Towards better content dissemination applications for Disruption Tolerant Networks
Amir Krifa · HAL (Le Centre pour la Communication Scientifique Directe) · 2012
The rapid proliferation of advanced mobile devices has created a growing demandfor data content. Existing approaches cannot keep up with the large volume ofcontent generated and requested, without the deployment of new expensive infrastructure.Exchanging content of interest opportunistically, when two nodes are inrange, presents a low cost and high bandwidth alternative for popular, bulky content.Yet, efficiently collecting, storing, and sharing the content while preventingselfish users from impairing collaborative ones, poses major challenges.In this thesis, we start by discussing the state of the art in terms of proposedsolutions for both point-to-point content routing and point-to-multipoint contentsharing solutions in DTN(s) (Delay Tolerant Network). Our main observationswere (i) despite a large amount of effort invested in the design of efficient routingprotocols for DTN, there has not been a similar focus on storage management andscheduling policies, and (ii) in addition to dealing with the resources managementchallenges, distributed (or peer-to-peer) content sharing systems over non-altruisticmobile devices have one more important issue to deal with: to ensure enough nodescollaborate to make the system interesting to participants. This latter goal is oftenconflicting with optimal resources management policies.Following this preliminary study, we try to solve the highlighted problems intheir foundations. We focus first on the problem of optimal resource managementin the context point-to-point content routing through a DTN. This problem wasfirst studied in scenarios related to environment and habitat monitoring based onsensor networks [1, 2], in project willing to connect rural villages [3, 4, 5, 6, 7], andeven in scenarios related to space technologies based on DTN protocols [8, 9]. Wepropose a practical and efficient joint scheduling and drop policy that can optimizedifferent performance metrics, such as average delay and delivery probability. Wefirst use the theory of encounter-based message dissemination to derive the optimalpolicy based on global knowledge about the network (GBSD, Global knowledgeBased Scheduling and Drop). Then, we introduce a method that estimates allnecessary parameters using locally collected statistics. Based on this, we derive adistributed scheduling and drop policy that can approximate the performance of theoptimal policy in practice (HBSD, History Based Scheduling and Drop). Finally,we study how sampled statistics can reduce the signaling overhead of our algorithmand examine its behavior under different congestion regimes.In a second effort, we revisit the problem of optimal resource management inthe context of large scale interests-driven content sharing over non-altruistic mobiledevices. Our ultimate goal is to enable people, through a channel based architectureto express their interests, head out in the real world and wait to get notified whenevera content that matches their interests is retrieved. To achieve this, we proposeMobiTrade as a candidate architecture. MobiTrade is a utility driven trading systemfor efficient content sharing on top of a DTN. It does not only take care of thenetwork and device resources, but also carefully considers: (i) the propagation ofinterests of participating users, (ii) the matching of these interests to individual node mobility patterns, and (iii) the willingness of involved users to collaborate.While simple tit-for-tat (TFT) mechanisms can force nodes to give one to get one,dealing with the inherent tendency of peers to take much but give back little, theycan quickly lead to deadlocks when some (or most) of interesting content must besomehow fetched across the network. To resolve this, MobiTrade relies on a tradingmechanism that allows a node merchant to buy, store, and carry content for othernodes (its clients) so that it can later trade it for content it is personally interestedin. To exploit this extra degree of freedom, MobiTrade nodes continuously profilethe type of content requested and the collaboration level of encountered devices.An appropriate utility function is then used to collect an optimal inventory thatmaximizes the expected value of stored content for future encounters, matched to theobserved mobility patterns, interest patterns, and collaboration levels of encounterednodes.Both our resources management solutions for point-to-point DTN routing(HBSD) and our channel based content sharing architecture (MobiTrade) have beenvalidated respectively through extensive NS-2 and NS-3 simulations along with amultitude of synthetic mobility models and real mobility traces. Furthermore, inorder to ensure the feasibility of our protocols and offer them to users, we implementedrespectively HBSD and MobiTrade for the DTN2 reference architecture andon real Android powered mobile devices and did further experiments in real environments.