Runtime performance is an aspect of virtual machine qualification that often goes overlooked. Many IT shops wrongly assume that virtualization performance is tied to the underlying hardware, discounting the impact of the controlling virtual machine monitor or hypervisor. In reality, the efficiency of the virtualization layer can literally make or break the scalability of a virtual machine-based deployment. And as the test results show, it’s often the lightest-weight solution that delivers the best raw VM performance.
To assess the performance and scalability of these four desktop virtualization participants from Microsoft, VMware, Parallels, and InnoTek, I constructed a simple client/server workload simulation spanning two concurrently executing virtual machines. Using the free DMS Clarity Studio tool from Devil Mountain Software, I constructed a pair of multiprocess test scenarios featuring client/server database (ADO/SQL Server), workflow (MAPI), and multimedia (Streaming Windows Media) tasks. One VM hosted Windows Server 2003, IIS 6.0 and SQL Server 2000. The other hosted Windows XP SP2 and ran the DMS Clarity Studio Framework shell to drive the client/server transactions against the first VM.
The results were quite revealing (see charts). The lone hypervisor-based solution, Parallels Workstation for Windows 2.2, was the also the overall performance leader, outpacing the more sophisticated — but also more bloated — VMware 6.0 and even challenging the always-speedy Microsoft Virtual Server 2005 R2 (used as a baseline reference point). Microsoft Virtual PC 2007 was a distant third, while VirtualBox 1.3 — which eschews hardware-based virtualization in favor of its own software-based implementation — came in dead last and well off the pace.