Brett Mahar writes:
>> I'm trying to use qemu to host a Fedora 13 guest.
>
> On that note, be aware there are vast differences in the speed of
> applications running on qemu virtual machines, depending on the host
> system. Eg, BOINC on Debian or Suse running inside a virtual machine, with
> an Ubuntu 10.04 host is about 10 times faster than on Slackware 13 host
> (MIPS measured by BOINC software).
That sounds suspiciously like you are either running fully emulated (eg: no
KVM support) on Slackware, or that you don't have virtio on that platform.
> I am guessing this is due to Ubuntu kernel being built with kernel same page
> merging, and Slackware without, but am not sure.
That seems unlikely: other than the cost to scan for pages to merge (which
hurts Ubuntu) that would only make a difference if you were pushing the limits
of your host memory capacity *and* KSM shared enough to avoid I/O in the
guest.
It would be interesting, to me, to know if either of those guesses were true,
because I have a current need to improve my skills diagnosing KVM performance
issues and feedback helps with that. :)
> Also inexplicably, having the host OS boot from an external usb hard drive
> causes major drop in the speed of the virtual machine, far beyond what would
> be caused by disk access times. I am not sure why this happens.
Well, it would slow down the host OS some, because USB is a fairly slow
and high latency protocol for disk access, but otherwise it shouldn't
fundamentally alter the running kernel.
Is the backing store for the virtual machine also on the USB disk? If so it
would point a finger toward guest I/O performance is the pain point, and the
introduction of USB would just be making a painful service slower.
Daniel
--
✣ Daniel Pittman✉ dan...@rimspace.net☎ +61 401 155 707
♽ made with 100 percent post-consumer electrons
--
SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/
Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
