Decision Support for Virtual Machine Re-Provisioning in Production Environments

Management of virtual machines (VMs) on a large scale remains a significant challenge today. We lack general and vendor-independent quantitative criteria/metrics by which to describe the state of infrastructures for virtualization. This data is essential to both expressing administrative policy goals and to measure ongoing compliance with them in production settings.

In this work, we consider VM management in production environments. We investigate VM applicability to real-world scientific computing problems by comparing application performance in a controlled environment of physical servers with the Manage Large Networks (MLN) tool's implementation of the same scenario on virtual machines. Based on our observations, we propose three new metrics by which to describe and analyze the infrastructure in order to incorporate virtual machine management closer into policy. The metrics have been implemented into the latest release of our MLN tool for virtual machine management.