Peer-to-peer data searching using relevant peer selection and local availability estimation

Abstract

Due to the dynamic, robust and resource-intense environment of the peer-to peer systems, traditional approach of searching has become obsolete as systems becomes bigger and wider. Lack of knowledge, administrative and peers' churning are among the obstacles for researchers to develop an efficient searching mechanism. Proper peer selection is one way of deriving effective searching mechanism. Peers that are considered of relevance, should be chosen to receive the search query in order to increase the search efficiency and obtain better retrieval rates. In this thesis, two approaches have been developed to produce efficient searching mechanism in peer-topeer system. The first contribution is the flood-based relevant search method where peer relevance determination is based on the query feedback data. Peer relevance is based on whether the peer has answered previous queries and how much the queries answered have content similarity with the ongoing query. A metric graph has been proposed to calculate the relevance of peers based on the values of their query hits and query similarity with the current ongoing query terms. The method is then improved by implementing a dynamic threshold value for relevant peer selection based on the nearest-neighbor concept. The experimental results showed that, the proposed method managed to record high efficiency in search results by having high query hits with low number of messages used. The research also proposed a trust based search approach for unstructured peer-to-peer system. It is developed to reduce the effect of nodes churning on the system. In the trust-based method, three components, namely peer relevance, peer availability and query hops are used. The model uses the boolean concept in decision making for selecting trusted peers. The results obtained showed that the effect ofnodes churning in query hits have been reduced from four percent to two percent.