查看文章

Peer-to-Peer (P2P) file sharing networks are widely used by Internet users but because of their anonymity and self-organisation nature, they are targeted by malicious users. Therefore, reputation systems are used as a way to distinguish between good and malicious peers. In this project we review some of the existing reputation systems and we then focus on DRank which is one of the latest systems. DRank uses the Bayesian model in reputation and trust calculations and its designers claim that it can fight pollution in P2P streaming robustly in a network of 200 nodes. In this project we examine it in a larger network. Moreover, we propose randomness to reputation sharing and peer selection mechanisms to improve the DRank results. For evaluation purpose, a simulator is built using a P2P simulator and DRank is examined with and without our proposed improvements. The experiments showed that when the network size increases by more than 200 nodes the performance of DRank decreases slightly and peers are more exposed to malicious peers. In addition, it is found that our proposal of random reputation sharing and the selection of best peers improves DRank performance.