Research On Searching Strategy Of Rare Items In Unstructured P2P Networks

With the rapid development of internet, peer-to-peer overlay networks have become more and more popular. Millions of users have started using peer-to-peer file sharing systems, such as Gnutella and BitTorrent, to publish their contents and search for desired files. Due to the distributed, self-organized nature of peer-to-peer systems, resource locating has been a fundamental issue and draw a lot of researchers'attentions worldwide. The objective of a search scheme is to successfully locate resources while incurring low overhead and delay. A number of search algorithms have been proposed for effective resource locating.This paper has intensively studied the search algorithms for unstructured P2P networks in the past. Through the investigation, we have found that existing search method is efficient for locating popular items, and it's less efficient when locating rare items. Therefore increasing the search efficiency for rare items would drastically improve the network services and user experience.Base on the questions above, a novel search algorithm which especially concerns the rare items locating problem is proposed. We first provide a comprehensive theoretical survey of various state-of-the-art search schemes in unstructured P2P networks for file-sharing applications then we analysis the random graphs theory and Power-law networks for P2P networks modeling. Then, we present a data structure TDBF for calculating the file's local popularity based on which nodes can decide to place replica and search schemes. The idea of TDBF is derived from Bloom Filter, a data compress technique which controls the storage cost in a low level. Second a replica strategy and a dynamic search strategy for rare items are proposed. In the process of our dynamic search strategy, a query issuer first calculates the popularity of a resource that issued in the query. Then it can decide which search scheme will be chosen for item locating. We also proposed a node join/leave management method which decreases the negative effect of churn on P2P networks. In this algorithm we use high degree nodes to exploit the heterogeneity of the P2P networks and also consider the node load balance. Besides, a theoretical analysis of algorithm efficiency is also presented. At last, we conduct comprehensive simulations to evaluate the algorithms. Results show that our method significantly improves the search quality especially on rare items as well as reduces system traffic cost compared with existing approaches.