Use of Shrinking Quorums to Improve Efficiency of Replication Protocols

Abstract

Quorums is a well-known concept extensively used in designing distributed algorithms requiring coordination among a number of nodes to ensure i) consistency of replicated data/perceived events/computed results etc. and ii) availability of the hosted service. Quorum is a subset of nodes present in a replicated server system satisfying some conditions. A lot of research has been done proposing various strategies to carefully engineer construction of quorums for efficient solutions. However, size of a quorum is dependent on the total number of nodes and is mostly defined as a constant parameter at system start. We propose a novel concept of variable quorum size, shrinking as the system encounters node failures while moving through different stages of its life. It is believed that such a construction can contribute to reduced message traffic and better performance results. We incorporate this concept in a total order protocol developed earlier and present the modified version as an example application.