Recently our friends at VMware .
The paper shows Xen 3.0.3 performing poorly in comparison with ESX Server 3.0.1 on a set of standard OS benchmarks run in a Windows Server 2003 R2 VM, including Netperf, Passmark, SPECCPU and SPECJBB. In almost every test the VMware build of Xen performed worse than ESX – indeed so badly that VMware concluded that Xen 3.0.3 is in no way ready for enterprise use. VMware also concludes that Intel VT, which Xen uses for hardware assisted virtualization of Windows, is a major contributor to Xen’s apparently shoddy performance.
I’m afraid I have to admit I find this more than a little embarrassing.
Embarrassing, that is, that a company with the technological and marketing prowess of VMware did not think a little harder before publishing something so transparently foolish. VMware is tilting at windmills once again. For a start, Xen 3.0.3 is not a commercial product, it’s a code base. Second, Xen 3.0.3 only had partial support for hardware virtualization, and third, the VMware results are off – it seems ESX can get more than 1Gb/s out of a GigE NIC! Also, we have no idea what VMware did to build their Xen bits – a good example of why we care so much about what can and and cannot be called Xen™.
We have of course run exactly the same benchmarks, on the same kind of machine, pitting our commercial XenEnterprise product against ESX. As a result, I now understand why the ESX EULA forbids us from publishing the results without VMware’s approval – they would find the results extremely embarrassing.
All of XenSource’s commercial products match or beat ESX performance for Windows in all but a couple of benchmarks. This for a new product, and using the HVM feature set that has never been tuned. For Linux, we absolutely thrash ESX, which should come as no surprise. We’ll publish all of our results… just as soon as we get permission from VMware, that is.
How can I convince you? It’s simple: Download XenExpress and see for yourself in under 10 minutes what free, high performance virtualization for Windows & Linux is all about. And while we wait for the ‘researchers’ at VMware to approve our benchmarking methodology*here’s a paper* with our results, but with ESX performance numbers deleted.
The bottom line? Embarrassing.