Boosting the Performance of Content Centric Networking using Delay Tolerant Networking Mechanisms
Hasan M A Islam, Dimitris Chatzopoulos, Dmitrij Lagutin, Pan Hui, Antti Ylä-Jääski
11 Boosting the Performance of Content CentricNetworking using Delay Tolerant NetworkingMechanisms
Hasan M A Islam , Dimitris Chatzopoulos , Dmitrij Lagutin , Pan Hui and Antti Yl¨a-J¨a¨aski Aalto University, Hong Kong University of Science and Technology
Abstract —Content-Centric Networking (CCN) introduces aparadigm shift from a host centric to an information centriccommunication model for Future Internet architectures. It sup-ports the retrieval of a particular content regardless of thephysical location of the content. Content caching and contentdelivery networks are the most popular approaches to deal withthe inherent issues of content delivery on the Internet thatare caused by its design. Moreover, intermittently connectedmobile environments or disruptive networks present a significantchallenge to CCN deployment. In this paper, we consider thepossibility of using mobile users in improving the efficiencyof content delivery. Mobile users are producing a significantfraction of the total internet traffic and modern mobile deviceshave enough storage to cache the downloaded content that mayinterest other mobile users for a short period too. We present ananalytical model of the content centric networking frameworkthat integrates a Delay Tolerant Networking (DTN) architectureinto the native CCN, and we present large scale simulationresults. Caching on mobile devices can improve the contentretrieval time by more than 50 % , while the fraction of therequests that are delivered from other mobile devices can bemore than 75 % in many cases. I. I
NTRODUCTION
Today’s Internet architecture relies on the fundamentalassumption that there exists an end-to-end path between thesource and destination during the communication session.However, the vast majority of Internet usage is dominatedby content distribution and retrieval involving a large amountof digital content and this makes the conventional Internetarchitecture inefficient. In response, Information-Centric Net-working (ICN) [1] emerges as a paradigm shift from a hostcentric to an information centric communication model. Itsupports the retrieval of a particular content without anyreference to the physical location of the content.
Named data is the central element of ICN communication instead of itsphysical location. When a node needs content, it sends arequest for a particular content. If any node on the route of therequest has the content in its content store, it replies with thatcontent to the request. The main argument for this architecturalshift is that named data provide better abstraction than namedhosts.Among all the ICN proposals, Content Centric Networking(CCN) architecture [2] is gaining more and more interest forits architectural design. CCN supports two types of messages:
Interest and
Data . Each CCN node maintains three datastructures; the
Content Store (CS), Pending Interest Table (PIT)and Forwarding Information Base (FIB) . CCN communicationis consumer driven, i.e., a consumer sends
Interest packet towards the content source based on the information storedin the FIB. When a node receives an interest, it checks itslocal cache for the matching content. Otherwise, the nodeforwards the
Interest packet to the interface(s) based on theFIB table until the
Interest packet reaches a content sourcethat can satisfy the interest. Intermediate nodes store theinterests in the PIT so that the data can be sent back tothe proper requester. In addition, PIT is used to suppress theforwarding duplicate interests over the same interface andprovides response aggregations. CCN interests that are notsatisfied within a reasonable amount of time are retransmitted.As CCN senders are stateless [2], the consumer is responsiblefor re-expressing interests if not satisfied.Intermittently connected network topology or network dis-ruption means a significant challenge for ICN deployment. Forinstance, name resolution may fail due to network disruptions,especially when the elements of the distributed resolutionservices are affected by network partitioning. Delay-tolerantnetworking [3] architectures are proposed for such scenar-ios, which are characterised by long delay paths, frequentunpredictable disconnections, and network partitions. Sucharchitectures provide flexible and resilient protocols that buildan opportunistic network on top of existing underlying Layer 2and Layer 3 protocols. This is achieved through asynchronouscommunication along with the use of underlying ConvergenceLayer Adapters (CLA) (TCP, UDP, Bluetooth, etc.).DTN is based on store-carry-and-forward models that utilisepersistent storage that is distributed in the network. Data arecached in the network and are available for opportunistictransmissions. In particular, content based routing has beenexplored in DTN architectures [4]. The multitude of thenetwork interfaces in modern mobile devices allows DTNmechanisms to work in parallel with conventional ones. Forinstance, mobile users who are connected to the internet viathe cellular interface, can also use the WiFi-direct interface toexchange messages with their neighbours. DTN architecturesassimilate properties of ICN architectures and vice versa.In this work, we adapt mechanisms from DTN networkingto the ICN architecture in order to improve the efficiency ofthe content retrieval procedure of mobile users. In more detail,we consider the scenario where mobile users request content via a CCN mechanism. These requests can be of many types,such as a single piece of data (e.g a request for the mapof the current location of the user to Google Maps), a datastream (e.g., the homepage of a news website) or related to aspecific type of information (e.g., opened restaurants close to a r X i v : . [ c s . N I] M a r A BC D C e ll u l a r ne t w o r k c o v e r age U s e r ne t w o r k c o v e r age Fig. 1: The examined ecosystem that combines Content Cen-tric and Delay Tolerant architectures.the user). Some of the requests can be served more effectivelyby the cellular network but there are cases, like the third typeof the request, that can be served locally. Such requests canpotentially be served more effectively by nearby devices orby the cellular towers that have cached the requested contentbecause another user requested that earlier. We considered thecase, where contents are cached in the cellular towers butcan also be requested from other nearby devices that havestored them. To achieve this, we modified the PIT table of thenative CCN while operating on an opportunistic network. Themodified PIT table stores the pending requester(s) informationin the PIT table. The motivation behind this change is the factthat the original PIT table of the CCN keeps track of the arrivalinterfaces of the
Interest packets in a way that is not feasiblein a highly dynamic network.The contributions of this work can be summarised in thefollowing list: (i)
We explain what are the required modifications in theconventional CCN mechanism in order for it to be func-tional in a DTN environment. (ii)
We propose a Content Centric DTN network architecturefor mobile devices and introduce the required modifica-tion for the native CCN so that the native CCN can bridgewith the Content Centric DTN protocol. The ContentCentric DTN protocol operates independently of existingDTN routing protocols, i.e., DTN routing protocols runon top of the Content Centric DTN protocol. Whiledesigning our proposed architecture, we leverage theinherent properties of CCN and DTN architecture [3]. (ii)
We discuss the ways via which a mobile user can receivea requested content and show that the download time ofa content can be decreased significantly via caching inthe cellular access points and in other mobile devices. (iv)
We show that the underlying routing protocol does nothave a substantial effect on the download time of contentsdue to the limited number of hops in the DTN.Figure 1 depicts the examined scenario where at any timemobile users are connected to the cellular network and areable to potentially communicate directly with other mobileusers, depending on the distance between them and the un-derlying communication framework for the device-to-devicecommunication. All the contents are stored in an origin server,which is located in a cloud infrastructure and can be cachedto cellular access points and to mobile devices. Depending onthe placement and the number of the cellular access points, theproportion of the content receptions from other mobile usersdiffers significantly and, as we can see from our large scalesimulations (Section V), mobile users are able to successfullyhandle the requested contents in various different cases ofrouting schemes. After discussing the related work in thenext section, we provide a more detailed explanation of theexamined ecosystem in Section III. Next, in Section IV, wepresent the proposed protocol which is evaluated in SectionV. Finally, in VI we conclude the paper and list our futurework. II. R
ELATED W ORK
Based on the publish/subscribe paradigm, there exist nu-merous research efforts [5] on device-to-device (D2D) com-munication in cellular networks, which is defined as directcommunication between two mobile users without interveningBase Station (BS) or core network. This concept was first pro-posed in [6]. Although D2D, from an architectural perspective,seems similar to Mobile Ad-hoc Networks, the key differencebetween these two is the involvement of the Cellular Accesspoint. Casetti et al. [7] presents content-centric routing in aD2D architecture based on Wifi Direct. The content-centricrouting is based on two data structures: PIT of the nativeCCN and the Content Routing table (CRT). CRT provides therouting information to reach the content items. However, itis not feasible to maintain CRT and the PIT table in dynamicnetworks where mobile users provide intermittent connectivity.In contrast, our proposed scheme exploits the different PITtable which stores the requester(s) information instead of thearrival face of the original CCN so that the reverse path canbe different from the forwarding path of the
Interest packets.Nevertheless, most recently, Garcia et al. [8] have concludedthat
Interest aggregation should not be an integral componentof Content-Centric Networks and propose far smaller and moreefficient forwarding data structures (e.g., CCN-DART [9].Another similar effort has been proposed in [10] that allowswireless content dissemination between mobile nodes withoutrelying on infrastructure support. The proposed architectureis based on the publish/subscribe paradigm. Their focus ismainly on implementation aspects based on 802.11 in ad-hocmode. In contrast, our architecture is based on CCN and DTNarchitecture and hence there are many architectural differencesbetween their effort and our proposal. Most recently, Liu etal. [11] presents detailed descriptions on content routing basedon ICMANET, and describes a concept model for content routing, and categorizes content routing into proactive, reactiveand opportunistic types, then analyzes representative schemes,which can be referred to for the study of joint optimizationbetween content routing and caching in ICMANET. Thereare also several research efforts in the DTN environment[12], [13]. In [12], the author investigates the possibility ofintegrating the ICN and the DTN principles into a sharedICDTN architecture. Combining the ICN and the DTN hasbeen demonstrated in a recent effort called RIFE architecture[13]. The RIFE is a universal communication architecturethat combines the publish/subscribe based POINT architecture[14] and the DTN through a number of handlers for existingIP-based protocols (e.g., HTTP, CoAP, basic IP) which aremapped onto appropriate named objects within the ICN core.The IP endpoints are connected through the ICN using agateway. In contrast, our proposed model exploits the DTNarchitecture in the native CCN architecture that results ina Multihop Cellular Network (MCN) [6]. Amadeo et al.[15] have discussed the potential of the ICN paradigm as anetworking solution for connected vehicles. The authors havesummarized ICN-VANETs relevant literature and presentedthe open challenges in this area. Nevertheless, the analysisof their work shows that the native design principles of ICNwell match the main distinctive features of VANETs and thetargeted wide set of future vehicular applications. The authorsof [16] presents IP-based data DTN routing mechanisms usingCCN on the sparsely-connected real vehicular testbed andvalidate the performance and usability of CCN over VANET.However, their proposed schemes have not considered theforwarding loop and duplicates at the content level whileoperating on IP-based routing mechanisms. Our proposedmodel operates independently of DTN routing and can detectthe duplicates, and forwarding loop at content level.User-centric data dissemination in DTNs has been widelyexplored from various points of view [17]–[20]. Authors of[21] proposed a user-assisted in-network caching scheme,where users who request, download, and keep the contentcontribute to in-network caching by sharing their downloadedcontent with other users in the same network domain. Sourlas et al. [17] proposed an information-resilience scheme in thecontext of Content-Centric Networks (CCN) for the retrievalof content in disruptive, fragmented networks depending onthe in-network caching of its attached user. The proposedscheme enhanced the Named Data Networking (NDN) routerdesign as well as the
Interest forwarding mechanisms so thatusers can retrieve cached content when the content origin isnot reachable. To achieve this, the authors introduce a newtable, called Satisfied Interest Table (SIT), which keeps trackof the Data packets that are forwarded to users. In case thecontent origin is not reachable, the proposed scheme exploitsthe cache of the other users following SIT entries. However,the proposed scheme performs well only if the users listedin the SIT entries are connected. In [22], the authors presentagent-based content retrieval on top of CCN which providesinformation-centric DTN support as an application modulewithout modifications to CCN message processing. However,their proposed scheme may suffer from PIT bottleneck indelay tolerant environment. In contrast, our proposed scheme exploits the opportunistic communication of mobile usersusing DTN mechanisms.From a social-based point of view, the authors of SocialCast[18] proposed a routing framework that exploits the social tiesamong users for effective relay selection, while Yoneki et al.in [19] discussed the design of a publish-subscribe commu-nication overlay based on the distributed detection of socialgroups by means of centrality measures. However, this routingmechanisms can be complementary to our proposed scheme,which operates independently of any routing algorithm. Lu etal. at [20] used the K-means clustering algorithm to buildthe social level forwarding scheme in order to reduce thetransmitted messages. This approach raises several inevitablelimitations: (i) the interest may fail to reach the encounterednode with the same social level that might have the contentto satisfy the interest, (ii) the request from the higher sociallevel will never reach a content provider with a lower sociallevel, (iii) the proposed scheme cannot detect the routing loopof the
Interest packet and (iv) the authors have not consideredhow to optimise similar interests from multiple users. Theselimitations are addressed in our solution.D2D communication highly depends on the participation ofmobile users in sharing contents. Mobile users may be selfishand would not be willing to forward data to others due tolimited resources (e.g., memory, battery power). To handlethis issue, a number of incentive mechanisms [23]–[25] hasbeen proposed to motivate users to work in a cooperative way.D2D is still immature and faces many technical challengesand issues regarding aspects such as device discovery, relayselection, security and interference mitigation. The authors of[26] presents an incentive mechanism for data centric messagedelivery in DTN that exploits the social relationships. Thismechanism prevents users from becoming selfish and moti-vates them to relay the most popular content. Nevertheless,the incentive mechanisms are complementary to our proposedmodel and can be applied on top of our solution. In thiswork, we assume that all mobile users are participating ina cooperative way. III. S
YSTEM M ODEL
A. Preliminaries
We analyse a CCN architecture where mobile users makea request for named data contents c ∈ C . We consider a setof mobile users M that browse in a large scale metropolitanarea and produce their requests for content . A set of CellularAccess Points (CAPs) A are deployed in the area and weassume that at any time t any mobile device m ∈ M isassociated with one cellular access point and we denote thisby m a ( t ) ∈ A , while any CAP a has N a ( t ) mobile devicesassociated with it at time t .Each CAP a also operates as a CCN node by being con-nected to the fixed network and it maintains a Pending InterestTable P a and a Forwarding Information Table F a . Also, partof its storage S a is used for caching contents and works as a Content Store . The cache of each CAP is measured based on The terms user, node and device are used interchangeably depending onthe context. the proportion of total contents that it can store S a = α A |C| , < α A << . In addition to the three traditional tables thatare used in CCN architectures we add one table, motivatedby the work of [17], which stores the satisfied interests. Wedenote that table with D a . The entries of D a are of the form: < content, user, time > and work in a similar wayto the forwarding interest table, but with the difference thatthey keep who has satisfied its interest. In addition, we addanother table that stores the Pending Requester Informationtable (PRIT) that stores the requester information instead ofthe arrival interface. PRIT is used when the CAP receivesrequests from the DTN interface.Any mobile node m is able to communicate directly withits neighbours N m ( t ) , whose number depends on the mobilityof the users, and the interface used for the connectivitybetween them . We also denote by N km ( t ) the mobile users m being able to communicate in k > hops, at time t .Similarly to the CAPs, each mobile node m keeps three tablesa Pending Requester Information Table P m , a ForwardingInformation Table F m and a Satisfied Interest Table D m andhas a Content Store , S m . The cache of each node is measuredbased on the proportion of total contents that it can store S m = α M |C| . We assume that the storage capabilities ofcellular access points is much higher than that of the mobiledevices (e.g., some Terabytes compared to a few Megabytes), < α M << α A << . Table I contains the introducednotation . B. Problem Formulation
Each mobile user m ∈ M requests contents c ∈ C at therate r cm . We use the vector r c ∈ R |M| + to denote all the requestrates of all the mobile users for content c and the zero norm of r c , || r c || to indicate the number of the mobile users that arerequesting content c . The request rate may depend on multiplefactors, but in this work we consider only the popularity ofthe content π c and the profile of the user u m , which indicatesthe probability of a mobile user requesting each content. Wedenote the profiles of all users with the vector u . So the requestrate of content c by user m is given by: r cm = u m · π c (1)The service rate of an expressed interest from a user m and acontent c depends on the popularity of the content π c and thecontent placement strategy that will be explained in detail inSection IV. An interest in a content from a mobile user canbe served in three ways:
1) Core Network:
The mobile device, via the cellularnetwork, sends the
Interest packet and the content is retrievedin the traditional CCN way from the Content Store of anyintermediate node or from the server of origin, where thecontent was initially placed upon its creation. At any time t there exists at least one node that has the required content. In Bluetooth has a coverage radius of some tens of meters, WiFi-direct of afew hundreds and the soon-to-be-available LTE-direct is expected to have acoverage radius of half a kilometer. To avoid listing the same variables for both mobile devices and CAPs weuse x , X and y (i.e. x = { m, a } , X = {M , A} and y = { , , } ). The zero norm of a vector equals to the non-zero elements of the vector.
TABLE I: Notation Table C set of available contents M set of mobile users A set of cellular access points α X cache capacity of mobile users M or CAPs AN x ( t ) mobile users accessible by user m or CAPs a at time tP x Pending Interest Table of user m or CAP aF x Forwarding Infromation Table of user m or CAP aS x Content storage of user m or CAP aD x Satisfied Interest Table of user m or CAP aπ c Popularity of content cu m Content request profile of mobile user mr cm Request rate of content c from mobile user ms cN y ( t ) Service rate of content c at time t through network N y such a case, the service rate of content c is denoted by s cN anddepends on the popularity of the content and the characteristicsof the network (load, bandwidth, etc) and the caching policy(e.g., LRU, FIFO, LFU), c N . Without loss of generality weassume: s cN ( t ) = c N ( t ) · π c (2)
2) Cellular Access Point:
The mobile device downloadsthe cached content from the cellular tower because anotheruser had requested the content earlier. Given that the availablecache of each cellular tower is limited compared to the storagesize of the server of origin, the cached contents are limited( α A |C| ) but, depending on the caching policy, can achieve ahigh hit rate due to the popularity distribution of the contentsand the spatial skewness [27]. In that case, we denote theservice rate with: s cN ( t ) = α A · c N ( t ) · π c (3)
3) Delay Tolerant Network:
The mobile device gets thecontent from another mobile device via a single-hop or amulti-hop path. The number of the hops depends on (i) thephysical distance between the users, (ii) the number of theusers and (iii) the popularity of the content. Popular contentsare more probably found closer to the user who initiated therequest. Although mobile devices are not able to cache manycontent items, the social relationship between mobile users thathave, with high probability, similar mobility patterns, makes itprobable for two socially close mobile users to express interestin similar items [28]. In that case, we denote the service ratewith: s cN ( t ) = α M · c N ( t ) · π c · (cid:89) m ∈N m ( t ) u m (4)We employ a Markov process { X c ( t ) , ≤ t < ∞} withstationary transition probabilities that shows the number of thenodes in the whole ecosystem (mobile users, cellular accesspoints, the server of origin as well as the network componentssuch as switches that are part of the CCN ecosystem that havethe content c in their caches). If at any time ˜ t , X c (˜ t ) = 0 , this will mean that the content is not available at all, which can betrue only in the case of a very unpopular item that is not cachedin any node and the server of origin is not accessible becauseof network partitioning. However, although this is not realistic,we can use the Markov process as a birth-death process witha single absorbing state, which we define to be X c ( t ) = 0 inorder to then use the absorption time formula [29] that includesthe cost parameters for each type of network as an objectivefunction to optimise. In more detail [29]: T cn = ∞ (cid:88) i =1 λ ci ρ ci + n − (cid:88) k =1 ρ ck ∞ (cid:88) j = k +1 λ cj ρ cj , (5)if ∞ (cid:88) i =1 λ ci ρ ci < ∞ (6)and T cn = ∞ , if (cid:80) ∞ i =1 1 λ ci ρ ci = ∞ , where: λ cn is the birth rateof the process at state n , µ cn is the death rate and ρ cn = n (cid:89) i =1 µ ci λ ci (7)The birth rate of the process at state n and for content c , λ cn ,depends on the request rates for the examined item of each ofthe users r cm . λ cn ∼ (cid:88) m ∈M|| r c || = n r cm (8)while the death rate depends on the type of the service rate andthe caching policies. The required time for the Markov processto reach the absorption state depends on the initial state andthe difference between the service rates and the request rates.The service rate depends on the probability of a contentbeing placed close to the mobile users that generate requestsfor it. The probability of a content c being cached in the CAPwhich mobile user m is associated with at time t , m a ( t ) is: p cm a ( t ) := P [ c ∈ S m a ( t ) ] (9)and the probability of c being stored in at least one of m ’sneighbours is p c N m ( t ) := 1 − (cid:88) j ∈N m ( t ) P [ c / ∈ S j ( t )] , (10)while for K hops away from m , the probability of c beingcached is: p c N Km ( t ) := 1 − (cid:88) j ∈N Km ( t ) P [ c / ∈ S j ( t )] . (11)So the probability for a mobile user not being able to retrieve c from the access point that he or she is associated with andfrom any mobile user whose distance is at most K -hops is : p cm ( K, t ) := 1 − p cm a ( t ) − p c N m ( t ) − K (cid:88) k =2 p c N km ( t ) . (12) Small values of K are enough for successful content discovery [30]. S t a t i c D y na m i c U se r n e t w o r k c o ve r a g e m j j k k ab b b Fig. 2: The cellular access points connect the CCN with thestatic links between the nodes to the highly dynamic andunpredictable DTN.We also define the probability of a content c being cached inthe cellular access point a of at least one of the mobile devicesthat are associated with a : p ca [ N a ( t ) , t ] = 1 − P [ c / ∈ S a ( t )] · (cid:89) j ∈N a ( t ) P [ c / ∈ S j ( t )] (13)The size of the Content storage in the CAPs and mobiledevices, and more specifically the proportion of the totalitems they can store is what affects p cm ( K, t ) and p ca [ N a ( t ) , t ] .Another determinant parameter is the number of mobile usersthat are associated with the same access point as the user thatrequested a content item and, consequently, the diversity inthe subset of the objects that are cached in all these devices.We denote with C a ( t ) ⊂ C the set of the content items that arecached in at least one device that is accessible from CAP a orare cached in a . Then equation (13) can be expressed shortlyas P [ c ∈ C a ( t )] .Next, in Section IV, we present a protocol that determineswhich contents should be cached in each device and for howlong. The protocol is designed to consider highly dynamicmobile users with limited resources as well as the static accesspoints that operate as the glue between the dynamic users andthe fixed infrastructure.IV. P ROTOCOL
The original design of CCN is based on the fact thatthe multiple network interfaces can be integrated via themechanism of the forwarding information base [2]. Each entryon the FIB points to a list of interfaces that can be used toforward
Interest packets towards the desired content producer.At this point, the traditional CCN can be combined with DTNnetwork protocols, as presented in figure 2. The integrationof DTN architecture with the native CCN architecture resultsin a Multihop Cellular Network (MCN) [6]. The generalconcept of MCN comprises a cellular network in which userdevices can communicate with each other, either via meansof a conventional cellular mode or via means of direct D2Dcommunication if they are mutually reachable. To enable thisparadigm, the functionalities of the proposed protocol can bedecomposed into three parts: (1) The control plane that performs packet (
Interest/Data )management. The control plane is implemented on top ofthe DTN mechanisms, and its functionalities are responsiblefor performing specific actions based on the packet type(
Interest/Data ). To achieve this, the control plane inserts themeta-information in the DTN messages. (2) The forwarding plane that consists of two parts and,depending on the type of the node, it can be either the native
CCN forwarding or the
DTN forwarding (Store-carry-and-forward) . This module provides an interface between the CAPand the mobile nodes so that the CAP can hand over thepacket to a mobile node. The mobile node exploits DTNarchitecture to forward the packets in D2D fashion whileoperating in an opportunistic network without the interventionof the cellular network. The CAP includes a separate PITtable called the Pending Requester Information Table (PRIT)which stores the requester(s) information instead of the arrivalfaces of the
Interest packets. The mobile nodes only use ourproposed architecture while operating in a DTN environment,i.e., the control plane is implemented on top of the DTNforwarding plane and enables the host centric DTN to performin content centric fashion. To bridge between CCN and DTN,each message carries meta-information of the CCN mechanismthat assists the content centric operation in DTN environment. (3) The routing decision engine is the process by whichone router sends packets to another router by means of routingprotocols which decide the appropriate path for the packet.The routing protocol assists the router in choosing the bestpath out of many paths. The routing decision engine operatesindependently on top of our proposed model.The proposed protocol deals with two control decisions:1)
Request/Response Processing : Although the CAPs oper-ate as conventional CCN nodes regarding the forwardingand the routing of
Interest and
Response packets, it is notthe same for mobile users. Whenever a mobile user ofa CAP receives a content request or a content response,there is the question of what actions should be taken?2)
Content Management : Given that a mobile user or a CAPhas a content item, should it store it in the content storeor drop it? The CAPs have higher storage capabilitiesthan the mobile users, but still they can not cache all theavailable contents.
A. Request Processing
In the relatively static CCNs, the
Interest packets arepropagated as upstream towards the potential data sources,while leaving a trail of bread crumbs for the matching datapackets to follow back to the original requester(s). On theother hand, in dynamic environments the nodes are mobileand the connections are intermittent, which means that itis not feasible to keep track of the changes in the net-work topology. Unlike the conventional PITs in CCN, mo-bile users keep the address information of the requester(s)in the
P ending Requester Inf ormation T able ( P RIT ) so that they can forward similar content towards potentialrequester(s). PRIT is also used to detect forwarding loopand aggregate the similar interests. Mobile users exploit the Algorithm 1
Processing Interest Message key ← [ Interest ] if key in Local Cache then content ← Cache ( key ) end if if content (cid:54) = NULL then response ← createResponse ( content ) if current node = mobile user then requester ← [ Interest ] Send response to requester following P RIT else
Send response following P IT breadcrumb end if else if current node = mobile user then satisfied req provider ← lookup SRIT ( Interest ) if satisfied req provider (cid:54) = NULL then
Send Interest to satisfied req provider else pending requester ← lookup P RIT ( Interest ) if requester ∈ pending requester then drop the interest packet else Add requester to P RIT table forward the Interest to next Hop end if end if end if if current node = CAP then
F IB entry ← native CCN mechanism ( Interest ) satisfied req provider ← lookup SRIT ( Interest ) if F IB entry = NULL then if satisfied req provider (cid:54) = NULL then
Send Interest to satisfied req provider else pending requester ← lookup P RIT ( Interest ) if requester ∈ pending requester then drop the interest packet else Add requester to P RIT table forward the Interest to mobile user end if end if end if end if end if
Satisf ied Request Inf ormation T able ( SRIT ) to remem-ber all the satisfied interests of the requester(s) so that itcan provide information on the potential content source forthe similar interests in future. By doing that, an intermediatenode having an entry matching with the interest packet in theSRIT can forward the Interest packet towards those potentialcontent provider(s). The CAP acts similarly to a mobile nodeif it receives the
Interest packet from the DTN interface.Nevertheless, if the CAP has FIB entry for this
Interest , itcan also apply the native CCN mechanism. The overview ofthe request processing is presented in Algorithm 1.On the reception of an
Interest packet, a mobile nodeinitially searches in its Content Store and if there is no match,the node checks its SRIT table to verify if there is any entrymatching with the
Interest packet. If any matching is found,the node forwards the request towards those potential contentprovider(s) from SRIT. The node also enters the
Interest packet in the PRIT table. The PRIT is used to keep track of the IDs of the interest(s) creators that are used as destinations inthe response packets. In more detail, upon the reception of an Interest packet the mobile node checks its PRIT. If there isan older entry for the same content, it updates the entry onlyif the requester is different, otherwise it drops the
Interest packet. On the other hand, the CAP first applies the nativeCCN mechanism, i.e., it searches the content store to verify ifit can satisfy the request. If there is a match, the CAP sendsthe content back to the requester. If no matching is found, theCAP forwards the request further, based on the information ofthe FIB. The CAP can also forward the request to the mobilenode which runs our proposed architecture. Before forwardingthe request to the mobile node, the CAP will store the requesterinformation in the PRIT table, but only if it receives the requestfrom the DTN interface.Regardless of the total number of users, our proposal doesnot spread the
Interest packets all over the ecosystem becauseit is inefficient and not worthwhile doing since the mobilenodes are submitting their requests in parallel to both theCAPs they are connected to and their neighbouring mobiledevices. More importantly, the respective CAPs inform themobile nodes whether there exists another mobile node thathas the requested content in the same cell, and depending onthe level of the assistance from the CAPs, as will be discussedin the next section, the mobile nodes can either receive theirrequest via a multi-hop-but-short path from another node inthe same cell, or via a two hop path with the help of theCAP. So, a request as shown in Figure 3 can be served infour ways: (A) from the Content Store of the associated CAP, (B) via the associated CAP that retrieved the content fromthe conventional CCN network, (C) from another mobile nodethat sent the content via a multi-path among the other mobilenodes, and (D) from another mobile node that sent the contentto the CAP, which then forwarded the content to the requester.
B. Response Processing
Algorithm 2 presents an overview of the response process-ing on a network node. When the
Interest packet reaches anode having content matching with the
Interest packet, thenode constructs a response packet with the content and sendsit back to the originator of the request. If the intermediatenode is a mobile node, the node checks the PRIT table andremoves the entry if there is a match for the response packet. Ifthe PRIT entry has the information on multiple requesters, theintermediate node adds all the source IDs of those requestersto the response packet as meta-information. If the interme-diate node does not find any matches in the PRIT table, itsimply forwards the response packet to the next best contact.Subsequently, if the response packet reaches the target node,it checks the meta-information to verify if there is any otherpending requester(s) who requested this content. If there existsno pending requester information, the recipient node drops thepacket to avoid further transmission by the DTN mechanism.Otherwise, if the node finds other pending requesters, it willforward the response to those pending requesters. If the meta Without loss of generality we assume that the id of user m is m . Algorithm 2
Processing Response Packet if current node is mobile user then destination id ← [ Response ] content provider ← [ Response ] insert content provider in SRIT table if current node is the destination then notify application key ← [ Response ] pending requester ← lookup P RIT ( key ) if pending requester is empty then drop the packet return else forward response to pending requester return end if end if end if if current node is CAP then if response is received from DT N interface then key ← [ Response ] content provider ← [ Response ] insert content provider in SRIT table pending requester ← lookup P RIT ( key ) if pending requester is empty then drop the packet else forward response to pending requester end if else follow the native CCN mechanism end if end if A B CD C
Fig. 3: The four potential ways via which a mobile user canget the requested content item.information has multiple pending requesters, the node adds onerequester as the destination address for the response and otherrequester(s) as meta-information. If the intermediate node isthe CAP, it checks both PIT and PRIT to forward the responsein an appropriate manner. If the CAP finds a match in its PIT, itfollows the native CCN mechanism. A match in PRIT followsour proposed scheme.
TABLE II: Performance Metrics
Average end-to-end delay The average time passed to receive a content inresponse to a request.Packet drop The amount of packets (Interest/Data) that is ef-fectively suppressed by the content router.Traffic split The service rate, i.e., the amount of requests pro-cessed by different types of nodes in the network:mobile user, CAP, content source.Service load The amount of requests processed by the contentprovider.
C. Content Management
Mobile nodes can provide storage memory depending ontheir resource availabilities and policies. Using its storagememory, a mobile node can serve as the network mediumto share the content. Furthermore, this cache can also be usedby store-carry-and-forward based DTN protocols. However,the persistent storage of the DTN protocol keeps the messageuntil the successful delivery of the message to the next bestopportunistic contact. In our proposed architecture, the storagememory of mobile node keeps the response packet to satisfythe future requests. However, only in the ideal case is themobile node’s storage big enough to store all received content.Under the assumption that a mobile node can only store a smallproportion of the total contents, a caching policy is required.Additional information, such as the popularity distributionof the contents and the request profile of the mobile users,can be used by a caching policy to determine the probabilityof each content being requested by the user and, based onthat, a decision is made whether a newly received content bedropped or replaced with the unpopular one, given that there isno available space in the mobile node. However, the CAP hasinformation about the stored content in the whole cell that isnot utilised. We utilise this additional information by definingthe expected retrieval cost of each content by combining thisinformation: (i) the popularity of each content item, (i) theprofile of the users and the (i) estimated time required toretrieve the content via one of the aforementioned ways. Thiscosts are calculated with the assistance of the CAPs, which isable to recommend a mobile node on whether to keep an itemor not. V. E
VALUATION
We evaluate our proposed architecture using the Opportunis-tic Network Simulator (ONE) [31]. The goal of our evaluationis to investigate the performance of our proposal in terms of (i)
Average end-to-end delay, (ii) service load ratio, (iii) packetdrop ratio and (iv)
Traffic split. Table II contains a descriptionof each of these metrics.The ONE simulator contains map data of the Helsinkidowntown area (e.g., roads, tram routes and pedestrian walk-ways) and various Map-based Movement models: (1)
RandomMap-Based Movement, (2)
Shortest Path Map-Based Move-ment, and (3)
Routed Map-Based Movement. We employ theShortest Path Map-Based Movement since it is more realisticbecause the mobile users, after choosing a destination point on TABLE III: Simulation Parameters
Parameter ValueSimulation Duration 5 days (432000s)Number of Requesters 10Time interval of generating Interests 5minNumber of Relay Nodes 160Number of Access Points 30Cache of mobile users 10 itemsCache of Access points 50 itemsTTL value 500sTransmission range of Access Points 100mTransmission speed of Access Points 10Mbps.Transmission range of Mobile devices 10mTransmission speed of Mobile devices 2.5 Mbps. the map, follow the shortest path to that point from their cur-rent location. The destination point is chosen randomly froma list of Points of Interest (POI), which includes popular realworld destinations (e.g., shops, restaurants, tourist attractions).The simulation area approximately is 20km .In the simulation, we considered mobile users that are eitherwalking at a speed that is in the range of 1.8 kilometres perhour to 5.4 kilometres per hour or driving a car or using thetram. We categorised the mobile users into two groups: (i) requesters and (ii) intermediate users. The requesters were10 and the intermediate users were 150. All of them weredivided into four different groups and assigned with differentprobabilities of choosing the next group specific POI orrandom places to visit. Regarding the content generation, weconsidered content generated by 10 other mobile users or fromnon-mobile content generators (e.g., a news website). Apartfrom the mobile users, we also considered 30 CAPs that havecaching capabilities.None of the users had any content in the beginning of thesimulation, but whenever one requester imposed a request ona CAP, the content was retrieved from the content provider inthe cloud if it had not already been cached from a previousrequest, and delivered to the requester. The simulation timewas 5 days and we used the first day as a warm up phase. Allthe details of the simulation parameters are listed in Table III. A. DTN Routing
The Content-Centric functionalities of our proposal arerouting independent, and for that reason we examine theperformance of our proposal in four different cases regard-ing the routing strategies: (i)
Epidemic [32], (ii)
Spray-and-Wait [33], (iii)
First contact [34] and (iv) a hybrid one thatworks like the Epidemic in the forwarding step until reachingthe destination and also like the Spray-and-Wait in the reversepath creation step. Epidemic routing has no limitation ongenerating copies for each message. In this routing scheme,each node carries a list of all messages whose delivery ispending. Whenever a node encounters another node, theyexchange all that messages that are not common in their list.Spray-and-Wait generates a limited number of copies for everymessage and spreads initially. If a node does not find thedestination in the spray phase, it waits for the destinationto perform direct transmission. In our experiment, Spray-and-Wait generated 10 copies for every message in the spray phase.First contact generates only one copy per message. The hybrid T r a ff i c S p lit DTN CAP Source (a) Uniform content popularity T r a ff i c S p lit DTN CAP Source (b) Zipf content popularity
Fig. 4: The service rate, i.e., the distribution of the contentresponses from each type of source for each DTN routingprotocol.one sprayed
Interest packets (limited to 10 copies) until therequest reached the content providers and then used the Spray-and-Wait routing to deliver the content back to the requester.
B. Query Distribution
We generate user interest based on the available contents C , which we assume are 1000 (i.e. |C| = 100 ). We assumethat the i -th content c i is the i -th most popular one π i < π j ∀ i ≤ j . The users’ request profiles are randomly generated viathe uniform distribution. Content popularity is correlated withuser requests [35] and follows the well-known Zipf distribution[36]. In this work we consider two cases for the contentpopularity: (i) uniform and (ii) Zipf on initialising π c ∀ c ∈ C .For the Zipf distribution we initialised the parameter to 1 andthe normalizing constant to 0.2. C. Performance boost of Proposed Architecture
We measure the performance of our proposal using themetrics that are listed in Table II. Figure 4 shows how thecontent requests are served of each type of DTN routingprotocol. Practically, we show how the hit rates of each contentprovider type are related. Upon every request, the proposedmechanism uses all the possible ways in parallel in order todownload the content as soon as possible. As we can see fromboth Figure 4a and Figure 4b the content caches in the CAPstogether with the caches in the mobile nodes can handle morethan 90% of the requests. Only in the case of the Spray andWait routing protocol the requests are served by the contentproducer around 25% when the content popularity follows
Epidemic First Contact Spray and Wait Hybrid S e r v i ce l o a d by C on t e n t P r ov i d e r Proposed ModelDTN with user caching DTNProposed Model without user Caching (a) Load in the content provider in the case of contents withpopularity that follows the uniform distribution
Epidemic First Contact Spray and Wait Hybrid S e r v i ce l o a d by C on t e n t P r ov i d e r Proposed ModelDTN with user caching DTNProposed Model without user Caching (b) Load in the content provider in the case of contents withpopularity that follows the Zipf distribution
Fig. 5: Comparison of our proposal with other frameworksthat do not support either CCN functionalities or caching.the uniform distribution and 20% when they follow the Zipfdistribution.In order to measure the contribution of the CCN mecha-nisms and the caches in the mobile users and in the CAPs, weimplemented three simpler mechanisms and we compare themwith our proposal in Figure 5. The first one is a simple contentsearch using the DTN mechanism, i.e., that is operating asa request-response application on top of the DTN routingprotocols and is denoted by
DTN . The second one is animproved version of the first one that has content caches.Each content cache can store 10 objects. This mechanism isdenoted by
DTN with user caching . The last one is the sameas our proposal but without caching in the mobile users andis denoted by
Proposed Model without user Caching . As wecan see from both 5a and 5b, our model is under-loading thecontent providers more than the other competitors. It is worthmentioning that in the case of Epidemic routing, the contentprovider is overloaded because unlimited number copies ofeach request is generated until the request reaches the contentprovider.Next, we present in Figure 6 the benefit of using CCNmechanisms in conjunction with the DTN routing protocolsbecause they filter the requests and stop forwarding identicalpackets. We observe that if user caching is not used, thenumber of duplicate packets significantly increase. This ishappening because the request packets stay in the networklonger to reach the potential content provider, and, hencethe communication overhead in terms of additional traffic(interest/data packet) increases. Our proposal detects thoseduplicates and drops them accordingly. For instance, in thecase of uniform distribution and multi-copy routing (e.g., Spray and Wait Hybrid First Contact Epidemic P ac k e t d r op ( % ) Uniform Zipf
Fig. 6: The benefit of CCN in the DTN routing protocols interms of packet drop. S e r v i ce l o a d r e du c ti on on m ob il e u s e r ( % ) Proposed Model vs DTN with user caching Uniform Zipf
Fig. 7: Our proposal vs DTN with user caching but withoutCCN functionalities.Epidemic, Spray and Wait and Hybrid routing), we observethat more than 50% of the duplicate packets are reduced inour proposal as compared to our proposed model without usercache. This is happening because these protocols are producingmultiple copies per request and each content has the samechance of being requested multiple times and being found ina nearby user’s pending interest table. On the other hand, inthe case of First contact (single copy routing), we observe27% duplicate packet reduction. In the case of Zipf contentpopularity and First Contact, we observe more than 60%duplicate packets reduction. This is because without the usercache, the requests take a longer time to reach the potentialcontent provider, whereas the other three DTN protocols canpotentially reach the content provider faster than First Contact.We also examine the service load on mobile users withour proposal to compare
DTN with user caching . Figure 7shows that our model reduces the service load on the mobileuser in all routing. Especially in First Contact routing, theservice load on the mobile user is significantly reduced by57 % when the content popularity follows uniform distribution.On the other hand, when Hybrid and Epidemic routing is used,the service load is reduced by 37 % and 28 % respectively.This is because First contact generates single copy for eachrequest, whereas others use multiple copies. Multiple copiesincrease the probability of reaching the content provider faster.Service load is not significantly reduced (10 % ) by the Sprayand Wait routing due to a limited number of message copies.We observe that in the case of Zipf content popularity, serviceload reduction (approximately 10 % ) on mobile users by Sprayand Wait routing is almost similar to Epidemic routing.Furthermore, we examine the changes in the average delayof the content retrieval in Figure 8. As expected, we hada decrease in the delay in most of the cases because of -1-0.8-0.6-0.4-0.2 0 0.2 0.4 CCN and User Caching Without CCN Without User Caching A v e r a g e e nd - t o - e nd d e l a y Uniform Zipf (a) Epidemic routing -1-0.8-0.6-0.4-0.2 0 0.2 0.4
CCN and User Caching Without CCN Without User Caching A v e r a g e e nd - t o - e nd d e l a y Uniform Zipf (b) First Contact -1-0.8-0.6-0.4-0.2 0 0.2 0.4
CCN and User Caching Without CCN Without User Caching A v e r a g e e nd - t o - e nd d e l a y Uniform Zipf (c) Spray and Wait -1-0.8-0.6-0.4-0.2 0 0.2 0.4
CCN and User Caching Without CCN Without User Caching A v e r a g e e nd - t o - e nd d e l a y Uniform Zipf (d) Hybrid
Fig. 8: Average end-to-end delay change compared to themechanism that does not have CCN and caching function-alities.the caching mechanisms. Especially in the case of contentswith popularity that follows the Zipf distribution, the contentswere accessed faster because they were cached somewherenearby. However, there are cases where the delay can beincreased because there are not many requests for contents inthe Spray and Wait routing protocol with contents that followthe uniform distribution (Figure 8c).VI. C
ONCLUSION AND F UTURE W ORK
In this paper, we investigated the possibility of using mobileusers in improving the performance of content delivery. Forthis, we explain the necessary required modifications in theconventional CCN mechanism in order for it to be functional ina DTN environment. Furthermore, we present a mathematicalmodel of the content centric networking framework that ex-ploits the opportunistic communications among mobile users.The proposed framework is implemented in ONE simulator toevaluate the concept. The simulation result shows that caching on mobile devices and cellular access points can improve thecontent retrieval time by more than 50 % , while the proportionof the requests that are delivered from other mobile devicescan be more than 75 % in many cases. Our next steps will befocused on the development of caching policies and on varioustypes of contents that are application dependent. Moreover,we plan to consider incentives that motivate mobile users tocooperate and store other content.R EFERENCES[1] G. Xylomenos, C. N. Ververidis, V. A. Siris, N. Fotiou, C. Tsilopou-los, X. Vasilakos, K. V. Katsaros, and G. C. Polyzos, “A survey ofinformation-centric networking research,”
IEEE Communications Sur-veys & Tutorials , vol. 16, no. 2, pp. 1024–1049, 2014.[2] V. Jacobson, D. Smetters, J. Thornton, M. Plass, N. Briggs, andR. Braynard, “Networking named content,” in
Proceedings of the5th international conference on Emerging networking experiments andtechnologies . ACM, 2009, pp. 1–12.[3] V. Cerf, S. Burleigh, A. Hooke, L. Torgerson, R. Durst, K. Scott, K. Fall,and H. Weiss, “Delay-tolerant networking architecture,”
RFC4838, April ,2007.[4] P. Costa, M. Musolesi, C. Mascolo, and G. P. Picco, “Adaptive content-based routing for delay-tolerant mobile ad hoc networks,”
ResearchNotes RN/06/0808, University College London , 2006.[5] A. Asadi, Q. Wang, and V. Mancuso, “A survey on device-to-devicecommunication in cellular networks,”
IEEE Communications Surveys &Tutorials , vol. 16, no. 4, pp. 1801–1819, 2014.[6] Y.-D. Lin and Y.-C. Hsu, “Multihop cellular: A new architecture forwireless communications,” in
INFOCOM 2000. Nineteenth Annual JointConference of the IEEE Computer and Communications Societies.Proceedings. IEEE , vol. 3. IEEE, 2000, pp. 1273–1282.[7] C. Casetti, C. F. Chiasserini, L. C. Pelle, C. Del Valle, Y. Duan,and P. Giaccone, “Content-centric routing in wi-fi direct multi-groupnetworks,” in
World of Wireless, Mobile and Multimedia Networks(WoWMoM), 2015 IEEE 16th International Symposium on a . IEEE,2015, pp. 1–9.[8] A. Dabirmoghaddam, M. Dehghan, and J. Garcia-Luna-Aceves, “Char-acterizing interest aggregation in content-centric networks,” in
IFIPNetworking) . IEEE, 2016, pp. 1–7.[9] J. Garcia-Luna-Aceves and M. Mirzazad-Barijough, “A light-weightforwarding plane for content-centric networks,” in .IEEE, 2016, pp. 1–7.[10] ´O. R. Helgason, E. A. Yavuz, S. T. Kouyoumdjieva, L. Pajevic, andG. Karlsson, “A mobile peer-to-peer system for opportunistic content-centric networking,” in
Proceedings of the second ACM SIGCOMMworkshop on Networking, systems, and applications on mobile hand-helds . ACM, 2010, pp. 21–26.[11] X. Liu, Z. Li, P. Yang, and Y. Dong, “Information-centric mobile ad hocnetworks and content routing: a survey,”
Ad Hoc Networks , vol. 58, pp.255–268, 2017.[12] G. Tyson, J. Bigham, and E. Bodanese, “Towards an information-centricdelay-tolerant network,” in
Computer Communications Workshops (IN-FOCOM WKSHPS), 2013 IEEE Conference on , April 2013, pp. 387–392.[13] D. Trossen, A. Sathiaseelan, and J. Ott, “Towards an information centricnetwork architecture for universal internet access,”
ACM SIGCOMMComputer Communication Review
IEEECommunications Magazine , vol. 54, no. 2, pp. 98–104, 2016.[16] D. Saxena, V. Raychoudhury, and C. Becker, “Implementation andperformance evaluation of name-based forwarding schemes in v-ndn,”in
Proceedings of the 18th International Conference on DistributedComputing and Networking . ACM, 2017, p. 35.[17] V. Sourlas, L. Tassiulas, I. Psaras, and G. Pavlou, “Information resiliencethrough user-assisted caching in disruptive content-centric networks,” in
IFIP Networking Conference (IFIP Networking), 2015 . IEEE, 2015,pp. 1–9. [18] P. Costa, C. Mascolo, M. Musolesi, and G. P. Picco, “Socially-awarerouting for publish-subscribe in delay-tolerant mobile ad hoc networks,”
IEEE Journal on Selected Areas in Communications , vol. 26, no. 5, pp.748–760, 2008.[19] E. Yoneki, P. Hui, S. Chan, and J. Crowcroft, “A socio-aware overlayfor publish/subscribe communication in delay tolerant networks,” in
Proceedings of the 10th ACM Symposium on Modeling, analysis, andsimulation of wireless and mobile systems . ACM, 2007, pp. 225–234.[20] Y. Lu, X. Li, Y.-T. Yu, and M. Gerla, “Information-centric delay-tolerantmobile ad-hoc networks,” in
Computer Communications Workshops(INFOCOM WKSHPS), 2014 IEEE Conference on . IEEE, 2014, pp.428–433.[21] W. Gao and G. Cao, “User-centric data dissemination in disruptiontolerant networks,” in
INFOCOM, 2011 Proceedings IEEE . IEEE, 2011,pp. 3119–3127.[22] C. Anastasiades, T. Schmid, J. Weber, and T. Braun, “Information-centriccontent retrieval for delay-tolerant networks,”
Computer Networks , vol.107, pp. 194–207, 2016.[23] M. Noura and R. Nordin, “A survey on interference managementfor device-to-device (d2d) communication and its challenges in 5gnetworks,”
Journal of Network and Computer Applications , vol. 71, pp.130–150, 2016.[24] L. G. Jaimes, I. J. Vergara-Laurens, and A. Raij, “A survey of incentivetechniques for mobile crowd sensing,”
IEEE Internet of Things Journal ,vol. 2, no. 5, pp. 370–380, 2015.[25] X. Zhang, Z. Yang, W. Sun, Y. Liu, S. Tang, K. Xing, and X. Mao,“Incentives for mobile crowd sensing: A survey,”
IEEE CommunicationsSurveys & Tutorials , vol. 18, no. 1, pp. 54–67, 2016.[26] H. Jethawa and S. Madria, “Incentive mechanism for data-centricmessage delivery in delay tolerant networks,” in
Distributed ComputingSystems (ICDCS), 2017 IEEE 37th International Conference on . IEEE,2017, pp. 2551–2554.[27] S. K. Fayazbakhsh, Y. Lin, A. Tootoonchian, A. Ghodsi, T. Koponen,B. Maggs, K. Ng, V. Sekar, and S. Shenker, “Less pain, most of the gain:Incrementally deployable icn,” in
Proceedings of the ACM SIGCOMM2013 Conference on SIGCOMM , ser. SIGCOMM ’13. New York, NY,USA: ACM, 2013, pp. 147–158.[28] P. Hui, J. Crowcroft, and E. Yoneki, “Bubble rap: Social-based for-warding in delay-tolerant networks,”
IEEE Transactions on MobileComputing , vol. 10, no. 11, pp. 1576–1589, 2011.[29] H. M. Taylor and S. Karlin,
An introduction to stochastic modeling .Academic press, 2014.[30] L. Wang, S. Bayhan, J. Ott, J. Kangasharju, A. Sathiaseelan, andJ. Crowcroft, “Pro-diluvian: Understanding scoped-flooding for contentdiscovery in information-centric networking,” in
Proceedings of the2Nd International Conference on Information-Centric Networking , ser.ICN ’15. New York, NY, USA: ACM, 2015, pp. 9–18. [Online].Available: http://doi.acm.org/10.1145/2810156.2810162[31] A. Ker¨anen, J. Ott, and T. K¨arkk¨ainen, “The one simulator for dtn pro-tocol evaluation,” in
Proceedings of the 2nd international conference onsimulation tools and techniques . ICST (Institute for Computer Sciences,Social-Informatics and Telecommunications Engineering), 2009, p. 55.[32] A. Vahdat, D. Becker et al. , “Epidemic routing for partially connectedad hoc networks,” Technical Report CS-200006, Duke University, Tech.Rep., 2000.[33] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Spray and wait: anefficient routing scheme for intermittently connected mobile networks,”in
Proceedings of the 2005 ACM SIGCOMM workshop on Delay-tolerant networking . ACM, 2005, pp. 252–259.[34] S. Jain, K. Fall, and R. Patra,
Routing in a delay tolerant network .ACM, 2004, vol. 34, no. 4.[35] H. Liu, V. Ramasubramanian, and E. G. Sirer, “Client behavior and feedcharacteristics of RSS, a publish-subscribe system for web micronews,”in
Proceedings of the 5th ACM SIGCOMM conference on InternetMeasurement . USENIX Association, 2005, pp. 3–3.[36] L. Breslau, P. Cao, L. Fan, G. Phillips, and S. Shenker, “Web caching andzipf-like distributions: Evidence and implications,” in
INFOCOM’99.Eighteenth Annual Joint Conference of the IEEE Computer and Com-munications Societies. Proceedings. IEEE , vol. 1. IEEE, 1999, pp.126–134. Hasan M A Islam received his Bachelor de-gree in Computer Science and Engineering fromBangladesh University of Engg and Techologywhich is the top ranked university in Bangladesh in2008. In 2013, he received his M.Sc degree in Net-working and Services from University of Helsinki.He is currently working as a Doctoral Candidatein the Department of Computer Science, Aalto uni-versity. His research interests include InformationCentric Networking, Communication Network Ar-chitecture, and Network protocols.
Dimitris Chatzopoulos received his Diploma andhis MSc in Computer Engineering and Commu-nications from the Department of Electrical andComputer Engineering of University of Thessaly,Volos, Greece. He is currently a PhD student atthe Department of Computer Science and Engineer-ing of The Hong Kong University of Science andTechnology and a member of Symlab. His mainresearch interests are in the areas of device-to-deviceecosystems, mobile computing, mobile augmentedreality and cryptocurrencies.
Dmitrij Lagutin received his M.Sc (Tech) degreein 2005 from Helsinki University of Technologyand a D.Sc. (Tech) degree in 2010 from AaltoUniversity, Finland. He is currently working as aproject manager and postdoctoral researcher in theEU Horizon 2020 POINT project. Previously heworked as a researcher in several research projectsat Helsinki University of Technology and AaltoUniversity, including EU FP7 PSIRP and PURSUITprojects. His research interests include network se-curity and privacy, future network technologies, andthe Internet of Things.
Pan Hui received his Ph.D degree from the Com-puter Laboratory, University of Cambridge, andearned both his MPhil and BEng from the De-partment of Electrical and Electronic Engineering,University of Hong Kong. He is currently a facultymember of the Department of Computer Science andEngineering at the Hong Kong University of Scienceand Technology where he directs the HKUST-DTSystem and Media Lab. He also serves as a Dis-tinguished Scientist of Telekom Innovation Labo-ratories (Tlabs) Germany and an adjunct Professorof social computing and networking at Aalto University Finland. Beforereturning to Hong Kong, he spent several years in T-labs and Intel ResearchCambridge. He has published more than 150 research papers and has somegranted and pending European patents. He has founded and chaired severalIEEE/ACM conferences/workshops, and has been serving on the organisingand technical program committee of numerous international conferencesand workshops including ACM SIGCOMM, IEEE Infocom, ICNP, SECON,MASS, Globecom, WCNC, ITC, ICWSM and WWW. He is an associateeditor for IEEE Transactions on Mobile Computing and IEEE Transactionson Cloud Computing, and an ACM Distinguished Scientist.