A distributed hierarchical clustering algorithm for large-scale dynamic networks

Abstract

We propose an algorithm that builds a hierarchical clustering in a network, in the presence of topological changes such as those that occur in wireless mobile networks. Clusters are built and maintained by random walks, that collect and dispatch information to ensure the consistency of clusters. We then implement distributed communication primitives allowing clusters to emulate distributed algorithms; those primitives ensure that messages sent by a cluster are received and treated only once by their recipient, even in the presence of topological changes. Decisions concerning the behavior of the cluster are taken by the node that owns the random walk at this time. Based on this abstraction layer and the overlay network it defines, we present a distributed hierarchical clustering algorithm, aimed at clustering large-scale dynamic networks.