Abstract
Virtualization can provide many benefits for managing resources, including higher resource utilization, lower energy cost, faster fault recovery and more flexible re- source provisioning, etc. Hence, we have seen an increasing trend for adapting virtual machine in both datacenters and in-house clusters. However, virtualization also brings several new challenges, especially for scientific applications which have more complex runtime behavior and higher performance demand. In this work, we use real scientific applications and performance benchmarking tools to analyze the application performance of our in-house virtualized cluster. We found the perfor- mance degradation could be minimized with proper virtual machine configuration and the support of hardware virtualized InfiniBand, but the performance is still difficult to be modeled or predicted. Therefore, we developed an auto-tuning tool for finding the best resource provisioning in terms of both time and cost, and show that we can find close to optimal resource provisioning setting in much shorter time than searching though all possible settings.