Peer-to-peer (P2P) file-sharing applications such as BitTorrent have attracted a fair amount of controversy, due in no small part to their widespread use in disseminating copyrighted material. They are nevertheless a subject ripe for research. By splitting large files into pieces and allowing peers involved in a download to exchange missing pieces with one another, P2P protocols have allowed significant amounts of data to be distributed in a decentralized manner. Estimates from 2011 place BitTorrent traffic at 17.9 percent of all Internet traffic. Decentralized networks of such a scale provide an endless source of optimization problems.

Arguing that existing content distribution policies may still result in a significant deviation in the amount of upload bandwidth contributed by users, the authors of this paper aim to tackle the specific problem of fairness in P2P applications, and BitTorrent in particular. While the original tit-for-tat algorithm used by BitTorrent was intended to alleviate this problem, it relies on estimates of neighboring peers’ upload rates as a selection mechanism. According to the authors, this does not result in fair bandwidth exchange because much time is wasted discovering peers with similar upload rates. The authors propose instead a deficit-based distributed algorithm named FairTorrent that rewards peers in proportion to the amount of bandwidth they share.

The core contribution of the paper is a description of the FairTorrent algorithm itself, covering the behavior of both leechers (users seeking to download a file or part of a file) and seeders (users with the complete file). In summary, leechers prioritize peers to whom the most data is owed on a block-by-block basis when determining where to send the next data block. Seeders on the other hand simply send blocks on a round-robin basis to known leechers.

The authors argue that, beyond guaranteeing a high degree of fairness, the algorithm provides fast convergence, high utilization, and resilience to strategic peers. Eschewing theoretical analysis in favor of experimental validation, the authors run an implementation of their algorithm through a variety of scenarios and against a number of existing BitTorrent clients, including live BitTorrent swarms. The results show significantly improved download rates and faster rate convergence, even when surrounded by low contributing and dynamic peers.

It is debatable whether fairness, as defined by the authors, is a property sought by all BitTorrent users. However, there is no doubt that tackling the issue of free riding, whereby some users consume much download bandwidth while sharing little of their own upload capacity, is essential to ensure that these protocols remain a viable and scalable option for the distribution of data. This is particularly important if it is shown that download times are negatively impacted for those that do contribute. The authors have demonstrated in seemingly realistic case studies that the presented approach can substantially increase download speeds. As such, the results of this paper will be of great interest to those turning to P2P systems to achieve large-scale content distribution.