CBT: A proximity-aware peer clustering system in large-scale BitTorrent-like peer-to-peer networks

In a large-scale BitTorrent-like peer-to-peer file sharing system, the track server could be overloaded to update the state information of constantly arriving and leaving peers. Upon the connection request from a peer, the track server responses with a random list of peers and such randomly selected peers among the whole peer-to-peer network could create a long delay of file sharing between two peers. To improve the file sharing performance, we propose a hierarchical architecture to group peers into clusters according to their proximity in the underlying overlay network in such a way that clusters are evenly distributed and that the peers within each cluster are relatively close to each other. We achieve this by constructing the CBT (Clustered BitTorrent) system with two novel algorithms: a peer joining algorithm and a super-peer selection algorithm. We develop a fluid model to compare the performance of the proposed CBT system with a original BitTorrent system. With this model, we find that the CBT system quite effectively improves the performance of the system. Finally, simulation results are given, which demonstrate that the CBT system achieves better results than a randomly organized BitTorrent network, improving the system scalability and efficiency while retaining the robustness and incentives of the original BitTorrent paradigm.