values of the data sold per hour with variousBPP and DRR configurations are shown in Fig. 25. The vol-umes of data sold and the successful trade ratios exhibiteda similar trend. Fig. 25(d) shows that when DRR = 4 hours,the total utility earned was the lowest when BPP = 10%, butthe highest when BPP = 100%. There was a similar trend inthe volume of data sold when BPP was set to 70% and 100%,as shown in Fig. 26. Fig. 25 also show that the average datavolumes sold during the daytime hours were higher com-pared with those in the early morning and late night. Thistrend agreed with the trend in the number of trades whenDRR = 4 hours, as shown in Fig. 23(b). There was anincrease of 3000–6000 KB in the median values of the vol-ume of data sold when BPP increased from 10% to 70%, asshown in Fig. 25(b) and (f). Fig. 25(d) shows that the datavolume increased by up to 12,000 KB when DRR = 4 hours.These trends demonstrate that the BPP had a very signifi-cant effect on the data sold and number of trades duringthe day. This is intuitive because the number of trades isexpected to increase with the number of buyers.

To investigate this further, we also studied the hourlyvariation in the volume of data sold at different times ofthe day, and the results with BPP values of 10%, 30%, and70% with 1-h and 12-h configurations are shown in Fig. 26.When we fixed SPP = 100% and DRR = 12 hours, Fig. 26 showthat the data sold per day increased significantly with theBPP. There were maximum values of 8000 KB, 30,000 KB,and 70,000 KB when BPP = 10%, 30%, and 70% respectively.We also observed a median shift of 3000–8000 KB in the vol-ume of data sold when BPP increased to 30% and 70%,respectively. A similar trend was also observed when DRRwas set to 1 hour, as shown in Fig. 26.

4.5.3. Effect of the Seller Participation Probability (SPP)In the previous section, we reported the effects of BPP

on the ratio of successful trades and the volume of tradeddata. In this section, we report the effects of the SPP on theratio of successful trades and data volumes sold. We fixedthe BPP to 100% and set SPP at 10%, 30%, and 70%, with 1-hand 12-h refresh rates. The results obtained for the hourlyvariation in the ratio of successful trades are shown inFig. 27, whereas the variation in the volumes of data soldduring these trades is shown in Fig. 28.

Fig. 27 show that the SPP had a marginal effect on thesuccessful trade ratios, which may have occurred becausethe presence of an extra buyer directly affected the number

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

1 3 5 7 9 11 13 15 17 19 21 23

0.0

0.1

0.2

0.3

0.4

0.5

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

0.0

0.2

0.4

0.6

0.8

1.0

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

0.0

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

0.0

0.1

0.2

0.3

0.4

0.5

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

Fig. 24. Variation in the number of trades when seller participation probability (SPP) = 100%.


of trades and the volume of data traded. By contrast, thepresence of another seller did not affect the trade countand the volume of data traded. There was high variationin the successful trades because variation in the BPPdirectly affected the number of buyers present in the adhoc mesh. However, there was a significant median shiftin the successful trade ratios during the early morninghours of 2:00 am to 8:00 am when SPP varied from 10% to

30% and 70%. This occurred because the numbers of activedevices was relatively low in the early morning, so sellerparticipation was of significant importance. However, thenumber of potential sellers also increased throughout theday with higher social activity levels, which also led to anincrease in the successful trade ratios. For example, in theearly morning scenario, if we consider four active users,i.e., one seller and three buyers, there would have been a

Hour of Day

Tra

ded

Dat

a V

olum

e (K

B)

SPP

10%30%70%100%

050

0010

000

1500

0

Hour of Day

Tra

ded

Dat

a V

olum

e (K

B)

BPP

10%30%70%100%

Hour of Day

Tra

ded

Dat

a V

olum

e (K

B)

SPP

10%30%70%100%

Hour of Day

Tra

ded

Dat

a V

olum

e (K

B)

BPP

10%30%70%100%

Hour of Day

Tra

ded

Dat

a V

olum

e (K

B)

SPP

10%30%70%100%

1 3 5 7 9 11 13 15 17 19 21 23

Hour of Day

Tra

ded

Dat

a V

olum

e (K

B)

BPP

10%30%70%100%

050

0010

000

1500

00

5000

1000

015

000

050

0010

000

1500

00

5000

1000

015

000

050

0010

000

1500

0

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

Fig. 25. Hourly variation in the utility with the seller participation probability (SPP) and buyer participation probability (BPP).


loss of three potential trades if the seller decided not to par-ticipate in the auction. Furthermore, in the hours withhigher user activity levels, if we consider 10 active usersin the mesh, i.e., five sellers and five buyers, if one of thesellers decided not to participate, the buyers requirementswould still have been met by the remaining four sellersand the utility was not lost. It is important to note that thissituation is only valid when the number of sellers is small.

We also found that the average data sold with differentseller probabilities exhibited a similar trend. Fig. 28 showthat the SPP had a marginal effect on the volume of data soldand the increase in the data volume, simply because of theincreased user activity during the day. This trend agreeswith the fact that there would have been an equivalentincrease in data trading because users required data morefrequently. In addition, with different data refresh rates,

020

0040

0060

0080

00

Hour of Day

Trad

ed d

ata

Vol

ume

(KB

)

Hour of Day

Trad

ed d

ata

Vol

ume

(KB

)

050

0015

000

2500

0

Hour of Day

Trad

ed d

ata

Vol

ume

(KB

)

1 3 5 7 9 11 13 15 17 19 21 23

Hour of Day

Trad

ed d

ata

Vol

ume

(KB

)

020

000

4000

060

000

Hour of Day

Trad

ed d

ata

Vol

ume

(KB

)

Hour of Day

Trad

ed d

ata

Vol

ume

(KB

)

020

0040

0060

0080

000

5000

1500

025

000

020

000

4000

060

000

1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 231 3 5 7 9 11 13 15 17 19 21 23

Fig. 26. Hourly variation in the data sold when the seller participation probability (SPP) = 100%.


i.e., 1 hour and 12 hours, we found that SPP was not a majorfactor that governed the volume of data sold in the market,as shown in Fig. 25(a) and (e). However, when DRR was setto 4 hours, the utility curves for SPP = 30%, 70%, and 100%exhibited a similar trend, as shown in Fig. 25(c), whereasthis was not observed when DRR = 1 hour and 12 hours,i.e., there was a marginal increase in the data sold withincreasing probabilities, as shown in Fig. 25(a) and (e),

respectively. For the data sold on an hourly basis, Fig. 28shows that there were marginal variations in the volumeof data sold when SPP varied. For SPP = 10% (Fig. 28(a)),the transactions comprised 50,000 KB, whereas forSPP = 30% and 70%, the data sold comprised 80,000 KB, asshown in Fig. 28(c) and (e), respectively. The observed med-ian shifts with variation in DRR were also marginal, whichconfirmed the marginal significance of SPP in the market.

0.0

0.2

0.4

0.6

0.8

1.0

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

1 3 5 7 9 11 13 15 17 19 21 23

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

Hour of Day

Rat

io o

f suc

cess

ful t

rade

s

1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

0.0

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

1 3 5 7 9 11 13 15 17 19 21 23

Fig. 27. Variation in the number of trades when the buyer participation probability (BPP) = 100%.


4.5.4. Combined effect of seller and buyer participationIn our next experiment, we analyzed the variation in

the overall market utility by setting SPP and BPP to 10%and 30% with different data refresh rates, where DRR wasset to 1 hour, 4 hours, and 12 hours. The results are shownin Fig. 29, which indicate that a higher market utility wasobtained with increased buyer participation with differentDRR settings. However, the participation of sellers only had

a marginal effect on the overall utility. We also found thatthe maximum utility in the market was achieved whenDRR = 4 hours, as shown in Figs. 29(c) and (d). The trendsin the variation in the utility when DRR = 1 hour are shownin Fig. 29(a) and (b). There was a significant increase of upto 60,000 KB in the volume of data traded when BPP variedfrom 10% to 30%. A similar trend was observed when DRRwas set to 12 hours, as shown in Fig. 29(e) and (f).


Fig. 29(a) shows that in the month of January whenDRR = 1 hour, the data volume was about 7000 KB,whereas it increased to 8200 KB and 9500 KB whenDRR = 4 hours and 12 hours, respectively, as shown inFig. 29(c) and (e). A similar increase in the data volumeoccurred in the other months of the year. It shouldbe noted that only partial trace data were available forthe month of May, thus the market utility obtained wascomparatively lower in that month compared with other

010

000

3000

050

000

Hour of Day

Tra

ded

data

Vol

ume

(KB

)

1 3 5 7 9 11 13 15 17 19 21 23

Hour of Day

Tra

ded

data

Vol

ume

(KB

)

020

000

4000

060

000

8000

0

Hour of Day

Tra

ded

data

Vol

ume

(KB

)

020

000

4000

060

000

8000

0

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

Fig. 28. Hourly variation in the data sold when the

months. The effects of variations in the BPP on the overallmarket utility when DRR = 1 hour are shown in Fig. 29e,when BPP = 10%, the data volume was 7000 KB but whenBPP = 30%, it increased to 35,000 KB. A similar increaseoccurred when DRR = 4 hours and 12 hours. In this experi-ment, we found that the market utility was governedmainly by the DRR, BPP, and the social activity level ofusers. However, the SPP only had a marginal effect. Theseresults agree with our previous experiments.

010

000

2000

030

000

4000

0Hour of Day

Tra

ded

data

Vol

ume

(KB

)

Hour of Day

Tra

ded

data

Vol

ume

(KB

)

Hour of Day

Tra

ded

data

Vol

ume

(KB

)

020

000

4000

060

000

8000

00

2000

040

000

6000

080

000

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

1 3 5 7 9 11 13 15 17 19 21 23

buyer participation probability (BPP) = 100%.

050

0010

000

1500

020

000

Month Of Year

Tra

ded

Dat

a V

olum

e (K

B)

Month Of Year

Tra

ded

Dat

a V

olum

e (K

B)

Month Of Year

Tra

ded

Dat

a V

olum

e (K

B)

Month Of Year

Tra

ded

Dat

a V

olum

e (K

B)

Month Of Year

Tra

ded

Dat

a V

olum

e (K

B)

Jan Feb Mar Apr May Sep Oct Nov Dec

020

000

4000

060

000

8000

0

Month Of Year

Tra

ded

Dat

a V

olum

e (K

B)

050

0010

000

1500

020

000

050

0010

000

1500

020

000

020

000

4000

060

000

8000

00

2000

040

000

6000

080

000


Jan Feb Mar Apr May Sep Oct Nov Dec Jan Feb Mar Apr May Sep Oct Nov Dec



Fig. 29. Variations in the monthly utility with the seller participation probability (SPP), buyer participation probability (BPP), and the data refresh rate(DRR).


5. Conclusion

The increasing popularity of data-intensive mobileapplications presents a major challenge when trying toensure the quality and cost of services, and it demands

an infrastructure that facilitates collaborative data sharingamong nomadic mobile users. In this study, we argue thatharnessing the mobility patterns and social interactivity ofmobile users could facilitate cellular data offloading andapplication data delivery in a pure ad hoc setting.


Our approach, SmartParcel, allows users to fetch appli-cation content from their neighbors in an ad hoc setting.The system targets data dissemination via the interactionsamong mobile devices in a social context. However, thisapproach is susceptible to the ‘‘tragedy of the commons’’problem due to the lack of incentives to encourage usersto participate in these transactions. To address the issueof incentivizing transactions, we propose an ad hocmarket-based approach called CollabAssure. CollabAssureutilizes spatio-temporal coexisting mobile devicesarranged in a mesh and allows them to participate in anad hoc auction where sellers offer their bids to buyerswho request data services. The system uses a two-waytrade reporting mechanism to ensure fair pricing of theacquired and rendered services in the ad hoc tradingenvironment.

We implemented the SmartParcel and CollabAssure sys-tems in the Android framework and modified the permis-sion model in Android SDK to allow applicationdevelopers to include these services in their mobile appli-cations. However, the modular architecture is independentof the mobile OS and it can be implemented in other mobileOSs supplied by different vendors. Our simulation-basedresults demonstrate that a superior quality of service isensured using our approach. We observed significantincreases in content availability and earned utility whenthe framework was evaluated using real-world traces.

Acknowledgments

This research was supported partly by a FacultyResearch Grant from Fordham University (2013–2014)and by the National Science Foundation (NSF) under GrantNos. CNS-1247875, CNS-0953620, 1116644, and CNS-0845500.

References

[1] Comscore, 2012 mobile future in focus, 2012.[2] Cisco visual networking index: forecast and methodology, 2012–

2017, <http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11-481360_ns827_Networking_Solutions_White_Paper.html>.

[3] Ericsson, Traffic and market report, June 2012: on the pulse of thenetworked society, 2012.

[4] P. Mesh, Bringing low-cost coverage, Capacity and Symmetry toMobile WiMAX, A Tropos Networks white paper, 2007.

[5] D. Calin, H. Claussen, H. Uzunalioglu, On femto deploymentarchitectures and macrocell offloading benefits in joint macro-femto deployments, IEEE Commun. Mag. 48 (1) (2010) 26–32,http://dx.doi.org/10.1109/MCOM.2010.5394026.

[6] V. Chandrasekhar, J. Andrews, A. Gatherer, Femtocell networks: asurvey, IEEE Commun. Mag. 46 (9) (2008) 59–67, http://dx.doi.org/10.1109/MCOM.2008.4623708.

[7] Architecture for mobile data offload over Wi-Fi access networks,<http://www.cisco.com/en/US/solutions/collateral/ns341/ns524/ns673/white_paper_c11-701018.html>.

[8] Alvarion, Mobile data offloading for 3g and lte networks, 2012.[9] B. Han, P. Hui, V.A. Kumar, M.V. Marathe, J. Shao, A. Srinivasan,

Mobile data offloading through opportunistic communications andsocial participation, IEEE Trans. Mobile Comput. 11 (5) (2012) 821–834.

[10] K. Lee, J. Lee, Y. Yi, I. Rhee, S. Chong, Mobile data offloading: howmuch can wi-fi deliver?, in: Proceedings of the 6th InternationalConference, ACM, 2010, p 26.

[11] S. Milgram, The Individual in a Social World: Essays andExperiments, Addison-Wesley Publishing Company, 1977.

[12] B. Kaushik, H. Zhang, X. Fu, B. Liu, J. Wang, Smartparcel: acollaborative data sharing framework for mobile operatingsystems, in: 2013 IEEE 33rd International Conference onDistributed Computing Systems Workshops (ICDCSW), 2013, pp.290–295. http://dx.doi.org/10.1109/ICDCSW.2013.31.

[13] G. Perrucci, F. Fitzek, J. Widmer, Survey on energy consumptionentities on the smartphone platform, in: 2011 IEEE 73rd VehicularTechnology Conference (VTC Spring), IEEE, 2011, pp. 1–6.

[14] N. Eagle, A.S. Pentland, D. Lazer, Inferring friendship networkstructure by using mobile phone data, Proc. Natl. Acad. Sci. 106(36) (2009) 15274–15278.

[15] J.R. Kwapisz, G.M. Weiss, S.A. Moore, Activity recognition using cellphone accelerometers, ACM SIGKDD Explor. Newslett. 12 (2) (2011)74–82.

[16] W. Enck, P. Gilbert, B.-G. Chun, L.P. Cox, J. Jung, P. McDaniel, A.N.Sheth, Taintdroid: an information-flow tracking system for realtimeprivacy monitoring on smartphones, in: Proceedings of the 9thUSENIX Conference on Operating Systems Design andImplementation, 2010, pp. 1–6.

[17] W. Enck, D. Octeau, P. McDaniel, S. Chaudhuri, A study of androidapplication security, in: USENIX Security Symposium, vol. 2, 2011,p. 2.

[18] A.P. Felt, E. Ha, S. Egelman, A. Haney, E. Chin, D. Wagner, Androidpermissions: user attention, comprehension, and behavior, in:Proceedings of the Eighth Symposium on Usable Privacy andSecurity, ACM, 2012, p. 3.

[19] A.P. Felt, H.J. Wang, A. Moshchuk, S. Hanna, E. Chin, Permission re-delegation: attacks and defenses, in: Proceedings of the 20th USENIXSecurity Symposium, 2011, pp. 22–37.

[20] M. Nauman, S. Khan, X. Zhang, Apex: extending android permissionmodel and enforcement with user-defined runtime constraints, in:Proceedings of the 5th ACM Symposium on Information, Computerand Communications Security, ACM, 2010, pp. 328–332.

[21] Google, Android, <http://www.android.com/>.[22] E. Ferro, F. Potorti, Bluetooth and wi-fi wireless protocols: a survey

and a comparison, IEEE Wireless Commun. 12 (1) (2005) 12–26.[23] M. Pitkanen, T. Karkkainen, J. Ott, Opportunistic web access via wlan

hotspots, in: 2010 IEEE International Conference on PervasiveComputing and Communications (PerCom), IEEE, 2010, pp. 20–30.

[24] Signals Research Group, LLC, Femto forum femtocell business case –whitepaper, 2009.

[25] R.Y. Kim, J.S. Kwak, K. Etemad, Wimax femtocell: requirements,challenges, and solutions, IEEE Commun. Mag. 47 (9) (2009)84–91.

[26] I. Ashraf, L.T. Ho, H. Claussen, Improving energy efficiency offemtocell base stations via user activity detection, in: 2010 IEEEWireless Communications and Networking Conference (WCNC),IEEE, 2010, pp. 1–5.

[27] M.V. Barbera, J. Stefa, A.C. Viana, M. Dias de Amorim, M. Boc, Vipdelegation: enabling vips to offload data in wireless social mobilenetworks, in: 2011 International Conference on DistributedComputing in Sensor Systems and Workshops (DCOSS), IEEE, 2011,pp. 1–8.

[28] GPP, TS, 23.234 v6. 2.0, 3gpp system to wireless local area network(wlan) interworking, System description (Release 6), 2004.

[29] R. Braden, D. Clark, S. Shenker, Integrated services in the Internetarchitecture: an overview, June 1994, Tech. rep., Internet RFC 1633,2000.

[30] K. Fall, A delay-tolerant network architecture for challengedInternets, in: Proceedings of the 2003 Conference on Applications,Technologies, Architectures, and Protocols for ComputerCommunications, ACM, 2003, pp. 27–34.

[31] R. Lu, X. Lin, H. Zhu, X. Shen, Spark: a new vanet-based smart parkingscheme for large parking lots, IEEE INFOCOM 2009, IEEE, 2009, pp.1413–1421.

[32] L. Pelusi, A. Passarella, M. Conti, Opportunistic networking: dataforwarding in disconnected mobile ad hoc networks, IEEE Commun.Mag. 44 (11) (2006) 134–141.

[33] Z. Zhang, Routing in intermittently connected mobile ad hocnetworks and delay tolerant networks: overview and challenges,IEEE Commun. Surv. Tutor. 8 (1) (2006) 24–37.

[34] C. Boldrini, M. Conti, J. Jacopini, A. Passarella, Hibop: a history basedrouting protocol for opportunistic networks, in: IEEE InternationalSymposium on a World of Wireless, Mobile and MultimediaNetworks, 2007 (WoWMoM 2007), IEEE, 2007, pp. 1–12.

[35] C. Boldrini, M. Conti, A. Passarella, Contentplace: social-aware datadissemination in opportunistic networks, in: Proceedings of the 11thInternational Symposium on Modeling, Analysis and Simulation ofWireless and Mobile Systems, ACM, 2008, pp. 203–210.

http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11-481360_ns827_Networking_Solutions_White_Paper.html



http://dx.doi.org/10.1109/MCOM.2010.5394026



http://www.cisco.com/en/US/solutions/collateral/ns341/ns524/ns673/white_paper_c11-701018.html

http://www.cisco.com/en/US/solutions/collateral/ns341/ns524/ns673/white_paper_c11-701018.html

http://refhub.elsevier.com/S1389-1286(14)00318-1/h0045











http://dx.doi.org/10.1109/ICDCSW.2013.31





















http://www.android.com/















































[36] R. Ramanathan, R. Hansen, P. Basu, R. Rosales-Hain, R. Krishnan,Prioritized epidemic routing for opportunistic networks, in:Proceedings of the 1st International MobiSys Workshop on MobileOpportunistic Networking, ACM, 2007, pp. 62–66.

[37] S. Milgram, J.E. Sabini, M.E. Silver, The Individual in a SocialWorld: Essays and Experiments, Mcgraw-Hill Book Company,1992.

[38] E. Paulos, E. Goodman, The familiar stranger: anxiety, comfort,and play in public places, in: Proceedings of the SIGCHIConference on Human Factors in Computing Systems, Springer,2004, pp. 223–230.

[39] N. Agarwal, H. Liu, S. Murthy, A. Sen, X. Wang, A social identityapproach to identify familiar strangers in a social network, in:ICWSM, 2009, pp. 2–9.

[40] N. Agarwal, H. Liu, J. Salerno, P.S. Yu, Searching for familiar strangerson blogosphere: problems and challenges, in: NSF Symposium onNext-Generation Data Mining and Cyber-enabled Discovery andInnovation (NGDM), 2007.

[41] L. Sun, K.W. Axhausen, D.-H. Lee, X. Huang, Understandingmetropolitan patterns of daily encounters, Proc. Natl. Acad. Sci.110 (34) (2013) 13774–13779.

[42] H. Zong-you, W. Ding-Wei, Bidding strategies in sealed-bid reversemulti-attribute auctions, in: 2011 Chinese Control and DecisionConference (CCDC), IEEE, 2011, pp. 1762–1767.

[43] J. Shachat, L. Wei, Procuring commodities: first-price sealed-bid orenglish auctions?, Market Sci. 31 (2) (2012) 317–333.

[44] V. Krishna, Auction theory, 2002.[45] R. Cornes, T. Sandler, The Theory of Externalities, Public Goods, and

Club Goods, Cambridge University Press, 1996.[46] S. Scotchmer, Two-tier pricing of shared facilities in a free-entry

equilibrium, Rand J. Econ. (1985) 456–472.[47] K. Lai, L. Rasmusson, E. Adar, L. Zhang, B.A. Huberman, Tycoon: an

implementation of a distributed, market-based resource allocationsystem, Multiagent Grid Syst. 1 (3) (2005) 169–182.

[48] R. Wolski, J.S. Plank, J. Brevik, T. Bryan, Analyzing market-basedresource allocation strategies for the computational grid, Int. J. HighPerform. Comput. Appl. 15 (3) (2001) 258–281.

[49] R. Buyya, C.S. Yeo, S. Venugopal, Market-oriented cloud computing:vision, hype, and reality for delivering it services as computingutilities, in: 10th IEEE International Conference on HighPerformance Computing and Communications, 2008 (HPCC’08),IEEE, 2008, pp. 5–13.

[50] D. Hausheer, B. Stiller, Peermart: decentralized auctions forbandwidth trading on demand, ERCIM News 68 (2007)42–43.

[51] A. Mukherjee, H.M. Kwon, General auction-theoretic strategies fordistributed partner selection in cooperative wireless networks, IEEETrans. Commun. 58 (10) (2010) 2903–2915.

[52] S. Pal, S.R. Kundu, M. Chatterjee, S.K. Das, Combinatorial reverseauction based scheduling in multi-rate wireless systems, IEEE Trans.Comput. 56 (10) (2007) 1329–1341.

[53] P. Xu, S. Wang, X.-Y. Li, Salsa: strategyproof online spectrumadmissions for wireless networks, IEEE Trans. Comput. 59 (12)(2010) 1691–1702.

[54] P. Pałka, R. Schoeneich, Multi-commodity trade application to therouting algorithm for the delay and disruptive tolerant networks, in:New Trends in Databases and Information Systems, Springer, 2013,pp. 241–250.

[55] P. Maillé, B. Tuffin, Multibid auctions for bandwidth allocation incommunication networks, in: Twenty-third AnnualJoint Conferenceof the IEEE Computer and Communications Societies, INFOCOM2004, vol. 1, IEEE, 2004.

[56] P. Maillé, Auctioning for downlink transmission power in cdmacellular systems, in: Proceedings of the 7th ACM InternationalSymposium on Modeling, Analysis and Simulation of Wireless andMobile Systems, ACM, 2004, pp. 293–296.

[57] A. Mukherjee, H.M. Kwon, General auction-theoretic strategies fordistributed partner selection in cooperative wireless networks, IEEETrans. Commun. 58 (10) (2010) 2903–2915.

[58] H. Falaki, R. Mahajan, S. Kandula, D. Lymberopoulos, R. Govindan, D.Estrin, Diversity in smartphone usage, in: Proceedings of the 8thInternational Conference on Mobile Systems, Applications, andServices, ACM, 2010, pp. 179–194.

[59] G. Hardin, The tragedy of the commons, J. Nat. Res. Policy Res. 1 (3)(2009) 243–253.

[60] Apple App Store, <www.apple.com/iphone/from-the-app-store/>.[61] Google Play, <play.google.com/store>.

[62] G. Iosifidis, I. Koutsopoulos, Double auction mechanisms forresource allocation in autonomous networks, IEEE J. Select. AreasCommun. 28 (1) (2010) 95–102.

[63] I. Senturk, S. Yilmaz, K. Akkaya, Connectivity restoration in delay-tolerant sensor networks using game theory, Int. J. Ad Hoc Ubiquit.Comput. 11 (2) (2012) 109–124.

Bhanu Kaushik received his Bachelor oftechnology degree in Computer Science andEngineering from Kurukshetra University,Kurukshetra, India. He received his Masters ofScience and Doctorate degree in ComputerScience from the University of MassachusettsLowell, in 2012 and 2014, respectively. Hisresearch focuses on computer networks,mobile data off-loading, and opportunisticand ad hoc networking. His research interestalso include Big Data management, analysisand visualization. He is a student member of

IEEE and ACM. He has also provided his services as a designated reviewerfor Globecomm’14, CSNT’13 and Wireless Networks (WINET).

Honggang Zhang holds a PhD in ComputerScience (2006) from the University of Massa-chusetts, Amherst, USA. He received his BSdegree from the Central South University ofChina, and his MS degree from Tianjin Uni-versity of China. He also received a MS degreefrom Purdue University, West Lafayette, IN,USA. He is currently an Associate Professor inthe Department of Computer and InformationScience at Fordham University Bronx, NY. Hisresearch interests span a wide range of topicsin the area of computer networks and dis-

tributed systems. His current research focuses primarily on online socialnetworking, mobile computing/networking, data swarming systemsincluding peer-to-peer file sharing and media streaming networks, and
network security. He received the National Science Foundation (NSF)CAREER Award 2009–2014.
Xinyu Yang received the Diploma in Com-puter Science and Technology from Xi’anJiaotong University of China in 2001 and theBachelor, Master, and PhD degrees from Xi’anJiaotong University in 1995, 1997, and 2001,respectively. He has held positions with theDepartment of Computer Science and Tech-nology at Xi’an Jiaotong University, where heis a professor teaching in the Department ofComputer Science and Technology. His maintechnical interests lie in the areas of networksecurity, wireless communication, and mobilead hoc networks.

Xinwen Fu is an assistant professor in theDepartment of Computer Science, Universityof Massachusetts, Lowell. He received his BS(1995) and MS (1998) degrees in ElectricalEngineering from Xi’an Jiaotong University,China and the University of Science andTechnology of China, respectively. Heobtained his PhD (2005) in Computer Engi-neering from Texas A&M University. From2005 to 2008, he was an assistant professorwith the College of Business and InformationSystems at Dakota State University. In sum-

mer 2008, he joined the University of Massachusetts Lowell as a facultymember. His current research interests are network security and privacy.














































































http://www.apple.com/iphone/from-the-app-store/

http://play.google.com/store








Benyuan Liu received his BS degree in physicsfrom the University of Science and Technol-ogy of China (USTC) in 1994, his MS degree inphysics from Yale University in 1998, and aPhD degree in computer science from theUniversity of Massachusetts, Amherst in 2003.He is currently an associate professor ofcomputer science at the University of Massa-chusetts Lowell. His main research interestsinclude protocol design and performancemodeling for wireless ad hoc and sensor net-works. He served as the technical program co-

chair for the International Conference on Wireless Algorithms, Systems,and Applications (WASA) in 2009, the technical program co-chair of theCross-layer Design and Optimization Track for the IEEE International
Conference on Computer Communication Networks (ICCCN) in 2008, andtechnical program committee member for numerous computernetworking conferences. He was a US National Science Foundation (NSF)CAREER Award recipient in 2010. He is a member of the IEEE.
Jie Wang is Professor and Chair of ComputerScience at the University of MassachusettsLowell. He received his PhD in Computer Sci-ence from Boston University in 1991, and hisMS in Computer Science and BS in Computa-tional Mathematics from Sun Yat-sen Uni-versity in 1985 and 1982, respectively. Hisresearch interests include combinatorialoptimization, complex networks, networksecurity, wireless networks, and mathemati-cal modeling. His research has been fundedcontinuously by the NSF since 1991. IBM,

Intel, and Google have also funded his research. He has published over150 research papers, authored and co-authored five books, and edited andco-edited four books.

Date post:	05-Jun-2018
Category:	Documents
Upload:	domien
View:	216 times
Download:	0 times

Providing service assurance in mobile opportunistic networksxinwenfu/paper/Journals/14_CN_74... ·...

Documents