From the “things companies do in their End-User License Agreements to make their competitors’ lives more interesting” department – because we know you haven’t read enough about that angle in the last week or two…
As our CTO, Simon Crosby, noted in his blog, VMware recently published a performance paper that compared ESX Server 3.0.1 and open source Xen 3.0.3. His comments explain why we find this an absurd comparison — a lot like taking an engine out on the highway and complaining that it can’t outperform an entire car.
But I’m also interested in what it takes for us to get our side of the story – that XenSource XenEnterprise 3.1 is a more appropriate comparison and can hold its head up high – out to you.
VMware’s EULAprohibits the distribution or publication of test results for their product without their prior review and approval – a process they acknowledge in their process documentation can take a month or so.
I understand their original motivation for this – I was the product manager for ESX Server 1.x, back when server virtualization was new and scary, and some yahoos kept running micro-benchmarks that attacked the product for sub-sub-statistics like the number of cycles spent in the network stack that grossly exaggerated performance differences on real user workloads.
But it’s just not the same, I think, when you’re the incumbent product in a viable marketplace, you post your own tests, and make the other side wait for approval… I dunno. It seems to me like applying for a job and having to send your resume to one of the other applicants first for their approval… or wanting to ask someone out on a date and having to ask their ex for permission…
Anyway, here for your amusement, while we wait for approval of the real thing, is a not-quite-comparison of commercial hypervisors. You should still find the comparisons to native performance informative. And, of course, benchmarks like this are only a signpost — the best benchmark is your own applications’ performance.
EDIT: The final paper can be found here.