[fedora-virt] State of KVM paravirtualization

Pasi Kärkkäinen pasik at iki.fi
Mon Oct 26 08:58:12 UTC 2009


On Fri, Oct 23, 2009 at 01:00:44AM +0200, Dor Laor wrote:
> On 10/22/2009 02:07 PM, Ray Van Dolson wrote:
> >On Wed, Oct 21, 2009 at 06:39:01PM -0700, Giovanni Tirloni wrote:
> >>Hello,
> >>
> >>  I'm trying to understand where KVM stands regarding
> >>paravirtualization. I've seen the patches from 2007 by Ingo Molnar
> >>(http://people.redhat.com/mingo/kvm-paravirt-patches/) but couldn't
> >>find anything newer. Are they integrated to the mainstream kernel?
> >>
> >>  The reason I'm looking at that is because I've a server at home which
> >>doesn't have hardware-assisted virtualization support and running qemu
> >>in full virtualization mode is a bit slow. I've tried Xen and it sure
> >>is fast but I would like to experiment with KVM.
> >>
> >
> >Fortunately, RH will support Xen at least through RHEL 5's EOL.  I'm
> >sure the company I work at is not alone in having a lot of older but
> >still good hardware around that is great for Xen even if it's not cared
> >about by KVM developers so much :)
> 
> btw: it's not about care. It's not practical of kvm to implement it. By 
> the time it will work, even your Dell will be old... Also hw-assist virt 
> with nested paging is a killer, it outperforms any PV by any vendor.
> 

As I've understood this is the case with KVM, since pv-mmu stuff in KVM
isn't as fast as Xen has.

ie. with Xen PV is still faster than HVM (hw-assisted), at least for
some worksloads.

HVM (hw-assisted paging) has other/different performance hits.

-- Pasi




More information about the Fedora-virt mailing list