Cluster-based distributed consensus

Abstract

In this paper, we incorporate clustering techniques into distributed consensus algorithms for faster convergence and better energy efficiency. Together with a simple distributed clustering algorithm, we design cluster-based distributed consensus algorithms in forms of both fixed linear iteration and randomized gossip. The time complexity of the proposed algorithms is presented in terms of metrics of the original and induced graphs, through which the advantage of clustering is revealed. Our cluster-based algorithms are also shown to achieve an Omega(log n) gain in message complexity over the standard ones.