Experience with an object reputation system for peer-to-peer filesharing

In this paper, we describe Credence, a decentralized object reputation and ranking system for large-scale peer-to-peer filesharing networks. Credence counteracts pollution in these networks by allowing honest peers to assess the authenticity of online content through secure tabulation and management of endorsements from other peers. Our system enables peers to learn relationships even in the absence of direct observations or interactions through a novel, flow-based trust computation to discover trustworthy peers. We have deployed Credence as an overlay on top of the Gnutella filesharing network, with more than 10,000 downloads of our client software to date. We describe the system design, our experience with its deployment, and results from a long-term study of the trust network built by users. Data from the live deployment shows that Credence's flow-based trust computation enables users to avoid undesirable content. Honest Credence clients can identify three quarters of the decoys encountered when querying the Gnutella network.