Recent reports on potential scalability problems in Gnutella have focused attention on performance issues in decentralized filesharing networks such as Gnutella and Freenet. I propose an analytical framework for evaluating performance in these systems, based on the small-world model originally popularized by Stanley Milgram in the context of social networks and more recently applied to technological networks such as power grids and the World-Wide Web by Duncan Watts and others. Using simulation, I examine some aspects of scalability and fault-tolerance in Freenet and Gnutella, and attempt to explain contrasting behaviors between the two in terms of the small-world model. Finally, I hope to draw some conclusions for designers of future filesharing systems.

(This talk is based on the chapter on performance I am contributing to the forthcoming book, "Peer-to-Peer: The Disruptive Potential of Collaborative Networking" -- available on request.)