Self-stabilizing local k-placement of replicas with local minimum variance

Abstract

Large scale distributed systems require replication of resources to amplify availability and to provide fault tolerance. The placement of replicated resources significantly impacts performance. This paper considers local k-placements: Each node of a network has to place k replicas of a resource among its direct neighbors. The load of a node in a given local k-placement is the number of replicas it stores. The local k-placement problem is to achieve a preferably homogeneous distribution of the loads. We present a novel self-stabilizing, distributed, asynchronous, scalable algorithm for the k-placement problem such that the standard deviation of the distribution of the loads assumes a local minimum.