Research On Replication In Peer-to-Peer File Sharing System

With the rapid development of computer and Internet technologies, the traditional network model cannot meet tremendous emerging demands any more. The client/server model has to make possible transition, so that the new model, Peer-to-Peer network, is arising. P2P network is so fantastic that extracts much attention and has been widely used in varied fields.File sharing is a major application of P2P technology, as well as a power source of P2P development. As one of the main applications of Internet, P2P file sharing has much influence onto Internet usages and traffics. In P2P file sharing system, storage and communication overhead are distributed among network peers by copying and propagating files, to provide communication mode with high efficiency and low cost. Data replication technology is a usual and important data management mechanism in P2P file sharing system, which can improve availability and reliability of files and the overall system performance by increasing the number of files. Therefore, replication technology in P2P system is a subject of great research value, and the key guarantee for availability and reliability. The techniques about replica creation, consistency maintenance due to frequently file updating, and replica deletion are yet difficult issues without effective solutions.After investigating related works deeply, this dissertation focuses on three main issues of replication technology respectively: replica creation, replica consistency maintenance and replica deletion. Objects are to encourage peers providing storages, lower the overhead of consistency maintenance, make the best use of storage and improve network performance. The main structure of the dissertation is as follows:The dissertation introduces the development, definition and characteristics of P2P network, and gives its classification and applications. The dynamicity of network and noncooperation between peers are analyzed and discussed. Chapter 2 introduces the importance of replication, summarizes current researches, and analyzes disadvantages of existing replica technology.Aiming at replica creation, the dissertation presents replica strategy based on incentive mechanism using information economics about principal-agent relationship and incentive concept for reference. The cooperation of peers providing free storage space is measured by their cooperative-utility. Peers with higher cooperative-utility can obtain better service. In this way, peers are encouraged to provide resources and services, so that the success rate of replica creation can be improved, and then the system performance can also be promoted.Based on the incentive replica creation strategy above, this dissertation introduces presenter peers, proposes improved replica creation strategy based on multilevel-agents. When a peer does not save replica due to its objective conditions rather than its selfishness, then it can act as introducer to introduce appropriate peers with high cooperative-utility to the principal-peer. In this way, not only the success rate of replica creation but also efficiency of creation can all be improved.Focusing on the replica consistency maintenance, the dissertation proposes a replica consistency maintenance mechanism based on physical-location and bloom-filter to reduce consistency maintenance cost. According to peers'physical-location using community partition in complex networks, peers are partitioned into clusters, where physical-locations among peers are closer and updating is done within clusters, to lower time cost of consistency maintenance. Double-deck bloom-filter and redundancy secondary neighbors are used to improve updating-message propagation method, and reduce updating-message redundancy.In order to make the best use of storage resources, the dissertation presents replica deletion strategy based on gray prediction theory and cost, to delete the useless replications. Using gray model for reference, the use value of replica, also called replica activity, is forecasted, to improve accuracy of prediction. Then the cost of replica deletion is analyzed and quantified. The survival value of replica is gained by integrating the prediction results, deletion cost and other factors, and saving or deleting can be decided to avoid replica over removed and reduce thrashing.The simulation results show that the proposed replica creation strategy can improve the success rate of creation, the proposed replica consistency maintenance can lower time consuming and message cost, and the proposed replica deletion can select more appropriate replications to remove, avoid blind replications remove and reduce thrashing.