An efficient atomic multicast protocol for client-server models

Abstract

Distributed applications which are replicated to achieve fault tolerance usually are faced with the problem of maintaining consistent states. To eliminate potential divergence of replica states caused by unreliable communication primitives, an atomic multicast protocol is required which guarantees delivery of messages at all receivers in the same order. The authors propose a new efficient protocol in which the order of messages is established by a token passed from one receiver to the other on a logical token ring. The protocol works correctly in presence of up to k failures. An analysis of the protocol characteristics and comparison to previously published atomic multicast algorithms shows that the approach meets the requirements in modern client-server models best.<>