An enhanced Bayesian-based reputation system for P2P file sharing

Peer-to-peer (P2P) file sharing networks are widely used by Internet users but due to their anonymity and self-organizational nature, they are often targeted by malicious users. Therefore, reputation systems are used to distinguish between good and malicious peers. Many reputation systems propose the use of the Bayesian model and it has proved to be effective for reputation calculation in a small, fully distributed system. This paper focuses on the distributed rank-based reputation system (DRank), one of the latest systems applying the Bayesian model in P2P streaming. DRank developers evaluated it on a small network with a maximum 200 nodes. In this paper, DRank is examined in a P2P file sharing system and on larger networks. This examination showed a drop in reputation propagation and a rise in the number of polluted files downloaded. Therefore, we propose a number of optimizations to DRank, such as unstructured reputation propagation, and two mechanisms of peer selection: (1) the first peer in the rank, and (2) randomly from a set of top ranked peers. For evaluation purposes, a simulator is built using a P2P simulator and DRank is examined with and without our proposed improvements. The experiments showed that our proposals improve DRank performance in preventing malicious peers from obtaining high reputation values.