Re: [Xen-users] Xen system hang or freeze

[Martin Fernau <m.fernau@xxxxxxxxxx>]

Peter,

at least I would thank you for supporting in this situation!
To answer your questions:
> 1. How many vcpus does each of your domUs have?
This vary. Mostly I've set 'vcpu' to 2.

> 2. Do you define pinning, cap or weight for your domUs?
no. They are not restricted in this way. I just limited the number of cpu 
cores to the domains (cpus=...)

> 3. Does sar on the guests show a high %steal before the problem
> occurred ?
I didn't hat this tool running at the time of the crash. I also have 3 Windows 
guests which can't run this program.

> 4. Do you limit the number of vcpus that Dom0 has? If not, I would
> suggest that you try this an dsee if the problem occurs when you know
> that there should be free vcpus for each of your domUs.
I didn't limit Dom0 so far.

Because I have 8 cores for 4 guests I could do the following:

Assign CPU 1 to dom0 -> "(dom0-cpus 1)" in xend-config.sxp
2,3 to Guest1 (windows 2003)
4,5,6 to Guest2 (windows 2003)
7 to Guest3 (gentoo linux)
8 to Guest4 (windows xp)

With this settings all domains using its own processor.

Do I need to change vcpus too? I'm afraid if I change this back to 1 that 
windows wants to activate itself again...


Am Dienstag, 5. Mai 2009 schrieb Peter Booth:
> Martin,
>
> You and Nick both see this problem on a system with eight cores.
> You have 4 domUs, Nick has 3 domUs.
>
> I think a reasonable hypothesis is that this is some kind of resource
> starvation/livelock/deadlock scenario.
> Some questions that might help see  how similar your scenario and
> Nick's are:
>
> 1. How many vcpus does each of your domUs have?
> 2. Do you define pinning, cap or weight for your domUs?
> 3. Does sar on the guests show a high %steal before the problem
> occurred ?
> 4. Do you limit the number of vcpus that Dom0 has? If not, I would
> suggest that you try this an dsee if the problem occurs when you know
> that there should be free vcpus for each of your domUs.
>
> Peter
>
>
> Do you limit your Dom0
>
> On May 5, 2009, at 2:42 AM, Martin Fernau wrote:
> > Hello,
> >
> > did you find out anything helpful about this problem?
> > I just had this freeze this morning again. Server freezed and no
> > guests
> > working any more. After 5 weeks of normal running. It seems that
> > this is my
> > intervall of this problem.
> >
> > I really appreciated anything helpful about this! I'm totally stuck
> > with this
> > problem...
> >
> > Martin
> >
> > Am Freitag, 24. April 2009 schrieb Nick Anderson:
> >> On Thu, Apr 23, 2009 at 10:01:02PM -0400, Peter Booth wrote:
> >>> Nick,
> >>> When you got the "soft lockup" errors were they always with CPU0?
> >>
> >> Yes I believe so. But since I updated to the latest debian patched
> >> kernel I havent seen that soft lockup error.
> >>
> >>> Do you have any nesting in your LVM definitions?
> >>
> >> No
> >>
> >>> Is dom0 making use of LVM?
> >>
> >> No it is not.
> >>
> >>> Does the host have a wireless card?
> >>
> >> No
> >>
> >>> What graphics card does it have?
> >>
> >> Just the onboard graphics 01:05.0 VGA compatible controller: ATI
> >> Technologies Inc ES1000 (rev 02)
> >>
> >>> There is a Xen 2.6.18 kernel bug that might be related to what you
> >>> see,
> >>> as well as a patch:
> >>> http://www.mail-archive.com/debian-kernel@xxxxxxxxxxxxxxxx/msg40893.htm
> >>>l
> >>
> >> Yes I think that patch is in the latest kernel update which made the
> >> soft lockup messages go away. Odly everyone else that noted the soft
> >> lockup errors only noted them as an annoyance. I don't recall ever
> >> seeing anyone link the messages with any actual undesirable behavior
> >> outside of the msgs themselves.
> >>
> >>> I appreciate that this freezing occurs when the system is quiet,
> >>> but it
> >>> seems as if the hypervisor CPU scheduler might play some part in
> >>> this,
> >>> such that the guest vcpus don't get scheduled on a pcpu. If it
> >>> were me I
> >>> would try limiting dom0 to four vcpus - not as a long-term
> >>> configuration, but to see whether the problem recurred when we
> >>> know that
> >>> each domU should have an available CPU.
> >>
> >> I was considering that. I also thought it interesting that each of
> >> the
> >> domUs are running on vcpu0. I figured they would automatically
> >> distribute to a free vcpu.
> >
> > --
> > Mit freundlichem Gruß,
> > Martin Fernau
> >
> >
> > CPS Entwicklungsgesellschaft für EDV-Lösungen mbH
> > Gartenstraße 42 - 37269 Eschwege
> >
> > Telefon (0 56 51) 95 99-0
> > Telefax (0 56 51) 95 99-90
> >
> > eMail m.fernau@xxxxxxxxxx
> > Internet http://www.cps-net.de
> >
> > Handelsregister Eschwege, HRB 1585
> > Geschäftsführer, Wilfried Fernau
> > Steuernummer 026 230 40308
> > USt-ID-Nr. DE 178 554 522
> >
> > _______________________________________________
> > Xen-users mailing list
> > Xen-users@xxxxxxxxxxxxxxxxxxx
> > http://lists.xensource.com/xen-users

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