Supernode Election Algorithm in DHT Based upon Network Coordinate and District Partition

Abstract

Most P2P applications require to elect some supernodes(SN) from the peers to improve the performance, so the supernode election has become an important issue in such systems. To our best knowledge, most applications assume that nodes with enough bandwidth and powerful computation, having a public IP address can become SNs, but this strategy may have some disadvantages e.g. SNs may be clustered into some regions, and the access latency of ordinary nodes to SNs may be very large, the ratio of ordinary nodes to SNs may not be maintained, and some SNs may be overloaded. Though some applications use several SN administration strategies to improve the performance, but the improvement is limited especially in a large scale P2P network. In this paper, we bring forward a novel election algorithm based on network coordinate and district partition for DHT architecture to solve these problems. Through analysis and evaluation our algorithm could effectively accomplish SN election in a short time (less than 4s), scale to a large network, adapt to dynamic change, reduce the overhead to which most peers can afford, maintain the ratio of SNs to non-SNs, remarkably reduce the latency of non-SNs to SNs, especially the bandwidth consumption (less than 50kbps) is small no matter how much the node quantity is