Given the unique characteristics of vehicular net-works, specifically, frequent communication unavailability andshort encounter time, packet replication has been commonly usedto facilitate data delivery. Replication enables multiple copies ofthe same packet to be forwarded towards the destination, whichincreases the chance of delivery to a target destination. However,this is achieved at the expense of consuming extra already scarcebandwidth resource in vehicular networks. Therefore, it is crucialto investigate the fundamental problem of exploiting constrainednetwork capacity with packet replication. We make the firstattempt in this work to address this challenging problem. We firstconduct extensive empirical analysis using three large datasetsof real vehicle GPS traces. We show that a replication schemethat either underestimates or overestimates the network capacityresults in poor delivery performance. Based on the observation,we propose a Capacity-Constrained Replication scheme or CCRfor data delivery in vehicular networks. The key idea is to explorethe residual capacity for packet replication. We introduce ananalytical model for characterizing the relationship among thenumber of replicated copies of a packet, replication limit andqueue length. Based on this insight, we derive the rule foradaptive adjustment towards the optimal replication strategy.We then design a distributed algorithm to dictate how eachvehicle can adaptively determine its replication strategy subjectto the current network capacity. Extensive simulations basedon real vehicle GPS traces show that our proposed CCR cansignificantly improve delivery ratio comparing with the state-of-the-art algorithms.

PDF