Re: [Xen-devel] Phoronix Xen vs KVM vs Virtualbox benchmark with Ubuntu 11.10

[Alex Bligh <alex@xxxxxxxxxxx>]

--On 1 November 2011 15:26:39 +0200 Pasi KÃrkkÃinen <pasik@xxxxxx> wrote:

> Actually the first page of the article says:
> "The only Xen issue encountered when testing it with an Ubuntu 11.10
> guest and host was the need for manually loading the xen-blkfront driver
> for disk support."
>
> So it sounds like they actually did use PVHVM drivers..
Allegedly 11.10 supports Xen out the box. However, the installer does not
ship with the PV drivers in, which causes the IDE driver to be unplugged
and thus means the installer can't access its disk. It requires substantial
fiddling to fix this. It thus wouldn't entirely surprise me if they weren't
actually using PV drivers even if they thought they were.

An alternative explanation is that by default KVM uses write-through
caching (i.e. read caching) on block devices on Ubuntu in dom0. I don't
know what Xen uses by default on 4.1.1, but certainly this makes KVM
outperforms Xen 3.3 (where there is no such caching) using naive testing on
our platforms.

The article also does not say what the block devices were backended onto.
As discussed ad nauseam here at the end of last week, distributions do not
ship with blktap (certainly Ubuntu doesn't), so unless they are using
blkback and partitions, it will use (if I remember this right) the Qemu
drivers and be slow. Certainly my take away from the thread last week was
that there was no chance of decent performance with a file backed block
device (without loopback mount). I have to say, getting Xen 4.1 to work was
non-trivial, even though we have working 3.1 configs. Getting kvm to work
and work well was incredibly simple. So it's quite possible Phoronix
stumbled into a non-optimal configuration.

--
Alex Bligh

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel