Study On Searching Methods For Peer-to-Peer Networks Based On Ontology And Clustering

The Peer-to-Peer (P2P) system is an application layer network that enables the networked computers to share the resources in a distributed manner. An important problem in P2P systems is how to search efficiently the resources of the other peers. The existing search techniques for P2P sysyem are usually not highly efficient because most of them are based on the idea of flooding the network with queries and/or the way of keyword search. The former limits the scalability of P2P systems because of a large of messages generated, propagating each query to all reachable peers up to a specific number of hops from the initiator. The latter does not support the complex queries or semantic queries, thus limits the quality of the retrieved results. This paper applies Ontology and clustering techniques to study the P2P searching algorithms.According to the property of power law of P2P networks, the paper presents a feedback-based hybrid search algorithm combining BFS with DFS strategies. This algorithm forwards the query messages to a subset of neighbors at some high-degree nodes and to only ones at other nodes, and utilizes the feedback information from the previous searches to guide current ones in the searching process in order to select the right neighbors to forward queries. The experimental results show that this algorithm achieves high success rate and query hits and search efficiency, and reduces redundant messages efficiently. Furthermore, it has a good self-adaptive ability to adapt rapidly the dynamical changing network behaviors.In this paper, a shared peer ontology is designed for semantic extracting and knowledge representing for each node that the nodes can easily understand and process the knowledge of each other. It also gives out the relevant query representing and processing methods. In the searching process, each node records all relevant replies that it receives and periodically evaluates rank of each neighbor. By reconnecting to a new node with the higher query hits than that of one of its neighbors and discarding an old connection to a neighbor with the lowest rank, a node dynamically clusters the nodes with the most similar interests shared by it. Based on these works, a searching algorithm for P2P networks based on Ontology and clustering is proposed. The experimental results show that this algorithm achieves high precision and recall rate, and decreases about 2 hops of the searching depth required to find the desired results, compared to the gnutella search algorithm. It can return more relevant replies faster.