Load Balancing in Hypercubic Distributed Hash Tables with Heterogeneous Processors

Abstract

AbstractThere has been a considerable amount of recent research on load balancing for distributed hash tables (DHTs), a fundamental tool in Peer-to-Peer networks. Previous work in this area makes the assumption of homogeneous processors, where each processor has the same power. Here, we study load balancing strategies for a class of DHTs, called hypercubic DHTs, with heterogenous processors. We assume that each processor has a size, representing its resource capabilities, and our objective is to balance the load density (load divided by size) over the processors in the system. Our main focus is the offline version of this load balancing problem, where all of the processor sizes are known in advance. This reduces to a natural question concerning the construction of binary trees. Our main result is an efficient algorithm for this problem. The algorithm is simple to describe, but proving that it does in fact solve our binary tree construction problem is not so simple. We also give upper and lower bounds on the competitive ratio of the online version of the problem.KeywordsLoad BalanceCompetitive RatioOnline AlgorithmAddress SpaceDistribute Hash TableThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.