| The rapid development of P2P technology makes it as one of the most disruptive tools for the construction of large-scale distributed system over Internet. Therefore, as one of the most perspective applications, P2P content distribution system has become one of the hotspot of P2P research. BitTorrent is a popular P2P content distribution system, which with the central point of that: a sharing file is divided into many of equivalent file blocks. In the same time, each node can downloads and uploads lots of file blocks. However, as the asymmetric distribution of file blocks and the dissimilar capability of each peer, BitTorrent is confronted with the challenge of how to optimize peer selection strategy and file block selection strategy for improving system's performance, in order to increase the system resources utilization and reduce file download time.There will be an intensive study of peer selection strategy and file block selection strategy in this paper. BitTorrent adopts the rate-based Tit-For-Tat choke algorithm and Optimistic Unchoke (OU) algorithm as a peer selection strategy. Each upload peer uploads file blocks for five requesting peers, which employs the rate-based TFT algorithm to preferentially choose four requesting peers with the highest file blocks uploading rates for it and employs the OU algorithm to randomly choose one from other requesting peers. However, the upload peer who uses the rate-based TFT choke algorithm only chooses the neighbor who makes direct contribution for itself to use TFT upload quota. So this strategy restricts the selection range of TFT upload quota, which cause that many requesting peers are waiting for download while many peers' TFT upload quota being unoccupied. Moreover, the rate-based TFT can't effectively ensure fairness between download and upload. In order to solve the issues mentioned above, a global contribution-based peer selection strategy in BitTorrent from the angle of global and make the best of system resource is proposed in this paper: We calculate peer's global contribution in period of time from both direct and indirect contribution. The upload peer would preferentially choose the requesting peers who make more global contribution. In this way, it can avoid the problem of low resource utilization rate and unfairness between download and upload which due to partial contribution. Our experimental results show that the contribution-based peer selection strategy can improve the resource utility of TFT upload quota and decrease the average time consumption for completely downloading the whole file by more than 88%-95%.BitTorrent mainly uses Local Rarest First (LRF) algorithm as a file block selection strategy, in which the download peer preferentially chooses the file block which is lacking in itself and is rarest in its neighbors. Because each download peer only knows the local information of its neighbors' file blocks, local information is rarest doesn't mean global information is rarest. The LRF strategy is blinding in file blocks selection and can't ensure all of uploaded file blocks grow equally. In order to solve such problem, a Global Rarest First (GRF) file block selection strategy based on seedoverlay network from the angle of global is proposed in this paper. Each seed peer can retrieval the global information of file blocks uploaded through seedoverlay network, when a seed peer responds the download peer's request for a file block, it can upload the file block that is global rarest and interested by the download peer. Our experimental show that the global rarest first block selection strategy can ensure the equilibrium growth of all of uploaded file blocks and greatly improve the utility of system resource and decrease the expected average download time of file by more than 83%-99%. |
PDF Full Text Request