|
Abstract : |
While attention has been paid in the computer science literature to improving efficiency of search algorithms in P2P networks, little attention has been paid to 1) economic incentives that guide users to share content, 2) content-based measures of similarity of interests among users, and 3) cost implications of physical location of peers on the underlying network. Our work draws on ideas from the economics literature of club goods and the information retrieval literature to propose next generation P2P file sharing architectures that rely on content-based, self organizing communities of peers to address these issues. Using the model of ultrapeers and leaf nodes in Gnutella v0.6 architecture as context, we conceptualize an ultrapeer and its network of leaf nodes as a club (in Economic terms). We specify a simple utility-based model for a peer to determine which ?clubs ? to join, for clubs to manage their membership, and for ?clubs ? to determine to which other clubs they should be connected. We simulate performance of these models using unique real world dataset collected from the Gnutella v0.6 network. Our preliminary simulations demonstrate that our enhancements result in at least 300 % improvement in ultrapeers ? ability to satisfy their leaf node?s queries, thereby significantly decreasing congestion on the network and enabling more efficient and effective file sharing architectures., |
