Xen issues - Was Xen network performance problem during kickstart.

[Stephen C. Tweedie]

Hi,

On Tue, 2006-02-21 at 01:11 +0900, Naoki wrote:

> On Mon, 2006-02-20 at 08:41 -0500, Stephen C. Tweedie wrote:
> > >     ------------[ cut here ]------------
> > > kernel BUG at include/asm/xor.h:633!
> > > invalid opcode: 0000 [#1]
> > 
> > That was supposed to be fixed (bug 177644); I'll chase this up and see
> > what's happening here.
> 
> Cheers, just tried as of a few minutes ago and it's still with me..

I've just tried too, but can't reproduce.  Are you sure you're running
the 1955 kernel in both dom0 and domU, and the hypervisor from that
dom0?

> After install was done there were also a few errors on start up
> (something about /dev/map ).

Details would be helpful. :-)

> Just had it complete an install in an hour. Which is better.  Ran some
> tests.
> 
> wget over local net ( layer 2 switched no routing hops ) of
> openoffice-core RPM (77MB) :
> 
> 100%[====================================>] 80,504,759     1.52M/s
> ETA 00:00 
> 
> 932.47 KB/s
> 
> The same test from domain-0 :
> 
> 100%[=================================================================================================================>] 80,504,759    10.03M/s    ETA 00:00
> 
> 10.40 MB/s.

Chris Wright recently found a bug that could result in bad network
performance amongst other things; but I'd expect that to affect dom0 as
much as domU.

What sort of config is this box?  SMP/HT etc?

> That's a very big difference. Normally I'd say it was duplex problem but
> I can't check that because of this :
> 
> [root at localhost ~]# ethtool eth0
> Settings for eth0:
> No data available

There's no such thing as an eth setting for domU: the device is emulated
entirely in software.  For dom0, there is still a real physical device,
but the network scripts rename that to "peth0" and set up a virtual eth0
for bridging purposes.  You should still be able to run ethtool on the
peth0 device.

Cheers,
 Stephen