A Self-organization Reputation Evaluation Algorithm For HP2P Systems

With the development and application of P2 P technology, the security of the P2 P protocol becomes extremely important. The insecurity of P2 P protocol hinders the application of P2 P technology in some extent. Especially in the pure P2 P network system, each node is a server of some other nodes, the security challenges facing bigger than that in the traditional network system. A reputation system for estimating the trustworthiness of nodes in P2 P network is very important. In practical applications of P2 P protocols, there are malicious attack problems. Malicious nodes upload malicious files, making the network contain a large number of malicious files. A reputation system dynamically computes the trustworthiness of nodes by the behaviors of them. According to the trustworthiness of nodes, a node can judge if another node is a malicious node or not, so that normal nodes can disconnect from malicious nodes. Also, when multiple nodes have a same resource, a node can choose to download the resource from nodes with high trustworthiness.In order to solve the malicious node attack problem in an actual operation of the HP2 P protocol, a self-organizing evaluation algorithm called HP2 PSORT which is based on SORT is proposed. In this thesis, the filtering mechanism, the calculation method of the service trust value and the dynamic calculation of the threshold are used to improve the performance of SORT. In computing the service trust value of another node, a node can obtain some recommendation information from some nodes with which have traded. By filtering the recommendation information which is too high or too low, the influence of the malicious recommendation information given by malicious nodes to computing the service trust values is reduced. The experimental results show that, due to the collusions of malicious nodes, the service trust values of some good nodes are low. In order to further decide whether a node is malicious or not, a new method for calculating the service trust value of a node is proposed. Further, a dynamic method for calculating the threshold is proposed, so that a node can computing different threshold in different environments to make a better judgment whether a node is malicious or not. In an HP2 P simulation network, HP2 PSORT outperforms SORT in all four kinds of combined attack models.