Cloud environments offer the flexibility of virtual capacity and pay-as-you-go billing features, but cost-aware services must match computation capacity to workloads. Because cloud platforms provide virtual machines that run on top of typically heterogeneous hardware shared by multiple clients, their performance variability — that is, nonuniform VM performance — is inherently higher than it is on private hosting platforms. Consequently, capacity provisioning in a cloud must consider workload variability as well as varying VM performance.
In a paper presented at the IEEE fifth International Conference on Cloud Computing (Cloud 2012), researchers from the IBM Research Lab in Zurich and the University of Lugano propose an opportunistic service-replication policy that leverages both VM performance variability and the cloud's on-demand billing features. Evaluation results on traces collected from in-production systems show significant savings in costs and response times for the approach.
"Opportunistic Service Provisioning in the Cloud" and other papers from Cloud 2012 are available to IEEE Computer Society members and paid subscribers via the Computer Society Digital Library.