Maximizing Availability of Consistent Data in Unreliable Networks

We address the issue of maximization of the availability of replicated data that are distributed in a wide area network. We consider a system that uses majority voting, which is a common mechanism for providing consistency of replicated data in the presence of failures. The data availability provided by this mechanism critically depends on the vote assignment to the replicas. In this paper we formulate the problem of finding the optimal vote assignment into a specific form of a combinatorial optimization problem, namely the MAX-SMT problem. This formulation allows us to use a modern, fast MAX-SMT solver to solve the vote assignment problem. To evaluate the effectiveness of this approach, we build a failure-repair model of underlying networks and estimate the data availability using that model. The results of the estimation show that data availability can be significantly improved using the optimal vote assignment in the presence of failures.