Load balancing in peer-to-peer systems using a diffusive approach (Q455930)

A load balancing algorithm for peer-to-peer system is introduced. The objective is to equalize the available capacities of the system nodes such that they will have similar mean response times. The basic technique is the diffusive load balancing used in parallel computing system that is adapted for peer-to-peer systems. Each node collects the loads of its neighbors in the overlay network and decides the load transfers, without the need of extra connections between nodes or messages on random walkers in the system. A particular attention is given to the effect of the size of the services. The proposed algorithm converges faster for small-sized services than in the case of large-size services. Moreover it is proved that the impact of the churn rate on the bound of the standard deviation of the available capacities is larger than the impact of the sizes of services.