Uncover the Peer Distribution in BitTorrent

Peer-to-peer traffic constitutes more than 60% of today's Internet traffic, resulting in high bandwidth cost for ISPs. Recent efforts have been made to modify BitTorrent clients to reduce inter-ISP traffic. Although the results have been encouraging, recent research also reveals that global adaptation of such an approach may harm the download time as there is no clear evidence of persistent clustering in all ISPs. To this end, many large scale measurements on BitTorrent topology have been conducted by analyzing different snapshots of the BitTorrent network. However, the analysis overlooked the download time, the actual contributions of peers, and the distribution of peers throughout the file download period since the snapshots were obtained by querying the tracker for IP addresses of peers at a certain time. In this paper, we seek to understand to what extent the distribution of peers in BitTorrent relates to their contributions in data swarming and transmission rates by studying real BitTorrent download traces. In order to present an unbiased view, we collected the traces from over 100 different files, including books (in different languages), music (in different languages), movies, and software (for different operating systems). The file size ranges from 4 MB to 4 GB. We also compared traces from a regular BitTorrent client with an ISP-friendly BitTorrent client to examine the actual impact of an ISP-friendly algorithm on download time and peer contributions. Our major findings include that distance has no effect on the download rate in general, seeds or lechers cannot always be found within the same ISP, and a torrent can only benefit from an ISP-friendly approach in certain situations. Suggestions are given on how BitTorrent clients can be more ISP-friendly without sacrificing download rate.