Replication strategies in unstructured peer-to-peer networks
Proceedings of the 2002 conference on Applications, technologies, architectures, and protocols for computer communications
Super-peer-based routing and clustering strategies for RDF-based peer-to-peer networks
WWW '03 Proceedings of the 12th international conference on World Wide Web
Making gnutella-like P2P systems scalable
Proceedings of the 2003 conference on Applications, technologies, architectures, and protocols for computer communications
Evaluating GUESS and Non-Forwarding Peer-to-Peer Search
ICDCS '04 Proceedings of the 24th International Conference on Distributed Computing Systems (ICDCS'04)
Efficient search in unstructured peer-to-peer networks
Proceedings of the sixteenth annual ACM symposium on Parallelism in algorithms and architectures
PocketLens: Toward a personal recommender system
ACM Transactions on Information Systems (TOIS)
A Two-Level Semantic Caching Scheme for Super-Peer Networks
WCW '05 Proceedings of the 10th International Workshop on Web Content Caching and Distribution
Epidemic-Style management of semantic overlays for content-based searching
Euro-Par'05 Proceedings of the 11th international Euro-Par conference on Parallel Processing
Clustering in peer-to-peer file sharing workloads
IPTPS'04 Proceedings of the Third international conference on Peer-to-Peer Systems
Hi-index | 0.00 |
We study the effect of semantic overlay structure on the performance of decentralized search. Semantic overlays create communities of nodes that share particular interests. In peer-to-peer systems these communities can be designed to improve the recall of search algorithms. Such communities also play a role in balancing load between agents. An examination of these two performance metrics on some basic semantic overlay topologies shows that the choice of the best decentralized search algorithm can be influenced by differing design goals. We present an extensive experimental evaluation using data sets from eDonkey and Movielens. We find that, in general, these data sets do not exhibit obvious semantic clusters of nodes. For this reason, using a best-neighbors overlay, in which nodes individually choose their neighbors, to implement search produces better recall values than using an overlay that specifically clusters nodes into groups. Using best-neighbors overlays, on the other hand, can lead to highly unbalanced load distributions, a problem avoided in clustered overlays. We also find that forwarding search queries to "friends" in best-neighbors overlays does little to improve recall while further unbalancing load distributions.