Research On Overlay Management Strategy In Peer-to-Peer Based Platform

Development of traditional P2P applications is difficult and complex. Besides, isolation between different applications makes them not get full use of the whole resources. P2P platform is aimed at integrating the underlying heterogeneous resources, so it is convenient for the development of upper layer applications and sharing resources, improving system efficiency. Efficient and stable overlay management strategy is the core of a P2P platform. However, Current research of topology is mainly concerned about designing for some particular application, such as centralized topology, decentralized unstructured topology, hybrid topology, and etc. There are some advantages of each model, but their inherent disadvantages make them not able to be applied to platform.In order to solve the problem above, a new overlay management strategy is proposed in P2P platform Ripple, to construct a general-purpose structure for large scale system. The whole structure is divided into two layers, the upper part is composed by those peers with strong capability, called super nodes, while the lower part is composed by the others, called ordinary nodes. The relation between super nodes is not pure mesh. Every super node maintains a ring structure according to the delay of its neighbor peers, so it makes the upper nodes avoid using flooding mechanism. Every ordinary node maintains three parent nodes, including a directed one and two backups. The advantage is that if the direct parent node leaves, the ordinary node can find a new one quickly, decreasing turbulence of the lower peers. The standard of partitioning super node and ordinary node is according to their capacity. Peers with more capacity are considered as super nodes, while the others are as ordinary nodes, and their proportion would change according to the variation of the whole system's scale. There is a new mechanism for P2P search, including searching popular resource on neighbors based on Random Walker protocol, and searching rare resource on Distributed Hash Table, making the result more efficient and reliable.Ripple is based on Windows series and developed using Visual.NET platform. And the overlay management strategy of Ripple is implemented with C++. The performance of the system is evaluated by analyzing the log data which is generated by Ripple. The results show that the overlay management strategy is good for processing joining and leaving of the peers, whose login delay is all less than 1.6s. The number of neighbors on super nodes and number of father nodes on ordinary nodes is all close to the ideal number. 95% of the super nodes is among the most capable in the entire network, which improves the load balance of the system. The delay of resource search is within 2s, and 93% of peers can search the exact resource successfully. These show that the overlay management strategy obviously improves the platform's performance.