An evaluation of chord using traces of peer-to-peer file sharing

Abstract

We evaluated the effectiveness of the Chord [1] as a protocol for file sharing using measurements from a real P2P application. Chord has several problems when used to support content sharing. The heavy-tail distributions we measured relating to queries and transfers results in a very unbalanced workload for Chord peers. Caches at nodes are helpful, but do not remove a unbalanced workload with a power-law skew. Our measurements are from a real P2P file-sharing protocol implementation based on a centralized architecture: users sent queries to a server that maintained a database of the current peers and their shared libraries. Users explicitly consented to our anonymized logging of their actions. Twice a day, a snapshot of users and their shared files was dumped Since we have a record of every query, and every transfer between users resulting from queries, the only information we are missing is the exact time when files in users’ libraries are added to the system out-of-band. This data revealed downloading habits of users that are independent of the underlying network file transferring architecture. We obtained a Chord simulator from the Chord project homepage [1].