Load Balancing in a Cluster-Based P2P System

In this paper we study load balancing in a cluster based p2p system. The cluster based architecture is chosen after the study of various real systems. We then define the notion of load for that system and describe coefficients that measure the different parameters of load. We use these coefficients to define load factors that can be used to measure each node's load. These factors are also used in two algorithms that balance the load fairly among all peers in a cluster. We improve the algorithms' balancing performance using replication of popular objects. A set of metrics is used to evaluate the cost of each algorithm and the effect on system's performance. Our simulation results show that it is possible to improve the fairness on load distribution without replication but with a small reduce on system's throughput. And under the cost of replicating a small amount of objects we improve fairness without reducing system's throughput.