Virtualization has been part of the datacenter since the 1960s when it was exploited across mainframe systems [1]. Virtualization is experiencing a renaissance as this technology is finding its way to high-volume servers. To the IT technologist, virtualization brings the promise of solving several datacenter problems. Virtualization can reduce costs by consolidating older servers. It helps organizations become more nimble through fast provisioning of virtual servers. It improves equipment utilization and the end-user experience by enabling dynamic load balancing and improved disaster recovery capabilities. These benefits provide a strong motivation for accelerating server virtualization deployments.

Virtualization provides an abstraction of hardware resources enabling multiple instantiations of operating systems (OSs) to run simultaneously on a single physical platform. The abstraction provides isolation between the separate running partitions to prevent individual faults from affecting the entire system. The virtualization of the hardware also means that different OSs can be supported on a single platform simultaneously—even older OSs. Consolidating several physical servers that have workloads with non-overlapping peak utilization requirements over time allows better hardware utilization than if these were carried out on separate systems. These benefits are attractive in environments with legacy servers that, though important to the business, cannot justify the porting and validation activity to a newer OS [2].

In this paper we present a methodology and an example for characterizing the performance of servers using virtualization to consolidate multiple physical servers. We provide a general overview of two key virtualization usage models. We also briefly look at how contemporary methods can be applied to virtualization. We discuss the challenges generated by the virtualization abstraction layer and consolidation, and we present a systematic approach to performance measurement. Finally. an example workload, called vConsolidate, is defined to further clarify the methodology.