Virtual High Performance Computing Environments for Science Computing On-Demand

Science computing environment dynamic construction based on virtualization technology can be benefit in performance isolation, on-demand creation and resource controls. This paper has focused on approach to science computing application environment provisioning based on virtualization technology. We propose a system model based on the features of science computing application and the key problem of provisioning is analyzed. Based on those works, a host capability rank strategy that can select the better performance host to deploy the virtual appliance is proposed. As the experiment shown, the algorithm can decrease the deployment time. As an additional result, we have observed that application runtime performance does have a linear increase on different application cases.