Cyclic ranking in single-resource peer-to-peer exchange

Abstract

Peer-to-peer (P2P) sharing systems use incentives for resource exchange to encourage cooperation and ensure fairness. In bilateral strategies, such as BitTorrent Tit-for-Tat or deficit-based FairTorrent, individual decisions of peers utilize direct observations. It may result in low performance and unfair treatment. In this paper, we study a novel exchange strategy that applies Cyclic Ranking (CR). In addition to direct observations, a peer utilizes provision cycles---a shared history of effective exchanges. The PageRank algorithm runs for the locally collected cycles and computes the numerical ranks to estimate the reputation. The CR strategy incrementally augments known incentive-aware strategies. For evaluation we implement CR-BitTorrent and CR-FairTorrent variants. Our simulation model captures the dependence on network bandwidth and the number of seeders as well as selfishness and stability of the participants. The initial experiments show improved fairness and download times, compared to the original BitTorrent and FairTorrent. The performance of selfish and unstable peers decreases by as much as 50%. The CR strategy suits well in environments where direct reciprocity has shown little effect. Contrasted to existing solutions, the CR strategy rewards longevity and stability of peers.