Toward Convergent Search for Large Peer-to-Peer Networks

Abstract

An efficient structured peer-to-peer (P2P) network should guarantee the search always convergent. This paper presents analytical models to characterize the process of locating recourses and the way of building routing table. Our models provide insight into peer-to-peer search and give the conditions of how to build suitable routing tables for greedy routing. Both in the ideal and the real world situations, the 2^k system is the best choice. This can explain why most of the existing structured P2P networks choose the 2^k system to build their routing tables. However, in order to get a better search performance, using the 2^k system requires that each peer should maintain a lot of neighbors. It would generate large load on the network participants in large P2P networks. We design and implement an efficient search-convergent peer-to-peer network using collision-allowed DHT (CADHT). The simulation experimental results show that CADHT P2P network can reach the same search performance as the One Hop Aware Kade but better than original Kade, while incurring almost the same communication cost as the Kade network but less than the One Hop Aware Kade. We also analyze the deep reasons in a numerical way.