Study Of Optimal Strategies Of Search Performance In Peer-to-Peer Networks Based On Characterization Of Resource Distribution

Peer-to-Peer(P2P) is a new network application model which is different from traditional Client/Server(C/S) model. In P2P networks, each peer has equal functionalities: each peer can act both as server and client, and the resource can be passed directly among peers without relying on the centralized server. Furthermore, peers can join in or leave from the network at any time to form dynamic network environment. In addition, P2P network will be run normally while a certain peer is invalidation. Compared with the C/S model, P2P has many advantages, such as robustness, scalability, self-organization, and the higher utilization of network resources. For these reasons, P2P is a valuable application in the Internet.The share of resource is the most important intention of P2P network, which must be supported by the location of resource. First in this paper, the disproportion of resource distribution and its effect on search performance are analyzed. Then the distributions of resource popularity and requirement are defined. And the two distributions are obtained by gossip-based aggregation in the overlay of networks of resource copies, which is constructed in the process of searching. The resources with high popularity and requirement are broadcasted in P2P network, and flooding search method can be improved by the location of resource of high requirement using index and high popularity using adaptive Time-to-Live(TTL). Second, efficient communication ratio(ERC) of peer is defined to compare the response performance between peers, with the purpose of making the peer with higher ECR has bigger degrees. The optimization model of peer degree is proposed along with the peer connection strategy, by the average ERC and average degree of P2P network, which are gained based on gossip mechanism. In the final, the two optimization strategies are evaluated through simulation by Brite and Peersim. The simulation results show that the strategies proposed in this paper are effective to improve the performance of resource search.