Congestion aware data dissemination in social opportunistic networks

Recent developments in social-based opportunistic forwarding [1], [2], [10] have identified that load is unfairly distributed towards nodes which are better connected. Unfair load distribution and high unrestricted volumes of traffic produce congestion. Congestion in opportunistic and delay tolerant networks takes the form of persistent storage exhaustion [1][4]. Several solutions exist including slowing sources, using alternative routes, discarding traffic, or migrating messages to alternative storage locations. We focus on congestion aware forwarding algorithms that adaptively choose the next hop based on contact history, predictive storage and delay analysis in order to distribute the load away from the storage hotspots and spread the traffic around. We describe new forwarding heuristics that uses nearby nodes with available storage and low delays to store data that would otherwise be lost during opportunistic bulk data transfers. More specifically we describe the two metrics Receptiveness and Retentiveness, each is a measurement of how utilised a node is, regarding node delays and buffer capacity respectively. Our heuristics factors in a cost metric associated with storage and delays so that the traffic is spread away from ego-network-centric paths and only paths with more storage and lower delays are chosen at times of congestion. Our heuristics makes use of the diversity of forwarding paths in human contact networks [7] by forwarding packets along multiple paths similarly to the resource pooling principal [6]. For our evaluation, we perform an extensive set of trace driven simulations for an opportunistic peer to peer file casting application. We build a realistic interest-driven P2P content dissemination overlay on the top of our congestion aware forwarding protocol. Our results show that our availability heuristics dramatically increases robustness of content distribution in terms of shorter download times and higher ratio of queries being successfully solved in the face of increasing congestion levels while still maintaining high availability among nodes when compared to SimBetTs[11] and Fairroute[4]. We induce congestion levels by increasing the topics popularity (number of randomly chosen subscribers) and the number of randomly chosen file publishers from 3% to almost 100%.