Thursday afternoon, Microsoft released its virtualization product, Hyper-V, to manufacturing. Previously, the company had promised to make a production-supported version of Hyper-V available to Windows Server 2008 customers within 180 days of the official release of the operating system itself.
By releasing Hyper-V in late June, Microsoft beat its self-imposed deadline by about a month, although delivering less than was originally promised. There have been public release candidates available since early 2008. In this piece, I'll take a look specifically at the release-to-manufacturing, or RTM, edition, noting improvements and changes.
The most notable -- and the most significant -- change between the initial release candidate version (RC0) of Hyper-V and the RTM edition is better performance. Most of the performance work was done between RC0 and RC1, but not many people knew about it due to (a) not-so-wide a release and (b) a ban on performance testing by MS. The company just wasn't ready for it to be tested on a wide scale. The performance story between RC1 and RTM is identical.
QLogic, a vendor of storage area networks (SANs) and other components, tested Hyper-V RC1 on one of its own host bus adapters (HBAs), based on Fibre Channel. QLogic hooked the HBA up to a storage array, to test the number of input/output operations per second supported by a virtual machine as opposed to real, physical hardware. The results were impressive, and are equally so under RTM as the underlying performance plumbing didn't change.
In particular, a setup based on intensive I/O applications -- one that is common to mail server and database server configurations -- showed that 120,426 I/O operations per second were possible with real hardware, compared with 116,720 per second on a Hyper-V-based virtual machine. In other words, the virtual machine was able to saturate the hardware enough to achieve 97 percent of the storage performance of a physical server.
Similar tests were performed by QLogic in a variety of configurations, each mimicking a specific type of storage operation, and results were similar -- as low as 88 percent of real hardware in one scenario and as high as 99.93 percent in other. (These results are here.)
Indeed, performance has been good enough that Microsoft claims it has been running its popular msdn.microsoft.com and technet.microsoft.com sites from Hyper-V RC1 virtual machines now for months. These sites combined receive around 4 million hits per day. Each IIS7 virtual machine runs four virtual CPUs with 10GB of RAM, and the physical hosts have two quad-core CPUs with 32GB of RAM and host three virtual machines. Microsoft will be migrating to the RTM version of Hyper-V on these production machines very soon, as part of its major push to virtualize up to 25 percent of its internal IT infrastructure this year.
There were no major performance tweaks between RC1 and the RTM version, so the raw performance numbers are 100 percent intact within the RTM version.