On the Availability of Replicated Data Managed by Hierarchical Voting

Abstract

Cloud services often depend on a mechanism that provides a reliable data storage that is resilient to failures. Majority voting and hierarchical voting, an extension of majority voting, are both used to implement such a mechanism in an unreliable network. For example, Apache ZooKeeper adopts both of the two methods. This work considers the question whether or not hierarchical voting with two levels indeed improves data availability compared to majority voting. The simulation results obtained so far suggest that hierarchical voting is unlikely to increase data availability.