An empirical evaluation and improvement of the item balancing algorithm in P2P systems

Load balancing is one of the main issues in structured Peer-to-Pear (P2P) systems. An improvement in the load balancing algorithm can significantly enhance the resources utilization and fairness in P2P systems. Load balancing across multiple nodes has been widely studied in P2P systems. Some proposals exist for load balancing in P2P systems in which each node maintains log n virtual servers and allows the address space to be partitioned equally but the degree of each node is log²n which is not desirable. Karger & Ruhl proposed an algorithm that simultaneously offers a node degree of log n and an equal partition of the address space. We propose an improvement to Karger and Ruhl algorithm by adopting greedy randomize approach which improves the variance of number of items across all peers significantly using additional peers´ load information. We simulate the Karger and Ruhl item balancing algorithm, evaluate its load balancing results then simulate the improved item balancing algorithm and evaluate its load balancing results. The results of the improved algorithm are compared with the results of the original algorithm. The results of the performance comparison show better load balance using the improved algorithm; around 40% in the variance by using only 1% of the number of peers in the system.