A novel load balancing mechanism for P2P networking

Abstract

Peer to Peer (P2P) networking is a potential disruptive technology that can be used for the development of scalable, fully decentralized distributed applications. However, to realize its potential, P2P technology should address the needs of a variety of applications, other than file-sharing requiring support for exactmatch queries on the file names. Our work complements and contributes to existing P2P overlays that support multipleattributes and range queries, using the distributed K-Dimensional (K-D) tree structure for organizing shared information among participating peers. This guarantees that the time needed for node join - leave operations and query times are logarithmic with respect to the number of peers. In such systems, an open issue is load balancing of resources among peers, as only load-balanced data structures can guarantee that the complexity for resolving multi-attribute and range queries remains logarithmic (thus scalable) with respect to the number of participating peers. In this paper, we report a novel load balancing algorithm for dynamically keeping the resource load among peers balanced. We prove that the load balancing algorithm is robust and scalable, achieving an O(log2N) complexity, where N is the number of peers. We illustrate how our algorithm can be used to build a scalable Grid Information Service supporting multi-attribute and range queries on available services within the shared Grid infrastructure.