Research On Reputation Management Of P2P File Sharing System

In recent years, peer to peer network is widely used in file sharing, multimedia transmission, instant communication, collaborative computing and distributed data access and so on. Because as a new kind of network application mode, it has improved the efficiency of the network, fully utilized the network bandwidth, developed each peer's computation and storage capacity. What's more, it is good in scalability and fault-tolerance.When the P2P network is universally used, peers'selfish behaviors and malicious behaviors are also more and more serious, which has greatly influenced the usability of the network. The reason for this situation is lack of effective reputation mechanism between peers. In the P2P networking, traditional reputation models are divided into the local reputation and the global reputation. The local reputation model's rating is one-sided and reputation is difficult to collect. But in global reputation model, computation and communication will cost too much and it is easily cheated and attacked by associated malicious peers. In addition, this model will ignore the private feature of trust. The necessity and feasibility of calculating global trust value for each node in large-scale P2P networks are still doubtful.In view of the shortage of current P2P reputation model, this paper proposes a multi-level reputation model based on trust and decision-making process of individual transactions in sociology. Compared with the previous models, this model details the trading decision-making process. In different trade scenes different levels of reputation are established to compute trust value gradually, which makes this model more reasonable and cost less. In this multi-level trust model, trust between peers is divided into three levels: the local direct trust, the most trusted set trust and the interest group trust. The introduction of the most trusted set and interest group will link scattered peers together so as to improve the efficiency of the management of reputation. This thesis firstly gives the definitions, sociological supports and calculation formulas of the local direct trust, the most trusted set trust and the interest group trust. Then this thesis puts forward a group of protocols and algorithms to manage the most trusted set and the interest group. Then this thesis will describe the reputation decision-making algorithm before trading. Finally, to overcome the shortcomings of traditional broadcasting algorithm, this thesis will propose an optimized broadcasting algorithm to improve efficiency of collecting reputation reduce the network overhead.Designed and implemented the simulation system, simulated malicious attacks to the multi-level reputation model. The experimental results show that the multi-level reputation model can still guarantee the successful rate of trading above 80% in the network environment with large amount of and various kinds of malicious peers. Thorough statistics, it is found that multi-level reputation model's additional network overhead is far less than that of traditional reputation models.