An adaptive mechanism for epidemic communication

Abstract

Reliable group communication is essential for building applications in distributed computing systems. Epidemic-style algorithms for group communication have attracted increasing interest. They emulate the spread of an infection; each computing node communicates with its randomly selected partners and information is disseminated by multiple rounds of such communication. Previous research has revealed that they are highly scalable and easy to deploy. In this paper we propose an adaptive mechanism with the aim of enhancing resiliency of these algorithms to perturbations, such as node failures. The mechanism dynamically adjusts the fanout, the number of receiver partners each node selects, to changes in the environment. Two metrics are used for this purpose, which reflect the status local to the node itself and the behavior of the whole system. This mechanism is analogous to those that can be seen in many biological systems, where each constituent part behaves mainly independently but is controlled indirectly by the whole system. © Springer-Verlag 2004.