Effective Load Balancing in P2P Systems

In DHT based P2P systems, various issues such as peer heterogeneity, network topology, and diverse file popularity, may affect the DHT system efficiency. In this paper, we propose an effective load balancing algorithm for DHT-Based P2P systems. Our main contributions are: (1) we propose an fully distributed mechanism to maintain the history of file access information. This information is used to predict the future file access frequencies and support the load distribution and redistribution operations; (2) we design a novel load balancing algorithm, which takes the file access history and peer heterogeneity properties into account to determine the load distribution. Our algorithm can generate the best load distribution decision when a new peer comes, it can also be able to dynamically perform the load redistribution during system running time if overloaded peers appeared. In our algorithm, no virtual servers are used, thus we have less processing overhead on the expensive routing metadata maintenance; (3) finally, we design a topologically-aware data replication mechanism, the topological information of the peers are used for file replication decisions. A file is replicated only on a peer close to the group of peers which have high access frequencies.