A Redundancy-based Improvement On Partially Decentralized P2P Structure

P2P system, which is based on partially decentralized structure, combining with all the characteristics of decentralized topology and centralized topology, has features of good management of Napster and characterizes of robustness and complexity-supporting of Gnutella. Owing to the proper management and extensity, it is widely applied in the true world. However, this structure has its own shortages: the frangibility of super-peer may cause network unstable and be largely dependent on super-peer; as the reputation of super-peer is not reliable, it may leave maliciously; the heavy load may lead super-peer out of work.By learning P2P system based on partially decentralized structure, inheriting the advantages of partially decentralized system and finding the bottleneck problem of super-peer node model, this paper set up a new type of super-peer topology based on partially decentralized structure named SRSP strategy. In order to avoid increasing connection number between super-peers brought by redundancy mechanisms, using the strategy of redundancy, we make up a sharing mechanism to manage and organize redundant nodes, which reduced the pressure on the super-peer from original topology; Based on the degree of network nodes following the principle of power-law distribution, we design selection algorithm integrated hardware performance, reputation and degree of peer neighbor to select super-peer. Winners of the super-peer node will perform tasks more trustworthy. This can reduce the fluctuations in the network caused by super-peer malicious leaving, void the super-peer active and passive failure, achieve the balance of time and space complexity and load balance and system stability.In order to verify the effectiveness of improved strategies, we use simulation platform, rewrite the traditional partially decentralized protocol by adding strategy described in this article. Respectively running our rewriting model and traditional model in the simulation platform, we mainly compare the load situation of super-peers, the rate of successful search, the average delay and the average steps. By the test results we can see that the super-peer Situation is more balanced, and our improved model is more stable.