Research On Replication Management Technology In P2P File Storage System

With the development of computer and network technology, Internet expands in scale constantly; network bandwidth has also been improved, Internet-oriented large-scale distributed storage systems have also been developed rapidly. Efficient, scalable distributed storage structure became the object of extensive research.Distributed P2P-based file storage system is a data storage system based on peer-to-peer network, which can provide an efficient, scalable, and robust, load balancing of data access. However, due to the dynamic nature of P2P networks, the different popularity of files, the request of different files are not same, some peers become hot spots and bottlenecks of the system. In the P2P file storage system, replication is a key technology to enhance system performance and improve file availability. But the current replication management techniques have some problems, so designing an efficient, flexible, dynamic replication management policy has important and practical significance.This paper mainly studies on the replication management in P2P file storage systems. At first, this paper detailed analyzes current replication management techniques in the P2P file storage system, including the static and dynamic replication management policy, and pointed out problems in existing scheme. Based on the status and problems of current research and combined with the underlying network structure of P2P file storage systems, this paper presents two kinds of replication management policy: a super-peer based dynamic replication management policy and structured network based dynamic replication management policy, which are applied to different network structures. Replication management policy based on the super-peer network is that a super-node decides to make a copy of the file on the local or other super-peers according to the file access information of the nodes in its cluster; and Replication management policy based on structured network is that the nodes decides to store a replica on the key nodes in the query paths according to a graph of query paths. Finally, the two policies are simulated on the PeerSim simulation platform and the results of the experiment show that two kinds of policies can effectively reduce the load of nodes and address the hot issues, achieve load balancing, reduce latency and improve system performance in their own network environment.