Agent Selection And P2P Overlay Construction Using Global Locality Knowledge

It is estimated that Peer-to-Peer(P2P) traffic has already accounted for 50-70 percent of the total traffic on the Internet. This work aims to improve the performance of P2P file-sharing applications by constructing locality-aware P2P networks and enabling locality-aware downloading. A P2P agent's locality information, e.g, the prefix and autonomous system(AS) number in which it sits, can be abstracted from its IP address. This information can then be used to identify whether P2P agents belong to different ASes, the hop-distance between the ASes can be inferred using the global locality knowledge obtained from measurements of the AS-level Internet topology. It is expected that the file-sharing process should be more efficient when choosing to download from the nearest file provider, and the total amount of P2P traffic generated onto the Internet should be reduced as well. Preliminary simulation suggests that by using the global locality knowledge, the P2P enquiry process can be modified to return more file providers that are close to the enquirer without comprising the success rate of file searching.