THWARTING P2P POLLUTION

Much of the content in peer-to-peer filesharing networks is corrupt, damaged, or mislabeled. Such polluted content makes it difficult for correctly functioning peers to locate desired content, decreasing the overall usefulness of the network.

Credence is a practical distributed reputation system, designed to counteract content pollution in peer-to-peer filesharing systems. Credence enables a peer to determine the authenticity of online content; that is, how accurate the purported description of the object matches the object itself. Participants in the Credence network vote on objects; Credence collates these votes, and weights them by a novel similarity measure which weighs highly votes from like-minded peers while discounting the votes from peers engaged in vote-spamming. This novel voter correlation scheme provides an incentive for peers to vote honestly and mitigates the impact of dishonest peers.

We have implemented Credence on top of the LimeWire client for the Gnutella network. Our client provides a peer-based judgement that a given object will possess the properties with which it is labeled and enables users to evaluate search results for authenticity before downloading.

Many peer-to-peer reputation schemes have been proposed in academia. Credence is the first practical implementation of a peer-to-peer reputation scheme. As future systems become more connected and endpoints become more reliant on content provided by other nodes, reputation systems like Credence can play an important role in assessing the validity of remote content.

As of September 20, 2005, Credence has been downloaded more than 10,000 times. We are examining the emerging properties of the Credence network.