Routing And Searching In Bittorrent Network

As thers is more and more multimedia information appears on the internet, people need toexchange a huge number of binary files like video, audio, etc. The traditional way ofexchanging is Client/Server Paradiam. In this paradiam, clients receive data only, while serverssend data only. When the number of clients and the size of files increasing, servers soonbecome the bottle neck of the whole system. P2P (Peer to Peer) is a new service paradiam.Inthis paradiam, every node in the network is both client and server, and both sends and receivesfiles. P2P paradiam solves the bottle neck problem in Client/Server paramdiam, and makes fileexchanging much faster in a lower price.Bittorrent is one of the most important peer-to-peer file sharing systems, and is loved by moreand more people because of its friendly interface and high performance.In order to obtainresources in Bittorrent network, people need to find corresponding torrent file first.Unfortunately, there is no effective method of searching torrent files at the moment. Peopleneed to visit a lot of well-known torrent publishing sites, and find the right torrent manually.itis a waste of time and energy. We analyzed various properties of topology of Bittorrent networkand torrent files, and proposed a distributed way of torrent file searching method.Our mainwork in this thesis includes:(1) After analyzing a huge number of torrent files, we find them contained abundant metainformation, and very suitable for indexing and searching. After analyzing thetopology of Bittorrent network, we find there are many peers belonging to more thanone swarm, called switch peers. We can use these switch peer to connect isolatedswarms, and build a unified unstructed P2P network. These attributes prove it ispracticable to develop a Torrent Search Overlay-TSO.(2) We propose connecting policy of TSO, and anylize its connectivity. We find thatcoverage rate of switch peers is 99.5%, coverage rate of torrent files is 93.5%. Thedata above means the connectivity is quite well.(3) We improve Prinkey's Content Query Routing algorithm, so we can support fizzysearch and do not need global hash functions any more.(4) We propose an opertimized neighbour selection algorithm for Bittorrent network, andmake routing and searching more efficient. Our results indicate it achieves in somecases 84% recall rate by using only 43% of the messages used in the floodingalgorithm.