When we got VMware’s approval to post a white paper with a replica of their own benchmarks that showed XenEnterprise to be about equal on Windows and way better on Linux performance, I told our performance team that I was sure that within a few weeks VMware would come back, swinging. After all, if your competitor offers a completely *free, production quality virtualization platform*that equals or beats the performance of your $several-thousand per server product, then you are likely to invest in a bit of mud slinging, right?
So VMware is back, as predicted, and while XenSource re-runs their benchmarks, and comes back with a response (for which we will have to obtain VMware’s permission to publish), I thought I’d offer a few remarks that don’t require sign-off from VMware.

First, while benchmark wars go on all the time, it is important to remember that performance is always a matter of what the customer thinks. Today XenSource was proud to announce KBC Clearingas a customer. They had not implemented virtualization before, but had researched other vendors before deciding to virtualize with XenSource. The company made their decision after testing both VMware ESX and XenEnterprise, and finding XenEnterprise’s performance to be superior. Since deployment in July 2006, KBC Clearing has reduced its data center to an eighth of its original physical size, and cut costs by not needing to manage and purchase additional physical servers. Indeed, Rob de Wit, IT specialist at KBC Clearing, had the following to say:

“XenSource outperforms VMware ESX. When we did a performance test on both Linux and Windows, XenSource came out best.”

It is well worth considering the ground we’ve covered since our product started to gain serious traction in the market. First, VMware entered into a poorly thought outbenchmarking exercise against their own build of Xen version 3.0.3 (as packaged by Red Hat’s in RHEL 5). We pointed out the difference between open source software and product, and when we showed that our product equals the performance of ESX in all but a couple of benchmarks, they changed tack, and are now going flat out trying to break XenEnterprise. Why are they paying us so much attention? First they tried to hammer Xen as soon as our partners the Linux Distros delivered Xen to market. And now that XenEnterprise is gaining traction in the market, they are trying to pick it apart. Clearly, we’re starting to hit them where it hurts – their pocket books. Customers are evaluating both products, and many, like KBC Clearing, are choosing XenEnterprise.

Performance aside, I’m delighted that customers agree that XenSource has achieved another goal – taking the mystery out of virtualization. Or, to quote an excellent review of XenEnterprise: “As their “10 minutes to Xen” tagline suggests, easing the complexity of virtualization and management are clear goals with these packages, and they really do hit a home run in that regard.”

It’s also time to thank VMware for their contributon to the Xen community: Their benchmarking work keeps us on our toes, and needles our team to continually improve our product. For a lesson on open approaches to benchmarking, look no further than the recent Xen Summit, where there was a whole session dedicated to performance evaluation. As a result, the core team will quickly find new ways to optimize Xen and make it better. The attention VMware pays to Xen gives us a great platform from which to point out the unique advantages of the Xen community based development approach and highlight our architectural and performance advantages. And of course, all of the tuning that XenSource does goes right back to the community. VMware is arming their arch rival.