Directed Research Project(Nov 2006)
Keywords: BitTorrent; P2P; Performance
BitTorrent is one of the most popular Peer-to-Peer (P2P) content distribution applications over the Internet that significantly contributes in network traffic.
In BitTorrent, a file is divided into segments and participating peers contribute their outgoing bandwidth by providing their available segments to other peers while obtaining their missing peers from others. Characterization of BitTorrent is useful in determining its performance bottlenecks as well as its impact on the network.
In this study, we try to address the following two key questions through measurement: (i) What are the main factors that affect observed performance by individual peers in BitTorrent? , and (ii) What are the contributions of these factors on the performance of individual peers? To address these questions, first we examine the group-level and peer-level characteristics of BitTorrent using three tracker logs from different sources. Second, we use statistical analysis (namely rank correlation and linear regression) to determine and quantify the potential effects of both peer-level and group-level properties on the performance of individual peers.
We conclude that: (i) There is no single property that has dominant effect on the observed performance by individual peers, (ii) Outgoing bandwidth of each peer, average available content in the group and churn rate appear to have the most notable effect on peer performance and (iii) The behavior of the system in practice is rather complex due to the inherent dynamics in peer participation and content delivery as well as bandwidth heterogeneity and asymmetry.