Asymptotic behavior and performance constraints of replication policies

Spreading the workload among a pool of replicated servers is a technique typically adopted to reduce the response time and increase the throughput in complex systems. In several actual computing environments, virtual machines can be provisioned in a fast and convenient way, and the replication has assumed an important role for the efficient system management. However, in order to provide economically acceptable solutions, the number of replica should be limited to the minimum required to match the given performance goal. In this paper we propose a simple replication policy to match thresholds on the system response times. The analytical relationships that exist between the performance objective values of some metrics and the number of replica are derived. Analytical and experimental validations with single and multi-class workload are presented. Open and closed models, and NO-SQL database have been considered.