Freenet: a distributed anonymous information storage and retrieval system
International workshop on Designing privacy enhancing technologies: design issues in anonymity and unobservability
Chord: A scalable peer-to-peer lookup service for internet applications
Proceedings of the 2001 conference on Applications, technologies, architectures, and protocols for computer communications
A scalable content-addressable network
Proceedings of the 2001 conference on Applications, technologies, architectures, and protocols for computer communications
Wide-area cooperative storage with CFS
SOSP '01 Proceedings of the eighteenth ACM symposium on Operating systems principles
Search and replication in unstructured peer-to-peer networks
ICS '02 Proceedings of the 16th international conference on Supercomputing
Pastry: Scalable, Decentralized Object Location, and Routing for Large-Scale Peer-to-Peer Systems
Middleware '01 Proceedings of the IFIP/ACM International Conference on Distributed Systems Platforms Heidelberg
Adaptive Replication in Peer-to-Peer Systems
ICDCS '04 Proceedings of the 24th International Conference on Distributed Computing Systems (ICDCS'04)
Proactive Hot Spot Avoidance for Web Server Dependability
SRDS '04 Proceedings of the 23rd IEEE International Symposium on Reliable Distributed Systems
Hi-index | 0.00 |
Highly skewed query distribution in structured Peer-to-Peer system may cause huge amount of dropped queries and consequently lead to poor system performance. This paper describes a Prediction-based Fair Replication Algorithm (PFR), which aims to maintain excellent system performance when the query is highly skewed. For the purpose of fairly distributing load onto each node, nodes that host hot items always shed load onto light-loaded nodes by creating replicas along the query path. Through the use of a simple prediction method, we can foresee traffic surge and replicate beforehand. Consequently, the number of dropped queries will decrease. Further more, each node can fairly decide the load redistribution speed for itself merely based on local information. The experimental evaluation demonstrates the effectiveness of our approach, which can simultaneously reduce the number of dropped queries as well as the number of created replicas without introducing unaffordable overhead.