[Fedora-xen] Status of Xen on FC6 2.6.20.x - feedback

[Paul Wouters]

On Mon, 28 May 2007, John Casey wrote:

> After 7 days of testing, I have found a potential instability with
> Kernel-xen-2.6.20.2925 on FC6.   Symptom: machine spontaneously rebooted
> upon launch of tvtime.  Previous launches of tvtime, 10+, were successful.
> After reboot system was stable again (tvtime launched fine)
>
> Nothing useful in xend.log, xend.debug.log, log/messages.
>
> Environment:  FC6, Intel Cor2, Hauppauge WinTV PCI, 4G Ram, WinXP Pro guest
>
> Any suggestions on how to capture useful information if this happens again?

I have had similar experiences, however I found my problem. The machine is
overheating. With a non-xen kernel, I am seeing thermal throttling events in
syslog, but since xen does not support cpu frequency scaling, the cpu
actually overheats and the machine turns itself of before catching on fire.

The easiest way to track this is to use another machine with display to
login to the xen server and run:

while true; do sleep 1; sensors |grep Temp; done

When your xen machine reboots, you can see the last few temperature measurements,
to see if you have the same problem.

With a non-xen kernel, I could set the maximum scaling frequency to the
minimum cpu freq (2.8Ghz) using /etc/cpuspeed.conf. But the machine still
overheated. Getting a new cooler didn't help me either. Underclocking in
the BIOS failed. And updating the BIOS didnt make a difference.

I would never see the /proc entries for the fans, so I couldn't control
them. I know the fans can go faster, because when the machine boots,
before the OS loads, the fans blow like mad. But it seems I cannot set
them back in that mode from within Linux, so I've sort of written of
this machine now (Asus P5LD2-VM)

Paul