Research On Key Technique Of BitTorrent-like Peer-to-Peer File Sharing Systems

With the rapid development of Internet, lots of new network application systems appear.As an efficient content localization and distribution tool, Peer-to-Peer(P2P) file sharing sys-tems have occupied a majority of Internet traffics. Whereas content localization has raised alot of attention in the last few years, content replication has started to be the subject of activeresearch recently. BitTorrent is a very popular and scalable file distribution system basedon P2P paradigm. It has been successful at distributing large files quickly and efficientlywithout overwhelming the capacity of the origin server. This dissertation focuses on the re-search of key technique of BitTorrent-like P2P file sharing systems, such as service capacity,incentive mechanism and overlay topology. Our research is beneficial to the improvementsof the performance, stability and scalability of P2P file sharing systems. The contents andcontributions of this dissertation are as follows:(1) Research on service capacity of BitTorrent-like P2P file sharing systems:the filedistribution performance of BitTorrent-like systems is analyzed through an architecturemodel based on parallel treen, also the file availability and effectiveness of file sharing inBitTorrent-like systems is studied. Moreover, an optimized neighbor selection strategy isproposed in this dissertation.As a very popular and scalable P2P file sharing system, BitTorrent has attracted a lotof interest lately. Many measurement researches have verified that the BitTorrent systemperforms well. However, the service capacity of BitTorrent is still not fully studied. In thisdissertation, an architecture model based on parallel treen is employed to analyze the filedistribution performance of BitTorrent-like systems. With this model, we validate that theBitTorrent system has very good properties to support a large number of downloaders, andsheds light on the effect of the number of file pieces and concurrent uploading connectionsof a peer on the file distribution performance. Besides, we present a mathematical model forpredicting the file availability in BitTorrent-like systems. It is observed only when the piecesare distributed as evenly as possible across the system can the file availability be maximized.Also it is validated that the rarest first policy is beneficial to the file availability. Further, we find BitTorrent is very efficient in sharing files. Moreover, an optimized neighbor selectionstrategy is proposed to overcome the drawbacks of random neighbor selection in BitTorrentsystems: that is, tracker server selects the peers that have the close number of file pieces to apeer as its neighbors. Simulation results verify our analysis and findings, and show that theneighbor selection strategy can further increase the network bandwidth usage, and improvethe performance of the system effectively.(2) Research on incentive mechanism of BitTorrent-like P2P file sharing systems: a ?uidmodel with two different classes of peers is developed to study the incentive mechanism ofBitTorrent-like systems, and a seed bandwidth allocation strategy is proposed which is basedstrictly on the contribution of peers.Cooperation is essential to a P2P file sharing system. However, it is difficult to promotecooperation among all individual participants without an effective incentive mechanism. Asa collaborative P2P file distribution tool, BitTorrent inevitably confronts the problem of free-riding,so BitTorrent built incentive mechanisms to reduce free-riding and increase usercooperation. However, the capability of BitTorrent on preventing free-riding is still not fullystudied. This dissertation develops a ?uid model with two different classes of peers to capturethe effect of free-riding on BitTorrent-like systems. With the model, we find that BitTorrent'sincentive mechanism is successful to prevent free-riding in a system without seeds, but maynot succeed in producing a disincentive for free-riding in a system with a high number ofseeds. The reason is that BitTorrent does not provide an effective mechanism for seeds toguard against free-riding. Therefore, a seed bandwidth allocation strategy based strictly oncontribution of peers is proposed. Simulation results validate our analysis, and show thatthis allocation strategy not only penalizes free-riding but also improves the performance ofcontributors effectively.(3) Research on overlay topology of BitTorrent-like P2P file sharing systems:a hier-archical architecture, called CBT(Clustered BitTorrent) system, is presented in this disserta-tion, which groups peers into clusters according to their proximity in the underlying overlaynetwork. The peer joining algorithm and super-peer selection algorithm is proposed to buildsuch a CBT system.In a large-scale BitTorrent-like P2P file sharing system, the track server could be over-loaded to update the state information of constantly arriving and leaving peers. Moreover,the random connections among peers in the overlay network make it possible for unfavorablydistant peers to connect, resulting in a long delay of file transmission and debasing the net-work bandwidth usage. In this dissertation, a CBT(Clustered BitTorrent) system is proposed to improve the file sharing performance of BitTorrent systems, in which peers are groupedinto clusters according to their underlying proximity, and peers priority exchange file withpeers in the same cluster. To build such a CBT system, we present two algorithms: the peerjoining algorithm and super-peer selection algorithm. A ?uid model is developed to comparethe performance of the proposed CBT system with an original BitTorrent system. With thismodel, we find that CBT system quite effectively improves the performance of the system.Simulation results demonstrate that CBT system achieves better scalability and efficiencywhile retaining the robustness and incentives of the original BitTorrent paradigm.