Self-optimizing DHTs Using Request Profiling

Various studies on request patterns in P2P networks have confirmed the existence of the interest-based clusters [11, 12]. Some P2P networks that exhibit the small-world phenomenon contain clusters of peers that frequently communicate with one another [17]. The existence of interest-based clusters opens up the possibility of more efficient routing. In this paper we consider the problem of designing a self-optimizing overlay network and routing mechanisms to permit efficient location of resources by the periodic profiling of request patterns. Our self-optimization protocol uses selective replication of resources for restricting the sizes of the clusters, and proposes the deployment of inactive nodes for further reduction of the routing latency. The self-optimization protocol is demonstrated on the Chord network [22]. It leads to a routing latency that scales with the size of the clusters.