Distributed Detection of Selfish Routing in Wireless Mesh Networks

Community wireless mesh networks are vulnerable to free riders who refuse to forward data for others. As the naive selfish strategy of dropping data is readily detected, free-riders are motivated to subtly manipulate the routing protocols to minimize their workload while still evading detection. This paper presents an adaptive on-line algorithm to detect such selfish behavior based solely on local observations of messages exchanged by AODV-like routing protocols. We use a finite state machine model of locally observable protocol actions to generate a statistical description of the behavior of each neighbor and apply statistical analysis to cluster neighboring nodes on the basis of behavioral similarities and identify the selfish ones. Through simulation, we evaluate the performance of our method with respect to the probability of detecting selfish nodes and the rate of false positives against two generic selfish strategies— dropping route requests and dropping route replies. We also evaluate the effect of detection on an adaptive adversary who attempts to operate as selfishly as possible while still evading detection. We find that our technique can detect dropped route requests as well or better than a variant of the widely-used Watchdog detection method, with a lower rate of false positives. In the case of dropped route replies, a fundamental scarcity of observable routing events prevents any algorithm from performing well, suggesting the need to revisit the design of routing protocols.