Research On Reputation Models For P2P Networks Based On Cloud Model And Credible Recommender Set

The applications based on P2P technology is the most important applications of the Internet. P2P networking, where there are no centralized servers, has caused wide public concern as it overcomes defects in the usual web client/server model.However, a P2P system is large in scale, whose component nodes are dynamic, strange to each other and lack of necessary mutual trust. The security issues are very serious in P2P system. Reputation-based trust model, which imitate the trust mechanism in human society, has been proven to effective in enticing cooperation among peers, improving security and overall performance of P2P system, by building trust relationship between peers in P2P networks. Therefore, it is very important to study the reputation models in P2P networks.Based on the existing research on P2P reputation models, this thesis carried out studys in two aspects .The main achievement are as follows.(1) A cloud-based reputation model for P2P networks is proposed. Taking into account the uncertainty of trust, this model integrates cloud model into the traditional reputation model to measure the trust of peers and make trust decision in a more comprehensive way. The calculation of reputation value was considered in both these aspects: the historical average behavior the peer and the recent volatility of the peer. Reputation value was integrated with peer's recent volatility to severly punish the peers who behave in an oscillating manner. This model give priority to the peers with a higher behavioral stability and reputation value when making trust decisions to select peers performing the transaction.In experiments, this model was tested in different aspects and compared with the original reputation model. Experimental results show that this model can distinguish malicious node with complex strategies effectively and performs better in dealing with various dynamic behaviors of peers compared to the original model. What's more, this model improves the succeful transaction rate and can conquer the failure of the existing model under some cases.(2) A credible recommender set based reputation model for P2P networks is proposed.To select the sutible recommender and to solve the problem in existing models that peers'behavior change have effects on honesty checking in P2P system, each peer maintains a table to record the information of peers who has given a false recommendation. The recommendations from the peers not in the table which was called the credible recommenders constitute the computation of peers'reputation value. An update strategy is proposed based on the table structure to observe recommendation performance timely of peers who have record in the talbe. In experiments, this model was tested in different aspects and compared with calssical reputation model. Experimental results show that this model can effectively filter out the recommenders who give deceptive recommendations, reduce the miscarriage of justice for the honest recommenders when peers change their behavior and improve the succeful transaction rate of the P2P system.All in all, based on the research on reputation models for P2P networks, a metric for peers'reputation and a comprehensive trust decision method are proposed based on cloud model.A recommender selection method, dishonest recommender filtering method and a strategy to correct the honest peers that has been misjudged are also proposed in this thesis. The work in this thesis is not only important for the research on related fields of reputation models, but also has some guiding significance for improving the performance of existing reputation models by integrating the work into them.