Abstract
Several approaches for distributing content over peer-to-peer (P2P) networks have been proposed in the last few years. They try to address some of the shortcomings of client-server architectures, most notably in terms of scalability, efficiency, reliability and fairness. CrossFlux is a P2P system for media streaming that incorporates these properties from the ground up in its design. The objective is to maximize the uniform effective throughput of every peer, even though some may have too little upstream bandwidth to serve the full stream (e.g., ADSL). Therefore, instead of relying upon classical tit-for-tat strategies that cap the download rate to the upload rate, reliability is coupled with fairness by rewarding peers that contribute more to content distribution with a higher number of alternate sources. This coupling is realized by using links (1) for content distribution in one direction and (2) as backup link in the opposite direction in the case of a peer failure. Furthermore, adaptive algorithms are used to dynamically reorganize the peers and find a suitable position for newcomers, in order to maximize the throughput and distribute the load among all participants. The evaluation of CrossFlux shows that recovery of node failures is fast due to the backup links, efficiency is increased with the help of self-adaptive techniques, and the load is well distributed among the peers.