On Jul 26, 2010, at 12:03 PM, Gilberto Nunes wrote: [...] > What you thing about???
As far as running 15 VMs, whether your hardware is suited to do that depends on how many spindles worth of SAS drives you have (improves concurrency), how busy your VMs are (IO and proc), how much the guests are swapping in case you're not giving them enough memory. And if you're not running virtio drivers, you should! I don't have the numbers, but several months ago I tested a few different Iometer meter workloads on Server 2003 R2 guests on PE1950 hardware against equivalently-matched VMs on the 1.x version of a popular proprietary product (dedicated memory instead of its default swap-mem-to-host-disk; also running the guest extensions), Xen on CentOS 5.4 with a then-recent build of GPLPV (meadowcourt.org/downloads) on the guests, and KVM (also CentOS 5.4) with somebody's build of unsigned virtio Windows drivers (was on a /~public_html from redhat.com I think). Results: Xen+GPLPV beat out KVM+virtio enough to be considered significant, but their difference seemed small compared to the margin they beat the other contender by. The proprietary one also had massive CPU load on the guest generated by running the test that the others didn't have. Obviously that's all very vague, but then again I'm sure somewhere I've accepted a EULA that says I'm not allowed to share benchmarking results for certain products :-). I'll be keeping Xen (and therefore CentOS 5.x) around to run Linux guests blazingly fast on still-usedful hardware. Everything else I'm (slowly) migrating to KVM in the interest of tracking with upstream. Xen's slight performance edge on Windows will be missed. YMMV. Eric _______________________________________________ CentOS-virt mailing list CentOS-virt@centos.org http://lists.centos.org/mailman/listinfo/centos-virt
