February 26, 2013 - by rachelbalik
Do a quick Google search, and you’ll find enough literature on Xen performance to keep you reading until they’ve invented a computer that will do your laundry...on Mars. Ok, maybe not quite that much material, but certainly enough to give the impression that most of its performance issues have been resolved. What you won’t find are many detailed comparisons of Xen to other virtualization technologies, especially OS virtualization. And, for lack of evidence to the contrary, this might lead you to think that Xen is the best of what’s out there.
But you don’t need to remain in the dark for long. Brendan Gregg, Lead Performance Engineer at Joyent, has done an extensive analysis comparing the performance of Zones, KVM and Xen. In a post on the DTrace blog, he summarizes their performance in four ways: Characteristics, block diagrams, internals and results. Through looking at the I/O path (network, disk) and its overhead, he provides some minimum config results comparing performance between OS virtualization (Zones) and the hardware virtualization of Xen and KVM varieties.
Brendan’s full post offers deeper insight into how each of these technologies work, how each criterion for performance is evaluated and the code path for performing network I/O with each technology.
Read the full analysis on the DTrace blog. We don’t want to give away the ending, but if you were previously convinced that Xen performance was the best you could do, you might be in for a surprise.