Study On Search Scheme In Structured P2P Network

Peer to peer network (P2P), in which every node can share a large amount of data with others, is a revolution to the traditional Client/Server network architecture. With the scale of the P2P network growing, it can rapidly accumulate inflated information and provide a rich information warehouse for users.Present distributed system based on P2P can be classified into unstructured and structured systems. Unstructured systems are fit for the highly dynamic network environment of P2P, for they maintain loose network overlay. But because of the random query forwarding scheme each node applies, it's search efficiency is low. The limited flooding mechanism improves the search efficiency, but causes poor system scalability and low search success guarantee. Structured system employs a scalable scheme to guarantee search success, but it needs to maintain strictly controlled overly network, it is difficult to adapt to highly dynamic network environment of P2P. More important, the use of DHTs leads to the fact that only keyword exact-match is available. Abundant shared resources of P2P systems attract large amount of users, but it's difficult to find desirable resources from the huge amount of resources. Consequently, an efficient search scheme for users to retrieve needed data quickly and accurately is required emergently.Because of the diversity and complexity of network resources, it is necessary to characterize each resource with several properties from various points of view so as to express users' requirements more accurately and to make the search result meet the users' requirements better. Based on distributed hash table (DHT) idea, a Multiple Properties Publish System is designed in this paper. Using the ability of DHT that it is able to map a single key to a peer in the network, the system publishes the inverted indices of each property to certain peers respectively while publishing resources, and routes the search request to these peers possessing the inverted indices according to given searching conditions of each property while searching resources. It's required for the system to collect each property and compute the intersection of them as the result of search. So the system avoids the problem that structured P2P only supports exactly searching.Next, the author analyses some important techniques to improve the system, such as VSM, TFIDF and Bloom Filter. And it discussed how to apply them on searching text resource.At the end of this dissertation, it gives out a summary of the research work and proposes the further research in this field.