Based On P2p Distributed Storage System

With explosive growth of internet service, users'requirements of mass data are increasing. How to design high available and high reliable distributed storage system has become a hot issue. The mature of P2P technology pushes network storage ahead further. Nowadays, international internet corporations have raised"Cloud Storage"concept for organizing system resources that users only need to pay a small amount of money to get stable and reliable storage service.Current available distributed systems haven't considered for providing high efficient storage service by taking advantage of coordination of server and front-end. This thesis first introduces distributed storage and P2P technology, then development status is studied both home and abroad. After that, high performance P2P Distributed Storage System is designed for evolution of end storage by constructing network overlay with server and front-end according to project's real demand.The major work of this thesis is showed below:First, PPDSS's architecture is designed. Metadata indexes and data loads are distributed among all servers to get rid of central node. Basic file operations are implemented, such as publish, download, update and delete. Various kinds of timer are designed for dealing with timeout events.Second, load balance module is designed and implemented to make good use of server's resources and avoid bottleneck or shake problem in system.Third, P2P content distribution strategy is implemented. Single thread mode leads to high network throughput and efficient disk access. Based on the properties of P2P network, Super Node Selection Algorithm is implemented to pursue high performance of content distribution by sending data to neighbor nodes preferentially due to their higher upload speed.Fourth, Lazy mechanism is designed and implemented to guarantee data consistency.Clients are used to run tests against PPDSS. Tests show that basic file operations are running correctly to provide right data for clients. Selecting low load server to spread replica does balance system load and reduce effect to clients. Compared to multi-process server, CPU occupation of single thread server based on non-blocking network process and asynchronous disk access is brought down 60%. Clients'average downloading time is cut down by using Super Node Selection Algorithm. Tests prove 17% downaloading time is saved compared to normal selection strategy.