VALIDATING AN ANALYTICAL APPROXIMATION THROUGH DISCRETE SIMULATION
J.-F. Paris · 1997
While Markov models have been extensively used to study the availability of replicated data, they cannot handle effectively network configurations where sites failures and network partitions have to be simultaneously considered. We had proposed in a previous paper a hierarchical decomposition method aimed at overcoming this limitation. While our method could provide closed form estimates of the availability of replicated objects whose replicas reside on networks subject to communication failures. We present here a simulation study measuring the quality of our estimates and attempting to improve upon them. KEYWORDS fault-tolerance, replicated systems, redundancy, voting. INTRODUCTION Managing replicated data can be a demanding task particularly when the replicas are stored at different sites of a computer network. Special replication control protocols have been devised to perform this task without user intervention and maintain the replicated data in a consistent state. Evaluating ...