Improved load balancing algorithms in DHT-based dynamic P2P systems

Load balancing for DHT (Distributed Hash Table)-based peer-to-peer (P2P) systems has been studied extensively. The DHT-scheme simply assigns each object to a specific peer with a hash function. This could result in O(log N) load imbalance, where N is the number of peers in a system. It is natural that a P2P system experiences frequent join and leave of both objects and peers. This dynamic characteristic may intensify load imbalance and, therefore, could increase the overhead for load balancing. In this paper, we propose two load balancing schemes, the object lifetime-aware load balancing (OLLB) and the peer activity-aware load balancing (PALB) schemes. The OLLB scheme uses the predicted lifetimes of objects to deal with the trade-off between the load balancing and the load movement. And the PALB scheme manages intelligently the virtual server operations to reduce the load unbalance occurred by dynamic peer activities. The simulation results show that the proposed schemes improve the load balancing performance with less overhead in the dynamic environment.