Towards the coevolution of incentives in bittorrent

Abstract

BitTorrent is a peer-to-peer file sharing system that is open to variant behavior at the peer level through modification of the client software. A number of different variants have been released and proposed. Some are successful and become widely used whereas others remain in a small minority or are not used at all. In previous work we explored the performance of a large set of client variants over a number of dimensions by applying Axelrod’s round-robin pairwise tournament approach. However, this approach does not capture the dynamics of client change over time within pairwise tournaments. In this work we extend the tournament approach to include a limited evolutionary step, within the pairwise tournaments, in which peers copy their opponents strategy (client variant) if it outperforms their own and also spontaneously change to the opponents strategy with a low mutation probability. We apply a number of different evolutionary algorithms and compare them with the previous non-evolutionary tournament results. We find that in most cases cooperative (sharing) strategies outperformed free riding strategies. These results are comparable to those previously obtained using the round-robin approach without evolution. We selected this limited form of evolution as a step towards understanding the full coevolutionary dynamics that would result from evolution between a large space of client variants in a shared population rather than just pairs of variants. We conclude with a discussion on how such future work might proceed.