A Binary-Tree based Hierarchical Load Balancing Algorithm in Structured Peer-to-Peer Systems

Abstract

In structured P2P overlay networks, nodes and objects are assigned unique IDs in the same identifier space by using a consistent hashing function, and have been known to result in an imbalance factor of O(log n), where n is the number of nodes in the systems. Further imbalance may become severely due to the heterogeneity of objects loads and nodes capacities, and dynamic of objects and nodes in practical P2P network environment. Aiming at the issue of load distribution unbalancing in large scale DHT network, a binary-tree hierarchical load balance model is presented in this paper. Our proposed load balance model organizes the system into a hierarchy of balancing domain, therefore, load balancing decision are based solely on information pertaining to those nodes within each domain, and thus load balancing is performed independently. In addition, because our load balancing scheme is built on the proximity-aware hypercube overlay network, therefore, the loads on the heavily nodes can be transferred to lightly loaded nodes in a proximity-aware fashion. Consequently, it provides rapid convergence on load balance and reduces the load transfer cost. Simulations show that our proposed scheme can balance the load of the nodes in proportion to their capacity, and reduced the load transfer cost in the heterogeneous and dynamic network state.