SIMULATING THE HUMAN FACTOR IN REPUTATION MANAGEMENT SYSTEMS FOR P2P NETWORKS An Agent Based Model

A compelling problem in peer-to-peer (P2P) networks for file sharing, is the spreading of inauthentic files. To counter this, reputation management systems (RMS) have been introduced. These systems dynamically assign to the users a reputation value, considered in the decision to download files from them. RMS are proven, via simulation, to make P2P networks safe from attacks by malicious peers spreading inauthentic files. But, in large networks of millions of users, non-malicious users get a benefit from sharing inauthentic files due to the credit system. In this paper we show, using agent based simulation, that reputation systems are effective only if there is a widespread cooperation by users in verifying authenticity of files during the download phase, while the size punishment derived by the reputation systems is less relevant. This was not evident in previous works since they make several ideal assumptions about the behaviour of peers who have to verify files to discover inauthentic ones. Agent based simulation allows to study the human factor behind the behaviour of peers, in particular the advantage of spreading inauthentic files, of not checking as soon as possible their authenticity during the download, thus unwillingly cooperating to the spreading of files-point.