Using Diffusive Load Balancing to Improve Performance of Peer-to-Peer Systems for Hosting Services

Abstract : This paper presents a diffusive load balancing algorithm for peer-to-peer systems. The algorithm reduces the differences of the available capacities of the nodes in the system using service migrations between nodes in order to obtain similar performance for all nodes. We propose algorithms for handling homogeneous services, i.e., services with equal resource requirements, and for heterogeneous services, i.e., services with diverse resource requirements. We have investigated the effect of load balancing in a simulated peer-to-peer system with a skip-list overlay network. Our simulation results indicate that in case that the churn (nodes joining or leaving) is negligible, a system that hosts services with small resource requirements can maintain equal performance for all nodes with a small variance. In case that churn is high, a system that hosts homogeneous services with large resource requirements can maintain equal node performance within a reasonable variance requiring only few service migrations.