Skip to Main Content
This paper describes an experimental study that closely examines the underlying topologies of multiple complex networks formed in BitTorrent swarms. Our results demonstrate that the networks exhibit fundamental differences during different stages of a swarm, suggesting that the initial stage is not predictive of the overall performance. We also find a power-law degree distribution in the network of peers that are unchoked by others, which indicates the presence of a robust scale-free network. However, unlike previous studies, we find no clear evidence of persistent clustering in any of the networks, precluding the presence of a small-world that is potentially efficient for peer-to-peer downloading. These results suggest an interesting venue for improving BitTorrent's performance. We present a first attempt to introduce clustering into BitTorrent. Our approach is theoretically proven and makes minimal changes to the tracker only. Its effectiveness is verified through a series of simulations and experiments.