Research Of Peer Selection Algorithm Based On File Availability In BitTorrent System

Peer-to-Peer applications are extensively used on internet nowadays, and BitTorrent is one of the most successful P2P file sharing system. Statistics indicates that by the end of 2004, 35% of the internet traffics are caused by BitTorrent protocol.The characteristic that BitTorrent differs from other P2P systems is that it splits the files into smaller pieces. Clients that do not have the whole file can also provide some pieces to other node. So the clients who are in the process of downloading can also be service providers. This remarkably increases the quantities of service providers in BitTorrent system, and thus greatly improves the performance of the system.The key algorithms in BitTorrent protocols are piece selection and node selection. Plenty of studies show that the rarest first strategy and the choke algorithm work well. The first balances the number of copies of each data piece among the BitTorrent nodes to help peers exchange data, and to avoid the bottle-neck. The latter is used by peers to find the appropriate service provider and speed up their downloading.We propose the concept of file availability based on investigations of the BitTorrent protocol. The rarest first piece selection algorithm of BitTorrent protocol only takes the number of piece copies into consideration regardless of their availability. Neighbor nodes selection algorithm is a strategy that search its neighbor peer blindly, and may take quite a long time to reach its appropriate service providers. Based on file availability, we improve the service provider selection strategies to help peers find the right service providers in a relatively short time. Experiment results show that this approach can speed the downloading rates up to 30%.