Re: [Xen-users] Windows PV drivers destroying performance

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

I think the unknown product warning is benign. It is part of the unplug
protocol which removes the non-PV devices after boot, in the bit which
allows us to blacklist known bad versions, but a) we've never actually had
to do that and b) it won't blacklist an unknown set of drivers.

The "ACPI:debug:" messages are interesting, since there are lots, but
unless they are continuing to appear once the system is up and running I'm
inclined to suggest they are not the issue.

I don't know too much enough about the libvirt XML configuration format,
but nothing is jumping out at me.

With the GPL PV drivers apparently still being offline perhaps the
alternative ones which came out of the xenserver project but are now part
of xenproject are worth trying?See 
http://www.xenproject.org/developers/teams/windows-pv-drivers.html ?

Ian.

On Sat, 2015-07-25 at 19:05 +1000, Graham Crowe wrote:
> My config and qemu logs are attached. There is one rather telling line 
> in the qemu log...
> "Unknown PV product 2 loaded in guest"
> 
> I suspect it may be because I am using an old version of the PV drivers 
> (the most recent sites are offline, so I used an older one)
> 
> Is anyone able to point me to the latest PV drivers to try (WinXP, Win7 
> 32 bit and Win7 64 bit)?
> 
> As for the suggestion that I roll back to a snapshot. This is basically 
> a fresh Windows install (I am just only starting a trial of Windows VMs 
> - I already have a number of Linux VMs) so a snapshot won't help me much.
> 
> Thanks
> 
> 
> 
> On 24/07/2015 11:35 PM, Ian Campbell wrote:
> > On Fri, 2015-07-24 at 22:14 +1000, G Crowe wrote:
> > > I installed the PV drivers in some windows PVs as I thought it would
> > > improve performance, but the systems then became painfully slow to 
> > > use.
> > > 
> > > I ran some tests on hard disc performance on a newly installed Win7 
> > > 32
> > > bit system before and after installing the PV drivers from 
> > > univention.de
> > > (version 0.11.0.373) as the other sites appear to be offline. The 
> > > host
> > > machine is running Fedora 21 with kernel 3.19.3.
> > > Without the PV drivers (with QEMU hardware) hard disc performance was
> > > 4409 I/O per second and 9.03Mbps.
> > > With the PV drivers installed this dropped to 207 I/O per second and
> > > 0.43Mbps.
> > > 
> > > Without the PV drivers, the remote desktop screen would refresh 
> > > almost
> > > instantly, but with the PV drivers it could take up to 10 seconds 
> > > just
> > > to refresh the screen.
> > > 
> > > I don't understand this, as I though that the sole purpose of the PV
> > > drivers was to improve I/O performance,
> > Your expectation here is correct.
> > 
> > >   yet they have made the system so
> > > slow that it is virtually unusable.
> > > 
> > > Am I doing something wrong here?
> > It seems unlikely, I don't think there is much you can have done wrong,
> > it's supposed to just work ;-) more likely some sort of bad interaction
> > between some (possibly buggy) component and another.
> > 
> > I don't know much about the Windows PV drivers, so hopefully someone 
> > else
> > can chip in with some tips/knowledge.
> > 
> > Seeing your guest cfg file and any relevant logs from /var/log/xen 
> > might
> > have some sort of clue.
> > 
> > Ian.
> > 
> > _______________________________________________
> > Xen-users mailing list
> > Xen-users@xxxxxxxxxxxxx
> > http://lists.xen.org/xen-users
> 

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users