A Malicious Peers Detection Framework for Peer-to-Peer Systems

A number of reputation mechanisms are introduced in recent years to alleviate the blindness during peer selection in distributed P2P environment where malicious peers coexist with honest ones. They indeed provide incentives for peers to contribute more resources to the system, and thus, promote the whole system performance. However, little attention has been paid on how to identify the malicious peers in this situation. In this paper, a general framework is presented for detecting malicious peers in Reputation-based P2P systems. Firstly, the malicious peers are divided into various categories and the problem is formulated. Secondly, the general framework is put forward which mainly contains four steps, i.e. data collection, data processing, malicious peers detection and malicious peers clustering. Thirdly, an algorithm implementation of this general framework named IDEA is put forward based on Principal Direction Divisive Partitioning and K-means clustering. Finally, a series of simulation experiments are conducted to evaluate the effectiveness of IDEA. Simulation results have shown that IDEA can precisely distinguish and cluster the malicious peers. Subject Categories and Descriptors C.2.1 [Network Architecture and Design]: Distributed networks; I.5.3 Clustering]: Algorithms General Terms: Distributed networks, P2P, Clustering