Routing in Mobile Opportunistic Social Networks …downloads.hindawi.com › journals › wcmc ›...

Research ArticleRouting in Mobile Opportunistic Social Networks withSelfish Nodes

Annalisa Socievole1 Antonio Caputo 2 Floriano De Rango 2 and Peppino Fazio 2

1National Research Council of Italy (CNR) Institute for High Performance Computing and Networking (ICAR)Via Pietro Bucci 7-11C 87036 Arcavacata di Rende (CS) Italy2Department of Informatics Modeling Electronics and Systems Engineering (DIMES) University of Calabria87036 Arcavacata di Rende (CS) Italy

Correspondence should be addressed to Floriano De Rango derangodimesunicalit

Received 27 July 2018 Accepted 29 November 2018 Published 3 February 2019

Guest Editor Enrique Hernandez-Orallo

Copyright copy 2019 Annalisa Socievole et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

When the connection to Internet is not available during networking activities an opportunistic approach exploits the encountersbetween mobile human-carried devices for exchanging information When users encounter each other their handheld devices cancommunicate in a cooperative way using the encounter opportunities for forwarding their messages in a wireless manner Butanalyzing real behaviors most of the nodes exhibit selfish behaviors mostly to preserve the limited resources (data buffers andresidual energy) That is the reason why node selfishness should be taken into account when describing networking activities inthis paper we first evaluate the effects of node selfishness in opportunistic networks Then we propose a routing mechanism formanaging node selfishness in opportunistic communications namely SORSI (Social-basedOpportunistic Routing with Selfishnessdetection and Incentive mechanisms) SORSI exploits the social-based nature of node mobility and other social features of nodesto optimize message dissemination together with a selfishness detection mechanism aiming at discouraging selfish behaviors andboosting data forwarding Simulating several percentages of selfish nodes our results on real-worldmobility traces show that SORSIis able to outperform the social-based schemes Bubble Rap and SPRINT-SELF employing also selfishness management in termsof message delivery ratio overhead cost and end-to-end average latency Moreover SORSI achieves delivery ratios and averagelatencies comparable to Epidemic Routing while having a significant lower overhead cost

1 Introduction

Even if the Internet with its ubiquity has revolutionizedthe way in which we communicate there are still somescenarios where this network infrastructure is not availableExamples of these scenarios are disasterrecovery situationsbig sports events and music festivals or more in generalareas where the Internet is expensive not available oroverloaded In such situations an alternative communicationmedium is necessary Delay Tolerant Networks (DTNs) [1ndash3] and Opportunistic Networks [4] have gained a lot ofinterests in these last years thanks to their capability to facethe uncertainty of a fixed network infrastructure providingconnectivity between mobile devices In future Internet thereare some scenarios where it is not possible to guarantee anany-time end-to-end connectivity and it is not always known

a priori the topology or network infrastructure Examplesof these scenarios are disasterrecovery situations big sportsevents and music festivals or more in general areas wherethe Internet is expensive not available or overloaded Insuch situations an alternative communication medium ismandatory and at this purpose DTNs and ONs can becomea promising communication paradigm [5ndash7] Since an end-to-end path between mobile phone hand-on devices orother terminals is not always available opportunistic routingalgorithms need to be designed with different features incomparison with routing for traditional Mobile Ad HocNetworks (MANETs) Because nodes are not always part ofa precomputed path it is necessary for mobile nodes to storemessages waiting tomeet during their movement some nodeswith good characteristics (on the basis of the metric) whereto forward the message Because it is not known how long

HindawiWireless Communications and Mobile ComputingVolume 2019 Article ID 6359806 15 pageshttpsdoiorg10115520196359806

2 Wireless Communications and Mobile Computing

time a packet can be buffered novel forwarding strategies anda specific bundle layer need to be designed to guarantee ahop-by-hop data delivery Moreover a stronger store-carry-forward approach needs to be applied for each node andall nodes form a DTN domain due to their capability totolerate data delivery delay In this context opportunisticnetworking and forwarding can be a useful and mandatoryapproach to offer to each node an opportunity to send datapacket freeing space in its buffer [8ndash11] The opportunisticapproach assumes that nodes that want to transmit can usesome specific metric to select the next hop where forwardingthe data and the receiving node is assumed that want toreceive the data packet sent by node encountered Howeverin these last years the second assumption is not alwaysconsidered because it has been observed as in the real lifenodes can be also selfish instead of cooperative Becausesome resources are limited such as buffer size energy budgetor other constraints mobile nodes cannot see benefits inparticipating in a collaborative way to the communicationThis selfish behavior can also be evident not at the beginningof the communication but during the network dynamic onthe basis of the constraints and budgets that can be exhaustedIn this case if some countermeasures are not adopted thereis the risk that after some period some nodes stop to forwarddata packet increasing the data delay reducing the data packetdelivery ratio and losing the advantage of a flexible paradigmsuch as DTNs

This paper proposes a routingmechanism for opportunis-tic networks able to mitigate the effect of node selfishnessExploiting the social-based nature of human encounters wepropose a selfishness-aware opportunistic routing schemeable to maximize message delivery As a matter of factmessage dissemination in opportunistic networks can beefficiently performed using social-based routing metricsSeveral works such as [6 7 12ndash17] demonstrate that thenodes with a highly social behavior are good carriers andare thus able to improve message delivery However howto extract node sociality and define efficient routing metricsis still challenging First opportunistic networks are highlydynamic and it is not easy to reconstruct a nodersquos socialbehavior Second when representing the network as a socialgraph several metrics are able to define the sociality and thesocial position of a node (see for example the social metricsanalyzed in [18]) Each of these metrics is not always easilycomputable in a distributed way on a dynamic graph Thirdeach node may interact in different social contexts like realworld when moving online world using social networkingwebsites and so on As such each node has a sociality thatcan be defined ldquomultilayerrdquo [18]

Starting from the above considerations to select aneffective forwarding node the routing scheme we proposenamed SORSI (Social-based Opportunistic Routing withSelfishness detection and Incentive mechanisms) measuresthe forwarding capability of a node when compared toan encountered node in terms of node centrality (ie theimportanceposition of a node within a social graph) tiestrength and link prediction Each of these routing metricsis computed in a distributed way on a particular socialnetwork layer that describes a social dimension Since each

mobile user may have social relationships on several socialdimensionslayers (eg real life Facebook Twitter etc) thisprotocol exploits the user multilayer sociality for computingforwarding paths and improving message disseminationMoreover it adds a selfishness detection mechanism basedon the history of message exchanges for incentivizing theforwarding of packets Starting from the work in [19] wherethe effect of selfish nodes in the network performance hasbeen presented in this paper we propose a mechanism todetect selfish nodes and to discourage their behavior Thismechanism has been designed and implemented in a multi-layer social routing in order to test its effectiveness and it hasbeen compared with other well-known data disseminationtechniques for opportunistic networks So the contributionsof this paper can be summarized as follows

(i) We characterize the situations in which an oppor-tunistic network node may act selfishly

(ii) We analyze the effect of node selfishness in oppor-tunistic networks through simulations

(iii) We propose a multilayer social-based routing schemewith selfishness management and compare it to threereference opportunistic routing protocols over twoexperimental datasets of human mobility with differ-ent connectivity patterns

(iv) We show that in general as the proportion of nodesacting selfishly in the network increases the messagedelivery ratio decreases and the average end-to-endlatency increases

(v) We show that the use of multilayer social networkinformation together with selfishness detection isable to achieve delivery ratios and average latenciescomparable to epidemic delivery with a much loweroverhead cost

The paper has been organized as follows Section 2provides background information Section 3 discusses theselfish node problem in opportunistic networks and analyzesits effect on routing through simulations Section 4 describesour proposal SORSI In Section 5 we detail the simulationsetup for SORSI evaluation and discuss the results FinallySection 6 concludes the paper summarizing the resultsobtained and discussing the future research directions

2 Related Works

Many works in literature in these last years are focusing onthe selfish behavior and on its effect on network performanceHowever a few contributions till now have been relatedto DTNs and ONs because for these networks it is morechallenging to detect malicious behavior and it is not easyto propose effective incentive mechanism that can tradeoffbetween network performance and resource draining

21 Selfishness Management in MANET In MANET anddistributed wireless networks some monitoring techniquesto detect malicious behavior or selfish nodes are applied Oneof the most adopted techniques is the watchdog technique

Wireless Communications and Mobile Computing 3

such as that presented in [20] According to this approachnodes monitor each otherrsquos communications to ensure if theconsidered node behaves as expected or not On the basisof the higher or lower adherence to an expected behaviorit is assigned a reputation value The computed reputationvalue is then compared with some thresholds to decideif the node is reliable selfish or to be monitored Themain limitation of this approach however is that withinan opportunistic network many encounters might not beobserved by a third party In [21] authors use the currencya method based on purchasing credits to assign to mobilenodes for their forwarding service A general model forelectronic coupons where a central system spreads electroniccoupons among interested users via access points installedin shops is proposed In [22] authors analyze the effect inthe introduction of a trust management scheme applied inMANET to detect selfish nodes and to isolate them Also iftrust management can be useful to avoid malicious nodesit can be expensive in terms of energy if it is not appliedcarefully on a distributed network

22 Selfishness Management in DTNs and ONs The absenceof a predetermined path and a priori nodes belonging to apath can present some risks in the data forwarding if somemechanism to incentive the cooperation is not applied In[11] authors propose a mechanism where all nodes can storein their memory a list of topicsinterests When they meetsome other node that can be interested to one or more ofthese topics they can spread the topic if the other nodebehaves in the same way This means that noncooperativenodes will not share interesting topics in the network andthis penalizes the egoistic behavior In [23] authors proposecollaborative contact-based watchdog (CoCoWa) as a col-laborative approach based on the diffusion of local selfishnodes awareness when a contact occurs so that informationabout selfish nodes is quickly propagated This collaborativeapproach reduces the time and increases the precision whendetecting selfish nodes In [18] authors propose an incen-tive mechanism called IRONMAN It uses social networkinformation to bootstrap the detection and discouragementof selfishness through an asynchronous bilateral trading Inthis work authors try to give a rank to nodes that behavecooperatively and forward data In particular nodes thatforward data on behalf of other node can increase theirrank whereas other nodes that discard packets or behaveselfishly decrease their rank and they are not involved inthe data forwarding and sharing The social network such asFacebook is considered in the proposal to establish the initialrank value on the online social network among involvednodes In [24] Ciobanu et al propose SPRINT-SELF arouting mechanism for ONs that use preexisting online socialinformation for detecting communities and the predictionof future encounters for routing data This routing strategyallows node to keep info about past data transfers and batterylevel so that through these values they are able to computean altruism score that is used to select or not the next hopforwarding the data Only if this score related to the altruismis within certain range the message is sent

3 The Selfish Node Problem inOpportunistic Networks

From the viewpoint of an individual selfishness is commonlydefined as a set of attitudes and behaviors aimed uniquelyto achieve the proper personal interests A selfish individualpursues his goals even at the cost of damaging the interestsof others How this behavior is translated in the context ofopportunistic networks Many researchers such as Urpi etal [25] consider the tradeoff between the cost in terms ofenergy consumption and the benefits in terms of networkthroughput as a key aspect to define an eventual selfishbehavior of a node involved in a routing operation Anopportunistic network node may thus act in a selfish waywhen induced by energy constraints Another valid reasonfor a node to act selfishly is related to message buffer statusIf a node has its buffer full it may act selfishly deciding tofirst drop the other nodesrsquo packets Moreover consideringan opportunistic networking environment where the nodeshave high mobility and frequent disconnections the contactduration between them might be highly variable and this isa further potential reason for being selfish A node onceassessed to have many short contacts may decide to exploita contact for first exchanging its messages and subsequentlythe messages belonging to other nodes in order to leave thepossible fragmentation of a message due to the interruptionof a wireless contact only to other nodes Figure 1 depictsthe three main reasons for deciding to act selfishly in anOpportunistic Network Environment

Hypothesizing that a node acts in a selfish way forthe reasons described above the routing performance ofopportunistic forwarding may severely degrade thus caus-ing serious problems to the communications within theopportunistic network In this section we show through apreliminary analysis how routing performance is affected bynode selfishness For modeling node selfishness we proceedas follows Through a uniform distribution we randomlychoose a subset of network nodes labeling these nodes asselfish We consider 25 50 and 75 of selfish nodes andthe limit case when all nodes are selfish We hypothesize thateach of these nodes acts selfishly for one or more of the threereasons described above For this initial study we implementa basic node selfishness behavior When a selfish nodeencounters another node they will exchange their messagesas usually done in an opportunistic networking scenario witha certain routing scheme However the selfish node will dropthe messages received for which it is not the destination Thefollowing subsections describe the simulation setup and theresults of this preliminary analysis

31 Simulation Setup In order to test the effect of nodeselfishness on opportunistic routing we carried severalsimulations using the Opportunistic Network Environment(ONE) simulator [26] This simulation environment modelsa node with a radio interface persistent storage severalmovement models (including the simulation of realisticmobility traces) several routing capabilities a basic energyconsumption model and different application interactions


A

B

C

I amselfish

Energy saving Buffer overload Contact duration

Figure 1 Node A wants to send a message to node C using node B as forwarder Node A forwards the message to node B but node B is selfishand will not forward the message to node C because it wants to preserve his battery (energy constraints) it wants to leave space in the bufferdeleting the messages belonging to other nodes (buffer overflow) andor it wants to privilege its own messages during routing due to shortcontact durations (contact duration)

Table 1 Characteristics of the two experimental datasets

Experimental dataset SASSY LAPLANDEnvironment AcademicUrban ConferenceDevice type T-Mote ImoteRadio interface ZigBee BluetoothGranularity 667 s [120-600] sDuration 70 days 3 days of nodes 27 17

In the following subsections we describe the mobility tracesused the routing protocols we simulated and the perfor-mance measures analyzed

311 Human Mobility Traces We utilize two experimentaldatasets of human mobility where the mobile devices ransoftware logging contacts between them Table 1 summarizesthe main features of the selected datasets SASSY dataset [27]contains the ZigBee encounter logs of 27 participants carry-ing T-Mote sensors and their social network generated fromFacebook data self-declared by candidates at the beginningof the experiment The experiment took place at Universityof St Andrews (United Kingdom) for an overall duration of3 months between February 15 2008 and April 29 2008 Forour analysis we consider one week of encounters focusingon the week having the highest number of encountersLAPLAND [28] dataset spans a shorter period and wascollected during the ExtremeCom09 workshop in PadjelantaNational Park (Sweden) The Bluetooth colocation data of17 conference attendees were gathered during 4 consecutivedays of the experiment from August 9 2009 to August 122009 Each candidate was asked to carry Imotes with himdetecting devices in proximityThis dataset includes also eachparticipantrsquos Facebook friend list and interests in terms ofscientific topics

312 Routing Protocols In simulations we analyze threebenchmark opportunistic routing protocols Epidemic Rout-ing [29 30] Spray amp Wait [31] and Bubble Rap [13] In thefollowing lines a brief explanation motivates the reason forwhich we choose to consider these protocols for our study

(i) Epidemic Routing is a flooding-based protocol Whentwonodes encounter they exchange all theirmessagesso that these messages are spread like viruses bypairwise contacts between two nodesThis protocol isconsidered a reference for opportunistic routing sinceit determines an upper bound for message delivery

(ii) Spray amp Wait is a different kind of epidemic routingwhich floods the network with a fixed number ofcopies of a message The source node ldquospraysrdquo 119871message copies to 119871 distinct encountered nodes andthen ldquowaitsrdquo hoping that one of these nodes willcarry the message to the intended destination nodeIf the destination node is not found during the sprayphase each of the 119871 nodes holding a message copywill forward the message only to the destinationnode This algorithm has shown to have routingperformance near to the classic epidemic routing witha lower overhead cost

(iii) Bubble Rap [13] is a social-based protocol consid-ered one of the most efficient protocols in termsof messages delivered and delay in delivering themBubble Rap uses two centrality values associatedwith each node based on its global popularity in thewhole network and the local popularity within itscommunity or communities The forwarding schemeuses these centrality values so that a message istransferred to nodes with higher global centralityvalues until the carrier node meets a node with thesame community label as the destination nodeThena message is forwarded to nodes with higher localrankings until successful delivery Since opportunisticnetworks are characterized by a social nature due to


human mobility this protocol has shown to reducethe number of message replicas spread on the net-work while maintaining a good message deliveryFor our simulations we chose C-Window degree[13] for implementing node centrality since it can becomputed in a fully distributed way providing a goodestimate of the suitability of a node as message relayThis is a cumulative centrality measure averagingdegree centrality (ie the number of unique contactshad with the encountered nodes) over a fixed numberof temporal windows For detecting communities wechose k-Clique [32] Even if several community detec-tion methods are present in the literature we chosethis method since it finds overlapping communitieswhich are more similar to the communities formedin real life For this community detection method acommunity is defined as the union of all 119896-cliques(complete subgraphs with 119896 nodes) that can reacheach other through a series of adjacent 119896-cliqueswhere two 119896-cliques are said to be adjacent if theyshare 119896 minus 1 nodes

313 Performance Metrics We use two important oppor-tunistic performance metrics for our analyses First we studythe system throughput or delivery ratio which is computedhere as the number of delivered packets divided by thenumber of unique packets created in the system Then westudy the system delay or delivery delay which only considersthe delivered messagesThismetric measures the time it takesa packet to be delivered to the destination node

We evaluated these metrics as a function of the TTL(Time To Live) which represents the maximum time amessage can stay in the system after its creation The TTL isfundamental for studying the ability of a routing protocol tofind an adequate number of relay nodes within a certain timeThe values of themain parameters used in our simulations areshown in Table 4 We repeated each simulation 20 times andtook the average of each run as a result

32 The Selfishness Effects

321 SASSY Results We start evaluating the effects of nodeselfishness on opportunistic routing by analyzing the SASSYmobility trace The main simulation parameters are listed inTable 2 Delivery ratio and average latency for this trace areshown in Figure 2 By analyzing delivery ratio we observethat as the proportion of nodes acting selfishly in the networkincreases this network performance decreases for all therouting protocols considered This means that if the selfishbehavior can be in some way detected and discouraged itmight be possible to achieve the same performance as if nonodes behave selfishly even if they have a propensity for beingselfish Spray ampWait delivery compared to epidemic deliveryachieving the highest message delivery is slightly lower as itcan be expected since it disseminates a limited number ofmessage copies As far as Bubble Rap evaluation concerns wefound an interesting result as the percentage of selfish nodesincreases the system throughput does not vary significantly

Table 2 Values for the simulation parameters

Parameter ValueNetwork Buffer size 2000MB

Message lowast size 1 kBIntermessage creation interval 1800 s

Spray amp Wait L 5Bubble Rap C-Window duration 6 hours

C-Window of windows 5k (k-Clique) 3

lowast Eachmessage is exchanged between randomly selected source-destinationpairs

For a message TTL equal to one week for example theprotocol is characterized by themaximumdifference betweendelivery with no selfish nodes and with all selfish nodes thatis equal to 01 We consider this difference low comparedto Epidemic Routing and Spray amp Wait This social-basedprotocol if on one hand is characterized by a lower deliveryratio due to its more restrictive forwarding rules on the otherhand it is able to better manage node selfishness We notethat by choosing the relay nodes that are more social theselfishness effect can be better balanced Here we think thatby choosing only the most central nodes as node relays theprobability to find a relay node that can be also selfish isreduced In the case of Epidemic Routing on the contrarywhere each encountered node is chosen as relay node theprobability to find a selfish relay node is higher A similarthing happens in Spray amp Wait It can be further observedthat all algorithms deliver more packets to the destinations asthe TTL increases However as the TTL becomes high theincrement in the delivery ratio is marginal since the capacityof the network to forward packets becomes the performancebottleneck

On the system delay node selfishness causes a degra-dation of this performance as the TTL increases We notethat for TTLs greater than 3 days the average latency highlydepends from the percentage of selfish nodes that are presentwithin the network while for lower TTLs node selfishnessdoes not particularly degrade this protocol performanceThiseffect characterizes all the protocols consideredThis happensbecause the longer TTLs result in higher probabilities forthe packets to be relayed to selfish nodes during their pathtowards the destination These packets will be discarded bythe selfish nodes thus increasing the corresponding averagelatencyWe can further note that BubbleRap is less influencedby the percentage of selfish nodes The average latencyslightly varies from 0 to 100 of selfish nodes Similarlyfor the delivery case the social-based routing rules result ina better management of node selfishness Finally comparingthe delay values achieved by the three protocols consideredEpidemic Routing has the highest average latency followedby Spray amp Wait and Bubble Rap Even if Epidemic Routinghas usually the lowest delay due to the highest number ofmessage replicas injected into the network in this mobilityscenario the selfish behavior of some relay nodes togetherwith the particular node encounters dynamics that often


1 2 3 4 5 6 7Message TTL (days)

0 selfish25 selfish50 selfish

75 selfish100 selfish

0

02

04D

eliv

ery

Ratio

(a) Epidemic Delivery



times 104

2 3 4 5 6 71Message TTL (days)

0

5

10

15

Aver

age L

aten

cy [s

]

(b) Epidemic Latency




0

02

04

Del

iver

y Ra

tio

(c) Spray ampWait Delivery



times 104

0

5

10

15

Aver

age L

aten

cy [s

]


(d) Spray ampWait Latency




0

01

02

Del

iver

y Ra

tio

(e) Bubble Rap Delivery



times 104

0

5

10

15

Aver

age L

aten

cy [s

]


(f) Bubble Rap Latency

Figure 2 SASSY routing performance under different of selfish nodes

assign packets to these relays resulting in packets deliveredwith high delays

322 LAPLANDResults LAPLANDdataset covers a shorterexperimental period and has a smaller number of nodescompared to SASSY As such the TTLs have been variedfrom 1 to 7 hours As it can be observed from the results inFigure 3 similarly to SASSY the increase in the proportionof selfish nodes results in a lower delivery ratio for all theprotocols considered We thus conclude that also in a smaller

dataset this feature is present However Bubble Rap showsagain to better manage node selfishness as it can be observedby the trend of the curves in Figure 3(c) We can furthernote that again all the algorithms deliver more packets to thedestinations as the TTL increases

The average latency results confirm that the system delayincreases as the TTL and the percentage of selfish nodesincrease In particular for these lower TTL values theaverage latency is characterized by an almost linear trendwhich we also found in SASSY for TTLs lower than 3 days


2 3 4 5 6 71Message TTL (hours)

02

04

06

08D

eliv

ery

Ratio




0

2000

4000

6000

Aver

age L

aten

cy [s

]






02

03

04

05

Del

iver

y Ra

tio



(c) Spray amp Wait Delivery


0

2000

4000

6000

Aver

age L

aten

cy [s

]




02

03

04

05

Del

iver

y Ra

tio





0

2000

4000

6000

Aver

age L

aten

cy [s

]





Figure 3 LAPLAND routing performance under different of selfish nodes

Here the interesting results deal again with Bubble Rap forthis social-based protocol the values of average latency aresimilar both if the system contains selfish nodes and if thesystem has all altruistic nodes This result again confirmsthat the social-based protocols offer a promising researchdirection concerning nodes selfishness and opportunisticrouting Comparing the average latency results achieved bythe protocols we found an opposite trend with respect toSASSYHere EpidemicRouting has the lower average latencyfollowed by Spray amp Wait and Bubble Rap These results aremore similar to those usually found in the literature since inmost cases the epidemic delivery results in a lower latency(even if at a higher overhead cost) due to its flooding nature

4 SORSI Social-Based OpportunisticRouting with Selfishness Detection andIncentive Mechanisms

After being assessed through simulations that selfishness canseverely degrade opportunistic routing we now describe howthis problem can be mitigated through a selfish detectionmechanism incentivizing node collaboration Our schemenamed SORSI (Social-based Opportunistic Routing withSelfishness detection and Incentive mechanisms) exploitsnode sociality and history of encounters to detect selfishnessOnce selfishness is detected nodes do not forward messagesto selfish nodes As such selfish nodes are incentivized to


Table 3 List of symbols used to define SORSI social metric

List of symbols119894 119894119905ℎ node119895 119895119905ℎ node119889 destination node119905 119905119905ℎ time slot119879 number of time slots119872119894 number of neighbors of node 119894119872119895 number of neighbors of node 119895119897 119897119905ℎ online social network layer119871 number of online social network layers119890 encounter event119862119863119890119892119903119890119890 temporal degree centrality119862119862119863119890119892119903119890119890 cumulative degree centrality119879119878 online tie strength119879119878119879119874119879 total online tie strength119871119875 link predictor (common neighbors)119862119878 centrality utility score119879119878119878 tie strength utility score119871119875119878 link predictor utility score119872119871119878 SORSI utility score119864119878119878 SORSI encounter-based selfishness score

participate in forwarding if they want their messages to berouted

Social-based rules for opportunistic forwarding havebeen shown to result in low-cost forwarding paths (seefor example [6 7 33]) Moreover our previous analysisshows that Bubble Rap which is a reference for social-basedopportunistic routing protocols is able to better sustain theselfishness effect As such we choose to build SORSI routingscheme on a social-based logic However differently fromBubble Rap only considering offline sociality reconstructedfrom node encounters SORSI is based on a routing metricnamelyMLS which exploits three social dimensions wirelessproximity online friendships and interests Since each ofthese dimensions is able to represent a nodersquos sociality wechoose to consider this set of social features in order to have awider view of a nodersquos social behavior As a matter of fact theway in which we move interact online with our friends andshare our interests represents our socialityMLSmetric is thuscomputed using a combination of three measures (1) nodecentrality computed on the DSN (Detected Social Networkie the social network detected through nodesrsquo encounters)graph layer (2) tie strength computed on the OSN (OnlineSocial Network ie Facebook Twitter etc) graph layer(s)and (3) a link predictor computed on the Interest networklayer (ie a social layer constructed on node interests) Assuch we model the network nodesrsquo sociality as a multilayersocial network where each layer is a social graph representinga social dimension and computing MLS metric Table 3 liststhe symbols used to define the social metric

We consider centrality as one of the most importantfactors to choose a good message relay In graph theoryand network analysis centrality quantifies the structural

importance of a vertex within the graph A central node hasusually a stronger capability of connecting other networknodes We therefore compute centrality at the DSN layerwhere the corresponding social graph is leveraged throughencounters between mobile devices Here the DNS socialgraph of the multilayer social network is a dynamic graphwhere an edge between two nodes represents a wirelesscontact There are several ways to measure centrality [18]SORSI social metric computes node centrality for a node 119894119862119862119863119890119892119903119890119890(119894) using a long-term cumulative estimate of degreecentrality Degree centrality basically quantifies the numberof connections a node has The advantage in using this mea-sure is that it can be easily computed locally considering onlya nodersquos ego network while other centrality measures (egbetweenness closeness or eigenvector centrality) requireglobal knowledge of the network More specifically SORSIcomputes the number of unique nodes seen throughout aspecific time slot and then averages this measure with a setof previous measures Degree of centrality for a node 119894 duringa time slot 119905 is computed as follows

119862119863119890119892119903119890119890 (119894 119905) =119872119894

sum119895=1

119890 (119894 119895 119905) (1)

where

119890 (119894 119895 119905) =

1 119894119891 119894 119890119899119888119900119906119899119905119890119903119904 119895 119889119906119903119894119899119892 119905119894119898119890 119904119897119900119905 1199050 119900119905ℎ119890119903119908119894119904119890

(2)

representing an edge between node 119894 and node 119895 on the DSNgraph corresponding to the time slot considered (consideringthat theDSNgraph is a temporal graphwe forma static graphfor each time slot by amalgamating all contacts in that timeinterval) and 119872119894 is the number of nodes in 119894rsquos range Thecumulative degree119862119862119863119890g119903119890119890(119894) is then calculated by averagingthe nodersquos degree values over a set of119879 time slots including themost recent time slot and all the previous ones

119862119862119863119890119892119903119890119890 (119894) =1119879119879

sum119905=0

119862119863119890119892119903119890119890 (119894 119879 minus 119905) (3)

In that way SORSI provides a fully decentralized approx-imation for a nodersquos degree centrality which is easy to becomputed

Centrality described above is measured using the historyof contacts and does not consider future links availabilityConsidering that the links in the network are time-varyingan existing link to a central node may not be highly availableWe therefore include a tie strength indicator into SORSIsocial metric This indicator is able to identify the links thathave a higher probability to be activated and is measuredby considering online social ties between the individualscarrying the mobile devices This choice is driven by theconsideration that social ties on online social networkingwebsites such as Facebook Twitter (here we consider a tiebetween a user A and a user B if A follows B and viceversa) or LinkedIn are more stable and hence stronger thancontact network ties Typically an online tie between two





TTL (SASSY) 4 daysTTL (LAPLAND) 240 s

SORSI-NS Time slot 6 hoursT 5

SORSI Selfishness threshold 119905ℎ119903 50Initial selfish score ESS (altruistic node) 0Initial selfish score ESS (selfish node) 100

Time slot 6 hoursT 5

Bubble Rap C-Window duration 6 hoursC-Window of windows 5

k (k-Clique) 3SPRINT-SELF 1199081 07

1199082 03k (k-Clique) 3

Cache size (I and O) 100 MBlowast Each message is exchanged between randomly selected source-destination pairs

users does not change over time In Facebook for exampleldquointermittentrdquo friendships are highly improbable Moreoverit has been shown in [18 20] that nodes encounter otheronline socially connected nodes with a high probabilityConsequently online ties can be considered a good measureof whether a link on DSN will be activated SORSI calculatestie strength between node 119894 and node 119895 at OSN layer l as

119879119878 (119894 119895 119897)

=

1 119894119891 119894 119886119899119889 119895 119886119903119890 119888119900119899119899119890119888119905119890119889 119886119905 119897119886119910119890119903 1198970 119900119905ℎ119890119903119908119894119904119890

(4)

The total tie strength between two nodes is the sum of theindicators measured at each OSN layer

119879119878119879119874119879 (119894 119895) =119871

sum119897=1

119879119878 (119894 119895 119897) (5)

where 119871 is the total number of online social networkingwebsites considered

SORSI social metric takes into account a third measureuseful to predict future collaborations between two nodes Alink predictor is computed on Interest network layer where alink between two nodes exists if they have at least one interestin common Examining commonneighbors of a pair of nodeson Interest network layer we can predict future encounters towhich the transfer of information may arise Several works oncoauthorship or collaboration networks demonstrated thatthe probability of two nodes being connected by a link ishigher when the nodes in question have common neighborsIn [34] for example a network of scientific collaborations

was analyzed showing that examining coauthors of authorshelps in predicting future collaborations Since coauthorshipnetworks are networks of scientific collaborations and hencenetworks of shared scientific interests we chose Interestnetwork layer for link prediction assuming that scientificinterest networks and more general interest networks have asimilar behavior SORSI computes the link predictor 119871119875(119894 119895)of a possible future collaboration between node 119894 and node 119895as a common neighbor measure based on Jaccard coefficient

119871119875 (119894 119895) =10038161003816100381610038161003816119872119894 cap119872119895

1003816100381610038161003816100381610038161003816100381610038161003816119872119894 cup11987211989510038161003816100381610038161003816

(6)

where119872119894 is the number of nodes in 119894rsquos range and 119872119895 is thenumber of nodes in 119895rsquos range

For each measure SORSI determines the utility score ofnode 119894 for delivering a message to node 119889 compared to node119895 as follows

119862119878 (119894 119895) =119862119862119863119890119892119903119890119890 (119894)

119862119862119863119890119892119903119890119890 (119894) + 119862119862119863119890119892119903119890119890 (119895)(7)

119879119878119878 (119894 119895 119889) = 119879119878119879119874119879 (119894 119889)119879119878119879119874119879 (119894 119889) + 119879119878119879119874119879 (119895 119889)

(8)

119871119875119878 (119894 119895 119889) = 119871119875 (119894 119889)119871119875 (119894 119889) + 119871119875 (119895 119889) (9)

The SORSI social metric is given by the combination of thecontributing score values as follows

119872119871119878 (119894 119895 119889)

= 119862119878 (119894 119895) [1 + 119879119878119878 (119894 119895 119889) + 119871119875119878 (119894 119895 119889)](10)


(1) procedure encounterNode(119872119894)(2) exchangeCentralityValues()(3) exchangeOnlineContactsLists()(4) exchangeInterestNodeList()(5) exchangeForwardingHistoryList()(6) for every messagem in message buffer do(7) D larr997888 119898destination()(8) myMLS larr997888 computeMLScore()(9) encounterMLS larr997888 computePeerMLScore()(10) if 119890119899119888119900119906119899119905119890119903119872119871119878 ge 119898119910119872119871119878 || 119872119894==119863 then(11) forwardMessage(119898119872119894)(12) end if(13) end for(14) for every messagem in forwarding history list do(15) if 119905119894119898119890119904119905119886119898119901 gt last encounter with119872119894 ampamp 119898119904119892119878119900119906119903119888119890119868119863==119898119910119868119863 then(16) if last encounter with 119891119900119903119908119886119903119889119890119903119868119863 gt last encounter with119872119894 ampamp 119891119900119903119908119886119903119889119890119903119868119863receivedMessage(119898119904119892119868119863)==false

then(17) 119864119878119878 larr997888 119891119900119903119908119886119903119889119890119903119868119863selfishScore()(18) 119891119900119903119908119886119903119889119890119903119868119863setSelfishScore(119864119878119878+1)(19) end if(20) end if(21) end for(22) end procedure

Algorithm 1 SORSI message forwarding with selfishness detection

(1) procedure receiveMessage(119898119872119894)(2) if 119872119894 = 119898source() then(3) 119864119878119878 larr997888 119872119894selfishScore()(4) 119872119894setSelfishScore(119864119878119878-1)(5) receive(119898)(6) else(7) if 119872119894selfishScore() lt 119905ℎ119903 then(8) receive(119898)(9) else(10) discard(119898)(11) end if(12) end if(13) end procedure

Algorithm 2 SORSI message reception with selfishness detection

As can be observed MLS captures the relay significance of anodewhen compared to an encountered node across all socialnetwork layers in terms of centrality tie strength and linkpredictor Note also that node centrality is considered as thepredominant factor in message forwarding Both tie strengthand tie predictor utility scores are weighted with centralityutility score and then added to centrality utility score Inthat way tie strength and link predictor utility scores willreflect the centrality utility score (eg high low or medium)between the sender node and the encountered node

The forwarding process in SORSI is given by Algorithm 1The two nodes having an encounter exchange also theirforwarding history lists containing the forwarding activityof the encountered nodes Each past forwarding included

in the list of a node is logged through the ID of theencountered node forwarding the message to that node themessage ID the source ID of the forwarded message andthe time when the message has been received A historylist is thus implemented as a list containing [forwarderIDmsgID msgSourceID timestamp] tuples Then after havingcomputed the MLS metric to decide whether forwardingmessages to the encountered node 119873 is according to itssociality and selfishness the node checks the encounterednode forwarding history list to detect eventual selfish nodesNote that the messages in the buffer are received messagesthat have successfully passed the selfishness check If the nodedetects another node as selfish (ie 119873rsquos history does notcontain certain messages of the detecting node that shouldhave been passed by the selfish node) the detecting nodeincrements the encounter-based selfishness score 119864119878119878 forthe selfish node Similarly when a node receive (and willpotentially forward) a message for which it is not the source(see Algorithm 2) its altruism is awarded decrementing itsselfish score In this way an incentive mechanism to forwardmessages is realized Nodes do not receive the messages ofnodes that have a selfish score greater than a thresholdConsequently according to SORSI scheme nodes that havebeen labeled as selfish are obliged to forward the other nodesrsquomessages for forwarding their own messages

5 SORSI Performance Evaluation

51 Simulation Setup We validate our proposal againthrough the Opportunistic Network Environment (ONE)simulator using SASSY and LAPLAND as human mobilitytraces In simulations we compare SORSI to its version


LAPLANDSASSY

25 50 75 1000Selfishness Threshold

0

05

1

Del

iver

y Ra

tio

Figure 4 The effect of selfishness threshold on message delivery

without selfishness detection namely SORSI-NS in order toquantify the improvements brought by selfishness detectionand the mechanism incentivizing participation to EpidemicRouting Bubble Rap and SPRINT-SELF For this analysiswe do not analyze anymore Spray amp Wait since we onlyfocus on social-based schemes considering that we intendto demonstrate the improvements brought to opportunisticmobile social routing by SORSI However we still considerEpidemic Routing to have an upper bound on messagedelivery

As previously specified SPRINT-SELF protocol is social-based and employs a selfishness detection mechanism dis-couraging node selfishness Here we briefly describe itsfeatures Similarly to SORSI SPRINT-SELF uses social net-work information future node behavior prediction and aselfishness detection mechanism This is the reason why wechose to compare SORSI to this protocol However differ-ently to SORSI SPRINT-SELF does not consider a multilayersocial structure to model the opportunistic network andmakes use of communities knowledge Communities sincemost of the nodes are more likely to interact with themembers of its own community [20] are a fundamentalpart of this protocol They are both computed on-the-flythrough community detection algorithms such as k-Cliqueor through the social communitiesgroups directly extractedby online social networks (eg Facebook groups) Howeverin a recent study [18] we demonstrate that not alwaysonline communities correspond to the offline ones As suchonline community information is not always usable and maylead to suboptimal forwarding paths Moreover due to thecomputation of communities through community detectionalgorithms and of a utility value taking into account severalinformation on contact and message exchange history thecost of computing this utility value is sensibly higher thanSORSI MLS value Specifically SPRINT-SELF uses a utilityfunction which uses two parts one of

For this protocol when a node is labeled as selfish wesuppose its battery level is greater than a tolerance threshold(eg for 2 of battery level a node is not considered selfish ifit does not forward messages while for 30 of battery level itis)

For our analyses we use three opportunistic performancemetrics First we study again the system throughput Then westudy the system cost or overhead cost which measures thenumber of packets transmitted across the air divided by thenumber of unique packets created This performance indexis important for our study since the duplicated messages

injected into the network consume resources such as batteryand memory Finally similarly to the previous analysis westudy the system delay

As first experiment through a uniform distribution werandomly choose a subset of network nodes labeling thesenodes as selfish We consider again 25 50 and 75 ofselfish nodes and the limit case when all nodes are selfishhypothesizing that selfishness is driven by one or more ofthe three motivations described in Section 3 For EpidemicRouting Bubble Rap and SORSI-NS a selfish node alwaysdrops the messages for which it is not the destination Werepeated each simulation 20 times and took the average ofeach run as a result The values of the main parametersused in our simulations are shown in Table 4 Simulatingdifferent selfishness thresholds (see Figure 4) we have chosena threshold equal to 50 for which we obtained the best SORSIdelivery performance in both datasets As expected as thethreshold increases being less restrictive with selfish nodesthe delivery ratio starts decreasing

As second experiment we test and compare the effect ofenergy consumption and buffer occupancy on node selfish-ness and hence on opportunistic routing For each node weconsider the energy consumption model adopted

52 Results We start evaluating our proposal by analyzingthe SASSY mobility trace for the first experiment where werandomly choose a node as selfish Delivery ratio overheadcost and average latency for this trace are shown in Figure 5As expected when the proportion of nodes acting selfishlyin the network increases the delivery ratio decreases for allthe routing protocols considered Similarly the increase inthe percentage of selfish nodes degrades the average latencywhile the overhead cost decreases since the selfish behaviorsresult in a lower number of forwarded packets HoweverEpidemic Routing is more influenced by the proportionof selfish nodes showing a more rapid degradation of itsrouting performance indexes as the percentage of selfishnodes present within the network increases We further notethat when all nodes are selfish the protocols deliver the samenumber of messages resulting in equal costs and delays Thishappens because every node forwards only its own messagesWhen it is present there is a certain percentage of selfishnodes (25 50 75) on the contrary Epidemic Routingperforms the best achieving the highest delivery ratio valuesfollowed by SORSI which is able to outperform all the othersocial-based schemes (its classic version SORSI-NS BubbleRap and SPRINT-SELF) However SORSI is characterized bydelivery ratios and average latencies comparable to EpidemicRouting with a much lower overhead cost This result showsthat SORSI is able to manage node selfishness while sensiblylimiting the number of message replicas and hence theenergy consumption avoiding the development of selfishbehaviors Moreover SORSI with 50 of selfish nodes forexample is able to achieve almost the same delivery ratio ofBubble Rap with no selfish nodes Its multilayer social metrictogether with selfish management is thus able to achievecomparable routing performance of a reference social-basedprotocol with no selfish nodes Compared to its version


05

048

046

044

042

04

038

036

034

032

Del

iver

y Ra

tio

0 25 50 75 100

of selfish nodes

16

14

12

10

8

6

4

2

0

Ove

rhea

d C

ost

3600

3400

3200

3000

2800

2600

2400

2200

2000

1800

Aver

age L

aten

cy [s

]

EpidemicBubble RapSORSIndashNSSORSISPRINTndashSELF

0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


Figure 5 SORSI routing performances compared to Epidemic Routing Bubble Rap SORSI-NS and SPRINT-SELF for the SASSY humanmobility trace

Table 5 Protocols performance when nodes have 50 of residual energy (SASSY)

Epidemic Bubble Rap SORSI-NS SORSI SPRINT-SELFDelivery Ratio 0412 0317 0331 0397 0375Overhead Cost 7121 3136 1951 1875 1056Average Latency 240654 322935 2451779 2394832 2398569

Table 6 Protocols performance when nodes have 50 of residual energy (LAPLAND)

Epidemic Bubble Rap SORSI-NS SORSI SPRINT-SELFDelivery Ratio 0151 0091 0129 0147 0137Overhead Cost 5162 4125 3165 1551 2945Average Latency 1147sdot105 1211 sdot105 1171 sdot105 1145 sdot105 1149 sdot105



without selfishness detection we observe that SORSI is ableto reduce the effect of node selfishness behaving similarly toSORSI-NS with a significantly lower number of selfish nodes

Figure 6 shows the results obtained for the LAPLANDmobility trace This trace covers a shorter experimentalperiod and has a smaller number of nodes compared toSASSY As such we chose a shorter TTL of 4 minutesSimilarly to SASSY the increase in the proportion of selfishnodes results in a lower delivery ratio for all the protocolsconsidered Again SORSI outperforms Bubble Rap SORSI-NS and SPRINT-SELF achieving a message delivery similarto Epidemic Routing We thus conclude that also in a smallerdataset this feature is present The overhead cost trendsconfirm that SORSI is able to reduce the number of message

replicas achieving the lowest costs without compromising itsefficiency in terms of end-to-end delay

In Table 5 SORSI outperforms SORSI-NS and BubbleRap with performance close to Epidemic Routing in terms ofdelivery ratio Also average latency is lower in SORSI in com-parison with the other diffusion strategies However over-head cost of SORSI is higher than SPRINT-SELFThis meansthat the number of node involved in data dissemination isgrater in SORSI and SORSI-NS than SPRINT-SELFThe sametrend can be observed in Table 6 where SORSI in LAPLANDscenario outperforms all data diffusion techniques in allperformance metrics When the number of nodes switchingin selfish behavior increases such as that in Tables 7 and 8the performancemetrics of delivery ratio and average latency


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105

Figure 6 SORSI routing performances compared to Epidemic Routing Bubble Rap SORSI-NS and SPRINT-SELF for the LAPLANDhumanmobility trace



Table 9 Protocols performance when nodes have 50 of full buffer (SASSY)


Table 10 Protocols performance when nodes have 50 of full buffer (LAPLAND)


are always better than other strategies However the overheadcost of SORSI is worst in SASSY and it is better in LAPLANDscenarioThis means that when few TTLs are allowed and thescenario is small SORSI does not have enough time to involvemore nodes in data dissemination Similar considerations canbe applied when buffer occupancy threshold is used It canaffect the selfish behavior such as presented in Tables 9ndash12In these cases SORSI always outperforms in delivery ratioand average latency the other data dissemination strategieswith the exception of Epidemic Routing that is considered asbenchmark for the delivery ratioMoreover the overhead costslightly increases in the SASSY scenario where SPRINT-SELFis more performing than SORSI and SORSI-NS

6 Conclusions

In this paper we have presented a routing mechanism formitigating the effect of node selfishness on opportunisticnetwork routing First we have delineated the main reasonsfor which an opportunistic node may act selfishly andevaluated the effects of node selfishness on opportunisticrouting Then we have proposed SORSI a selfishness-awareopportunistic routing scheme using multilayer social net-work information and the history of forwarding for drivingrouting decisions Simulating two real-world mobility traceswe have demonstrated that SORSI is able to outperformBubble Rap SORSI version without selfishness detection






and SPRINT-SELF achieving epidemic delivery ratios withthe lowest overhead cost Moreover SORSI outperformsSPRINT-SELF and Bubble Rap also in cases where bufferthresholds or energy thresholds can affect the selfish behaviorincreasing or decreasing the number of nodes acting selfishlyIn these cases SORSI continues to perform better in termsof delivery ratio while reducing the average latency incomparison with the other social-aware data disseminationtechniques As a result SORSI is able to increase the numberof nodes cooperating within the opportunistic network whensome nodes have an initial attitude to act selfishly

In future work we plan to further validate SORSIextending the simulations to other datasets investigating alsothe impact of other parameters like TTL and the messagegeneration rate We also intend to analyze the theoreticalreasons behind SORSI performance

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

References

[1] K Fall ldquoA delay-tolerant network architecture for challengedinternetsrdquo in Proceedings of the 2003 conference on Applicationstechnologies architectures and protocols for computer communi-cations SIGCOMM rsquo03 pp 27ndash34 ACM New York NY USA2003

[2] V Cerf S Burleigh A Hooke et al ldquoDelay-tolerant net-working architecturerdquo 2007 httptoolsietforghtmlrfc4838httptoolsietforghtmlrfc4838

[3] A V Vasilakos Y Zhang and T Spyropoulos Delay tolerantnetworks Protocols and applications CRC Press 2012

[4] L Pelusi A Passarella and M Conti ldquoOpportunistic network-ing data forwarding in disconnected mobile ad hoc networksrdquoIEEE Communications Magazine vol 44 no 11 pp 134ndash1412006

[5] C Boldrini M Conti and A Passarella ldquoAutonomic behaviourof opportunistic network routingrdquo International Journal of

Autonomous and Adaptive Communications Systems vol 1 no1 pp 122ndash147 2008

[6] A Socievole E Yoneki F De Rango and J Crowcroft ldquoML-SOR Message routing using multi-layer social networks inopportunistic communicationsrdquo Computer Networks vol 81pp 201ndash219 2015

[7] F De Rango A Socievole and S Marano ldquoExploiting onlineand offline activity-based metrics for opportunistic forward-ingrdquoWireless Networks pp 1ndash17 2014

[8] K Xu P Hui V O K Li J Crowcroft V Latora and PLio ldquoImpact of altruism on opportunistic communicationsrdquoin Proceedings of the 2009 1st International Conference onUbiquitous and FutureNetworks ICUFN2009 pp 153ndash158 June2009

[9] P Hui K Xu V O K Li J Crowcroft V Latora and P LioldquoSelfishness altruism and message spreading in mobile socialnetworksrdquo inProceedings of the IEEE INFOCOMWorkshops pp1ndash6 IEEE April 2009

[10] Y Li P Hui D Jin L Su and L Zeng ldquoEvaluating the impactof social selfishness on the epidemic routing in delay tolerantnetworksrdquo IEEE Communications Letters vol 14 no 11 pp1026ndash1028 2010

[11] A Mei and J Stefa ldquoGive2Get Forwarding in social mobilewireless networks of selfish individualsrdquo IEEE Transactions onDependable and Secure Computing vol 9 no 4 pp 569ndash5822012

[12] A Mtibaa M May C Diot and M Ammar ldquoPeopleRankSocial opportunistic forwardingrdquo in Proceedings of the IEEEINFOCOM 2010 pp 1ndash5 2010

[13] P Hui J Crowcroft and E Yoneki ldquoBUBBLE Rap social-basedforwarding in delay-tolerant networksrdquo IEEE Transactions onMobile Computing vol 10 no 11 pp 1576ndash1589 2011

[14] A Socievole F De Rango and S Marano ldquoFace-to-face withfacebook friends Using online friendlists for routing in oppor-tunistic networksrdquo in Proceedings of the 2013 IEEE 24th AnnualInternational Symposium on Personal Indoor andMobile RadioCommunications PIMRC 2013 pp 2989ndash2994 September 2013

[15] S Gaito E Pagani and G P Rossi ldquoStrangers help friends tocommunicate in opportunistic networksrdquo Computer Networksvol 55 no 2 pp 374ndash385 2011

[16] G Bigwood and T Henderson ldquoBootstrapping opportunisticnetworks using social rolesrdquo in Proceedings of the 2011 IEEEInternational Symposium on a World of Wireless Mobile andMultimedia Networks WoWMoM 2011 pp 1ndash6 June 2011

[17] R I Ciobanu C Dobre and V Cristea ldquoSPRINT Socialprediction-based opportunistic routingrdquo in Proceedings of 7th


IEEE WoWMoM workshop on autonomic and opportunisticcommunications (AOC 2013) pp 1161ndash1166 Madrid Spain2013

[18] A Socievole F De Rango and A Caputo ldquoOpportunisticmobile social networks From mobility and Facebook friend-ships to structural analysis of user social behaviorrdquo ComputerCommunications vol 87 pp 1ndash18 2016

[19] A Socievole F De Rango A Caputo and S Marano ldquoSimulat-ing node selfishness in opportunistic networksrdquo in Proceedingsof the 2016 International Symposium on Performance Evaluationof Computer and Telecommunication Systems (SPECTS 2016)pp 1ndash6 July 2016

[20] R I Ciobanu C Dobre V Cristea and D Al-Jumeily ldquoSocialaspects for opportunistic communicationrdquo in Proceedings ofthe 11th International Symposium on Parallel and DistributedComputing (ISPDC rsquo12) pp 251ndash258 IEEE June 2012

[21] M Berlingerio M Coscia F Giannotti A Monreale and DPedreschi ldquoFoundations of multidimensional network anal-ysisrdquo in Proceedings of the 2011 International Conference onAdvances in Social Networks Analysis and Mining ASONAM2011 pp 485ndash489 July 2011

[22] A Lupia and F De Rango ldquoEvaluation of the energy consump-tion introduced by a trust management scheme on mobile ad-hoc networksrdquo Journal of Networks vol 10 no 4 p 240 2015

[23] E Hernandez-Orallo M D S Olmos J-C Cano C TCalafate and P Manzoni ldquoCoCoWa a collaborative contact-based watchdog for detecting selfish nodesrdquo IEEE Transactionson Mobile Computing vol 14 no 6 pp 1162ndash1175 2015

[24] S Ahmed and S S Kanhere ldquoHubcode message forwardingusing hub-based network coding in delay tolerant networksrdquoin Proceedings of the 12th ACM international conference onModeling analysis and simulation of wireless andmobile systemsMSWiM rsquo09 pp 288ndash296 ACM New York NY USA October2009 httpdoiacmorg10114516418041641853

[25] A Urpi M Bonuccelli and S Giordano ldquoModelling coop-eration in mobile ad hoc newtorks a formal description ofselfishness WiOpT03 Modeling and Optimization in MobilerdquoAd Hoc and Wireless Networks 2003

[26] A Keranen J Ott and T Karkkainen ldquoThe ONE simula-tor for DTN protocol evaluationrdquo in Proceedings of the 2ndInternational Conference on Simulation Tools and Techniques(Simutools rsquo09) pp 1ndash55 ICST Brussels Belgium March 2009httpdxdoiorg104108ICSTSIMUTOOLS20095674

[27] G Bigwood D RehunathanM Bateman T Henderson and SBhatti CRAWDAD trace set st andrewssassymobile (v 2011-06-03) httpcrawdadcsdartmouthedust andrewssassymobile 2011

[28] E Yoneki and F B Abdesslem ldquoFinding a data blackhole inbluetooth scanningrdquo ExtremeCom 2009

[29] A Vahdat and D Becker ldquoEpidemic routing for Partially-Connected ad hoc networksrdquo Tech Rep DukeUniversity April2000

[30] F De Rango S Amelio and P Fazio ldquoEnhancements ofepidemic routing in delay tolerant networks from an energyperspectiverdquo inProceedings of the 2013 9th InternationalWirelessCommunications andMobile Computing Conference (IWCMC)pp 731ndash735 Italy July 2013

[31] T Spyropoulos K Psounis and C S Raghavendra ldquoSprayand wait an efficient routing scheme for intermittentlyconnected mobile networksrdquo in Proceedings of the 2005ACM SIGCOMM workshop on Delay-tolerant networking

(WDTN rsquo05) pp 252ndash259 ACM New York NY USA 2005httpdoiacmorg10114510801391080143

[32] G Palla I Derenyi I Farkas and T Vicsek ldquoUncoveringthe overlapping community structure of complex networks innature and societyrdquoNature vol 435 no 7043 pp 814ndash818 2005

[33] F De Rango A Socievole A Scaglione and S Marano ldquoNovelactivity-based metrics for efficient forwarding over onlineand detected social networksrdquo in Proceedings of the 2013 9thInternational Wireless Communications and Mobile ComputingConference (IWCMC) July 2013

[34] M Newman ldquoClustering and preferential attachment in grow-ing networksrdquo Physical Review E-Statistical Nonlinear and SoftMatter Physics vol 64 no 2 pp 251021ndash251024 2003

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018


Active and Passive Electronic Components

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of



Journal ofEngineeringVolume 2018

SensorsJournal of



RotatingMachinery


Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation


Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom


time a packet can be buffered novel forwarding strategies anda specific bundle layer need to be designed to guarantee ahop-by-hop data delivery Moreover a stronger store-carry-forward approach needs to be applied for each node andall nodes form a DTN domain due to their capability totolerate data delivery delay In this context opportunisticnetworking and forwarding can be a useful and mandatoryapproach to offer to each node an opportunity to send datapacket freeing space in its buffer [8ndash11] The opportunisticapproach assumes that nodes that want to transmit can usesome specific metric to select the next hop where forwardingthe data and the receiving node is assumed that want toreceive the data packet sent by node encountered Howeverin these last years the second assumption is not alwaysconsidered because it has been observed as in the real lifenodes can be also selfish instead of cooperative Becausesome resources are limited such as buffer size energy budgetor other constraints mobile nodes cannot see benefits inparticipating in a collaborative way to the communicationThis selfish behavior can also be evident not at the beginningof the communication but during the network dynamic onthe basis of the constraints and budgets that can be exhaustedIn this case if some countermeasures are not adopted thereis the risk that after some period some nodes stop to forwarddata packet increasing the data delay reducing the data packetdelivery ratio and losing the advantage of a flexible paradigmsuch as DTNs

This paper proposes a routingmechanism for opportunis-tic networks able to mitigate the effect of node selfishnessExploiting the social-based nature of human encounters wepropose a selfishness-aware opportunistic routing schemeable to maximize message delivery As a matter of factmessage dissemination in opportunistic networks can beefficiently performed using social-based routing metricsSeveral works such as [6 7 12ndash17] demonstrate that thenodes with a highly social behavior are good carriers andare thus able to improve message delivery However howto extract node sociality and define efficient routing metricsis still challenging First opportunistic networks are highlydynamic and it is not easy to reconstruct a nodersquos socialbehavior Second when representing the network as a socialgraph several metrics are able to define the sociality and thesocial position of a node (see for example the social metricsanalyzed in [18]) Each of these metrics is not always easilycomputable in a distributed way on a dynamic graph Thirdeach node may interact in different social contexts like realworld when moving online world using social networkingwebsites and so on As such each node has a sociality thatcan be defined ldquomultilayerrdquo [18]

Starting from the above considerations to select aneffective forwarding node the routing scheme we proposenamed SORSI (Social-based Opportunistic Routing withSelfishness detection and Incentive mechanisms) measuresthe forwarding capability of a node when compared toan encountered node in terms of node centrality (ie theimportanceposition of a node within a social graph) tiestrength and link prediction Each of these routing metricsis computed in a distributed way on a particular socialnetwork layer that describes a social dimension Since each

mobile user may have social relationships on several socialdimensionslayers (eg real life Facebook Twitter etc) thisprotocol exploits the user multilayer sociality for computingforwarding paths and improving message disseminationMoreover it adds a selfishness detection mechanism basedon the history of message exchanges for incentivizing theforwarding of packets Starting from the work in [19] wherethe effect of selfish nodes in the network performance hasbeen presented in this paper we propose a mechanism todetect selfish nodes and to discourage their behavior Thismechanism has been designed and implemented in a multi-layer social routing in order to test its effectiveness and it hasbeen compared with other well-known data disseminationtechniques for opportunistic networks So the contributionsof this paper can be summarized as follows

(i) We characterize the situations in which an oppor-tunistic network node may act selfishly

(ii) We analyze the effect of node selfishness in oppor-tunistic networks through simulations

(iii) We propose a multilayer social-based routing schemewith selfishness management and compare it to threereference opportunistic routing protocols over twoexperimental datasets of human mobility with differ-ent connectivity patterns

(iv) We show that in general as the proportion of nodesacting selfishly in the network increases the messagedelivery ratio decreases and the average end-to-endlatency increases

(v) We show that the use of multilayer social networkinformation together with selfishness detection isable to achieve delivery ratios and average latenciescomparable to epidemic delivery with a much loweroverhead cost

The paper has been organized as follows Section 2provides background information Section 3 discusses theselfish node problem in opportunistic networks and analyzesits effect on routing through simulations Section 4 describesour proposal SORSI In Section 5 we detail the simulationsetup for SORSI evaluation and discuss the results FinallySection 6 concludes the paper summarizing the resultsobtained and discussing the future research directions

2 Related Works

Many works in literature in these last years are focusing onthe selfish behavior and on its effect on network performanceHowever a few contributions till now have been relatedto DTNs and ONs because for these networks it is morechallenging to detect malicious behavior and it is not easyto propose effective incentive mechanism that can tradeoffbetween network performance and resource draining

21 Selfishness Management in MANET In MANET anddistributed wireless networks some monitoring techniquesto detect malicious behavior or selfish nodes are applied Oneof the most adopted techniques is the watchdog technique









A

B

C

I amselfish





























0

02

04D

eliv

ery

Ratio




times 104


0

5

10

15

Aver

age L

aten

cy [s

]





0

02

04

Del

iver

y Ra

tio




times 104

0

5

10

15

Aver

age L

aten

cy [s

]






0

01

02

Del

iver

y Ra

tio




times 104

0

5

10

15

Aver

age L

aten

cy [s

]










02

04

06

08D

eliv

ery

Ratio




0

2000

4000

6000

Aver

age L

aten

cy [s

]






02

03

04

05

Del

iver

y Ra

tio





0

2000

4000

6000

Aver

age L

aten

cy [s

]




02

03

04

05

Del

iver

y Ra

tio





0

2000

4000

6000

Aver

age L

aten

cy [s

]
















119862119863119890119892119903119890119890 (119894 119905) =119872119894

sum119895=1

119890 (119894 119895 119905) (1)

where

119890 (119894 119895 119905) =

1 119894119891 119894 119890119899119888119900119906119899119905119890119903119904 119895 119889119906119903119894119899119892 119905119894119898119890 119904119897119900119905 1199050 119900119905ℎ119890119903119908119894119904119890

(2)


119862119862119863119890119892119903119890119890 (119894) =1119879119879

sum119905=0

119862119863119890119892119903119890119890 (119894 119879 minus 119905) (3)













1199082 03k (k-Clique) 3



119879119878 (119894 119895 119897)

=

1 119894119891 119894 119886119899119889 119895 119886119903119890 119888119900119899119899119890119888119905119890119889 119886119905 119897119886119910119890119903 1198970 119900119905ℎ119890119903119908119894119904119890

(4)


119879119878119879119874119879 (119894 119895) =119871

sum119897=1

119879119878 (119894 119895 119897) (5)




119871119875 (119894 119895) =10038161003816100381610038161003816119872119894 cap119872119895

1003816100381610038161003816100381610038161003816100381610038161003816119872119894 cup11987211989510038161003816100381610038161003816

(6)



119862119878 (119894 119895) =119862119862119863119890119892119903119890119890 (119894)

119862119862119863119890119892119903119890119890 (119894) + 119862119862119863119890119892119903119890119890 (119895)(7)

119879119878119878 (119894 119895 119889) = 119879119878119879119874119879 (119894 119889)119879119878119879119874119879 (119894 119889) + 119879119878119879119874119879 (119895 119889)

(8)

119871119875119878 (119894 119895 119889) = 119871119875 (119894 119889)119871119875 (119894 119889) + 119871119875 (119895 119889) (9)


119872119871119878 (119894 119895 119889)

= 119862119878 (119894 119895) [1 + 119879119878119878 (119894 119895 119889) + 119871119875119878 (119894 119895 119889)](10)













LAPLANDSASSY


0

05

1

Del

iver

y Ra

tio











05

048

046

044

042

04

038

036

034

032

Del

iver

y Ra

tio

0 25 50 75 100

of selfish nodes

16

14

12

10

8

6

4

2

0

Ove

rhea

d C

ost

3600

3400

3200

3000

2800

2600

2400

2200

2000

1800

Aver

age L

aten

cy [s

]


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes














0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105









6 Conclusions









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia










A

B

C

I amselfish





























0

02

04D

eliv

ery

Ratio




times 104


0

5

10

15

Aver

age L

aten

cy [s

]





0

02

04

Del

iver

y Ra

tio




times 104

0

5

10

15

Aver

age L

aten

cy [s

]






0

01

02

Del

iver

y Ra

tio




times 104

0

5

10

15

Aver

age L

aten

cy [s

]










02

04

06

08D

eliv

ery

Ratio




0

2000

4000

6000

Aver

age L

aten

cy [s

]






02

03

04

05

Del

iver

y Ra

tio





0

2000

4000

6000

Aver

age L

aten

cy [s

]




02

03

04

05

Del

iver

y Ra

tio





0

2000

4000

6000

Aver

age L

aten

cy [s

]
















119862119863119890119892119903119890119890 (119894 119905) =119872119894

sum119895=1

119890 (119894 119895 119905) (1)

where

119890 (119894 119895 119905) =

1 119894119891 119894 119890119899119888119900119906119899119905119890119903119904 119895 119889119906119903119894119899119892 119905119894119898119890 119904119897119900119905 1199050 119900119905ℎ119890119903119908119894119904119890

(2)


119862119862119863119890119892119903119890119890 (119894) =1119879119879

sum119905=0

119862119863119890119892119903119890119890 (119894 119879 minus 119905) (3)













1199082 03k (k-Clique) 3



119879119878 (119894 119895 119897)

=

1 119894119891 119894 119886119899119889 119895 119886119903119890 119888119900119899119899119890119888119905119890119889 119886119905 119897119886119910119890119903 1198970 119900119905ℎ119890119903119908119894119904119890

(4)


119879119878119879119874119879 (119894 119895) =119871

sum119897=1

119879119878 (119894 119895 119897) (5)




119871119875 (119894 119895) =10038161003816100381610038161003816119872119894 cap119872119895

1003816100381610038161003816100381610038161003816100381610038161003816119872119894 cup11987211989510038161003816100381610038161003816

(6)



119862119878 (119894 119895) =119862119862119863119890119892119903119890119890 (119894)

119862119862119863119890119892119903119890119890 (119894) + 119862119862119863119890119892119903119890119890 (119895)(7)

119879119878119878 (119894 119895 119889) = 119879119878119879119874119879 (119894 119889)119879119878119879119874119879 (119894 119889) + 119879119878119879119874119879 (119895 119889)

(8)

119871119875119878 (119894 119895 119889) = 119871119875 (119894 119889)119871119875 (119894 119889) + 119871119875 (119895 119889) (9)


119872119871119878 (119894 119895 119889)

= 119862119878 (119894 119895) [1 + 119879119878119878 (119894 119895 119889) + 119871119875119878 (119894 119895 119889)](10)













LAPLANDSASSY


0

05

1

Del

iver

y Ra

tio











05

048

046

044

042

04

038

036

034

032

Del

iver

y Ra

tio

0 25 50 75 100

of selfish nodes

16

14

12

10

8

6

4

2

0

Ove

rhea

d C

ost

3600

3400

3200

3000

2800

2600

2400

2200

2000

1800

Aver

age L

aten

cy [s

]


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes














0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105









6 Conclusions









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




















0

02

04D

eliv

ery

Ratio




times 104


0

5

10

15

Aver

age L

aten

cy [s

]





0

02

04

Del

iver

y Ra

tio




times 104

0

5

10

15

Aver

age L

aten

cy [s

]






0

01

02

Del

iver

y Ra

tio




times 104

0

5

10

15

Aver

age L

aten

cy [s

]










02

04

06

08D

eliv

ery

Ratio




0

2000

4000

6000

Aver

age L

aten

cy [s

]






02

03

04

05

Del

iver

y Ra

tio





0

2000

4000

6000

Aver

age L

aten

cy [s

]




02

03

04

05

Del

iver

y Ra

tio





0

2000

4000

6000

Aver

age L

aten

cy [s

]
















119862119863119890119892119903119890119890 (119894 119905) =119872119894

sum119895=1

119890 (119894 119895 119905) (1)

where

119890 (119894 119895 119905) =

1 119894119891 119894 119890119899119888119900119906119899119905119890119903119904 119895 119889119906119903119894119899119892 119905119894119898119890 119904119897119900119905 1199050 119900119905ℎ119890119903119908119894119904119890

(2)


119862119862119863119890119892119903119890119890 (119894) =1119879119879

sum119905=0

119862119863119890119892119903119890119890 (119894 119879 minus 119905) (3)













1199082 03k (k-Clique) 3



119879119878 (119894 119895 119897)

=

1 119894119891 119894 119886119899119889 119895 119886119903119890 119888119900119899119899119890119888119905119890119889 119886119905 119897119886119910119890119903 1198970 119900119905ℎ119890119903119908119894119904119890

(4)


119879119878119879119874119879 (119894 119895) =119871

sum119897=1

119879119878 (119894 119895 119897) (5)




119871119875 (119894 119895) =10038161003816100381610038161003816119872119894 cap119872119895

1003816100381610038161003816100381610038161003816100381610038161003816119872119894 cup11987211989510038161003816100381610038161003816

(6)



119862119878 (119894 119895) =119862119862119863119890119892119903119890119890 (119894)

119862119862119863119890119892119903119890119890 (119894) + 119862119862119863119890119892119903119890119890 (119895)(7)

119879119878119878 (119894 119895 119889) = 119879119878119879119874119879 (119894 119889)119879119878119879119874119879 (119894 119889) + 119879119878119879119874119879 (119895 119889)

(8)

119871119875119878 (119894 119895 119889) = 119871119875 (119894 119889)119871119875 (119894 119889) + 119871119875 (119895 119889) (9)


119872119871119878 (119894 119895 119889)

= 119862119878 (119894 119895) [1 + 119879119878119878 (119894 119895 119889) + 119871119875119878 (119894 119895 119889)](10)













LAPLANDSASSY


0

05

1

Del

iver

y Ra

tio











05

048

046

044

042

04

038

036

034

032

Del

iver

y Ra

tio

0 25 50 75 100

of selfish nodes

16

14

12

10

8

6

4

2

0

Ove

rhea

d C

ost

3600

3400

3200

3000

2800

2600

2400

2200

2000

1800

Aver

age L

aten

cy [s

]


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes














0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105









6 Conclusions









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




02

04

06

08D

eliv

ery

Ratio




0

2000

4000

6000

Aver

age L

aten

cy [s

]






02

03

04

05

Del

iver

y Ra

tio





0

2000

4000

6000

Aver

age L

aten

cy [s

]




02

03

04

05

Del

iver

y Ra

tio





0

2000

4000

6000

Aver

age L

aten

cy [s

]
















119862119863119890119892119903119890119890 (119894 119905) =119872119894

sum119895=1

119890 (119894 119895 119905) (1)

where

119890 (119894 119895 119905) =

1 119894119891 119894 119890119899119888119900119906119899119905119890119903119904 119895 119889119906119903119894119899119892 119905119894119898119890 119904119897119900119905 1199050 119900119905ℎ119890119903119908119894119904119890

(2)


119862119862119863119890119892119903119890119890 (119894) =1119879119879

sum119905=0

119862119863119890119892119903119890119890 (119894 119879 minus 119905) (3)













1199082 03k (k-Clique) 3



119879119878 (119894 119895 119897)

=

1 119894119891 119894 119886119899119889 119895 119886119903119890 119888119900119899119899119890119888119905119890119889 119886119905 119897119886119910119890119903 1198970 119900119905ℎ119890119903119908119894119904119890

(4)


119879119878119879119874119879 (119894 119895) =119871

sum119897=1

119879119878 (119894 119895 119897) (5)




119871119875 (119894 119895) =10038161003816100381610038161003816119872119894 cap119872119895

1003816100381610038161003816100381610038161003816100381610038161003816119872119894 cup11987211989510038161003816100381610038161003816

(6)



119862119878 (119894 119895) =119862119862119863119890119892119903119890119890 (119894)

119862119862119863119890119892119903119890119890 (119894) + 119862119862119863119890119892119903119890119890 (119895)(7)

119879119878119878 (119894 119895 119889) = 119879119878119879119874119879 (119894 119889)119879119878119879119874119879 (119894 119889) + 119879119878119879119874119879 (119895 119889)

(8)

119871119875119878 (119894 119895 119889) = 119871119875 (119894 119889)119871119875 (119894 119889) + 119871119875 (119895 119889) (9)


119872119871119878 (119894 119895 119889)

= 119862119878 (119894 119895) [1 + 119879119878119878 (119894 119895 119889) + 119871119875119878 (119894 119895 119889)](10)













LAPLANDSASSY


0

05

1

Del

iver

y Ra

tio











05

048

046

044

042

04

038

036

034

032

Del

iver

y Ra

tio

0 25 50 75 100

of selfish nodes

16

14

12

10

8

6

4

2

0

Ove

rhea

d C

ost

3600

3400

3200

3000

2800

2600

2400

2200

2000

1800

Aver

age L

aten

cy [s

]


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes














0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105









6 Conclusions









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia









119862119863119890119892119903119890119890 (119894 119905) =119872119894

sum119895=1

119890 (119894 119895 119905) (1)

where

119890 (119894 119895 119905) =

1 119894119891 119894 119890119899119888119900119906119899119905119890119903119904 119895 119889119906119903119894119899119892 119905119894119898119890 119904119897119900119905 1199050 119900119905ℎ119890119903119908119894119904119890

(2)


119862119862119863119890119892119903119890119890 (119894) =1119879119879

sum119905=0

119862119863119890119892119903119890119890 (119894 119879 minus 119905) (3)













1199082 03k (k-Clique) 3



119879119878 (119894 119895 119897)

=

1 119894119891 119894 119886119899119889 119895 119886119903119890 119888119900119899119899119890119888119905119890119889 119886119905 119897119886119910119890119903 1198970 119900119905ℎ119890119903119908119894119904119890

(4)


119879119878119879119874119879 (119894 119895) =119871

sum119897=1

119879119878 (119894 119895 119897) (5)




119871119875 (119894 119895) =10038161003816100381610038161003816119872119894 cap119872119895

1003816100381610038161003816100381610038161003816100381610038161003816119872119894 cup11987211989510038161003816100381610038161003816

(6)



119862119878 (119894 119895) =119862119862119863119890119892119903119890119890 (119894)

119862119862119863119890119892119903119890119890 (119894) + 119862119862119863119890119892119903119890119890 (119895)(7)

119879119878119878 (119894 119895 119889) = 119879119878119879119874119879 (119894 119889)119879119878119879119874119879 (119894 119889) + 119879119878119879119874119879 (119895 119889)

(8)

119871119875119878 (119894 119895 119889) = 119871119875 (119894 119889)119871119875 (119894 119889) + 119871119875 (119895 119889) (9)


119872119871119878 (119894 119895 119889)

= 119862119878 (119894 119895) [1 + 119879119878119878 (119894 119895 119889) + 119871119875119878 (119894 119895 119889)](10)













LAPLANDSASSY


0

05

1

Del

iver

y Ra

tio











05

048

046

044

042

04

038

036

034

032

Del

iver

y Ra

tio

0 25 50 75 100

of selfish nodes

16

14

12

10

8

6

4

2

0

Ove

rhea

d C

ost

3600

3400

3200

3000

2800

2600

2400

2200

2000

1800

Aver

age L

aten

cy [s

]


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes














0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105









6 Conclusions









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia












1199082 03k (k-Clique) 3



119879119878 (119894 119895 119897)

=

1 119894119891 119894 119886119899119889 119895 119886119903119890 119888119900119899119899119890119888119905119890119889 119886119905 119897119886119910119890119903 1198970 119900119905ℎ119890119903119908119894119904119890

(4)


119879119878119879119874119879 (119894 119895) =119871

sum119897=1

119879119878 (119894 119895 119897) (5)




119871119875 (119894 119895) =10038161003816100381610038161003816119872119894 cap119872119895

1003816100381610038161003816100381610038161003816100381610038161003816119872119894 cup11987211989510038161003816100381610038161003816

(6)



119862119878 (119894 119895) =119862119862119863119890119892119903119890119890 (119894)

119862119862119863119890119892119903119890119890 (119894) + 119862119862119863119890119892119903119890119890 (119895)(7)

119879119878119878 (119894 119895 119889) = 119879119878119879119874119879 (119894 119889)119879119878119879119874119879 (119894 119889) + 119879119878119879119874119879 (119895 119889)

(8)

119871119875119878 (119894 119895 119889) = 119871119875 (119894 119889)119871119875 (119894 119889) + 119871119875 (119895 119889) (9)


119872119871119878 (119894 119895 119889)

= 119862119878 (119894 119895) [1 + 119879119878119878 (119894 119895 119889) + 119871119875119878 (119894 119895 119889)](10)













LAPLANDSASSY


0

05

1

Del

iver

y Ra

tio











05

048

046

044

042

04

038

036

034

032

Del

iver

y Ra

tio

0 25 50 75 100

of selfish nodes

16

14

12

10

8

6

4

2

0

Ove

rhea

d C

ost

3600

3400

3200

3000

2800

2600

2400

2200

2000

1800

Aver

age L

aten

cy [s

]


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes














0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105









6 Conclusions









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia














LAPLANDSASSY


0

05

1

Del

iver

y Ra

tio











05

048

046

044

042

04

038

036

034

032

Del

iver

y Ra

tio

0 25 50 75 100

of selfish nodes

16

14

12

10

8

6

4

2

0

Ove

rhea

d C

ost

3600

3400

3200

3000

2800

2600

2400

2200

2000

1800

Aver

age L

aten

cy [s

]


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes














0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105









6 Conclusions









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



05

048

046

044

042

04

038

036

034

032

Del

iver

y Ra

tio

0 25 50 75 100

of selfish nodes

16

14

12

10

8

6

4

2

0

Ove

rhea

d C

ost

3600

3400

3200

3000

2800

2600

2400

2200

2000

1800

Aver

age L

aten

cy [s

]


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes














0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105









6 Conclusions









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


0 25 50 75 100

of selfish nodes


04

035

03

025

02

015

01

005

Del

iver

y Ra

tio25

20

15

10

5

0

Ove

rhea

d C

ost

128

126

124

122

12

118

116

114

112

Aver

age L

aten

cy [s

]

times 105









6 Conclusions









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia









Data Availability




References









































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia
























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


Routing in Mobile Opportunistic Social Networks …downloads.hindawi.com › journals › wcmc ›...

Documents

Transcript of Routing in Mobile Opportunistic Social Networks …downloads.hindawi.com › journals › wcmc ›...