I really have no practical use for this, but it sure was fun to set up and get to work

I spent some time in #fedora on Freenode tonight troubleshooting an issue I had in regards to virt-manager not being able to connect to KVM/Qemu. The libvirtd service was running, but virt-manager simply couldn’t connect to the hypervisor.

Apparently, that’s because it wasn’t there. This issue probably wouldn’t affect you if you installed from the “Beta” milestone release, but I installed from a newer snapshot, which means it looks like I shook up a bug. In Fedora 11, the QEMU and KVM RPM’s are being merged, and as a result, the yum group “Virtualization” was still trying to pull in the ‘qemu-kvm’ RPM when it should have been looking for simply “qemu”. Once I figured that out, and did a yum install qemu all was right with the world once again

Here is some further reading on the topic if you would like to do your own research.

https://bugzilla.fedora.us/wiki/Features/KVM_and_QEMU_merge

Now, now, don’t get excited, this isn’t a great, fantastic whiz-bang benchmark article. If you want that head over here to Phoronix where they did a great test comparing Ubuntu 8.04 running natively vs. Ubuntu 8.04 running virtualized on an Ubuntu 8.04 host running KVM. It’s a great read, and if you love all things virtual, as I do, be sure to check it out.

Now, on to my little “unscientific” test. I didn’t want to see how Ubuntu performed virtualized on Ubuntu - I will likely not be running that scenario quite frequently, other than testing Alpha and Beta builds of Ubuntu +1.

I will mostly be running Windows guest virtulized on KVM, so I wanted to get some indication of raw CPU bound performance, and with my quick ‘n’ dirty little test, I was quite pleased.

This test isn’t exactly apples to apples, but I think it’s as close as I could get, given the hardware at hand. The guest machine is a Windows XP x64 guest virtualized on KVM, with 1024MB of RAM allocated too it, and 2 Virtual CPU’s. The machine it’s being compared too is my old Core2Duo machine, which is downstairs running Windows XP x64 natively. This should actually be a pretty close match-up. The “native” machine is running on a Core2Duo E4600 2.4GHz dual-core CPU with 2MB L2 Cache. The “host” for the virtual is running a Core2Quad Q6600 2.4GHz quad-core CPU with 8MB L2 Cache. The differences, The E4600 sits on a 800MHz FSB and has 1MB L2 Cache per core. The Q6600 sits on a 1066MHz FSB and has 2MB L2 Cache per core, so really not apples to apples, but the same clock-speed, and I knew I would be able to interpret the results.

I was a little worried that Cinebench 9.5 wouldn’t run in KVM due to Video requirements but it fired up and worked just fine, so it’s my benchmark tool of choice tonight.

The results:

Native Machine xCPU test: 747

KVM Machine xCPU test: 753

Note: I ran the test twice, and took the second score from each machine…I tend to get more consistent results from Cinebench by testing in this manner.

Yup - that’s right, the virtualized machine won, and it actually won by about the same amount I would have expected “natively” given the slight advantange in L2 Cache and FSB speed.

I also want to note that I was very impressed at how Ubuntu handles processor scheduling with a multi-cpu virtual. When running the multi-threaded test in the guest machine, 2 of my CPU’s maxed out at 100% usage, rather than bouncing the threads around like Windows tends to do.