Re: [Xen-devel] Questions about GPLPV stability tests

[Roderick Colenbrander <thunderbird2k@xxxxxxxxx>]

On Sun, Dec 11, 2011 at 4:52 AM, Pasi Kärkkäinen <pasik@xxxxxx> wrote:
> On Fri, Dec 09, 2011 at 02:02:31PM -0800, Roderick Colenbrander wrote:
>> One interesting observation. This morning I had another of my stress
>> machines with the problem. I never had it on this problem before and
>> it didn't have any of the new logging / software updates yet.
>>
>> The system was in the same state as the other machine which I reported
>> about before. I tried SysRq stuff and other things. While I was about
>> to reboot the system, a login prompt appeared again on the VGA. I
>> don't know whether any of the stuff I did triggered it or not. Anyway
>> it means Linux is not really dead. I tried logging, but I don't even
>> see characters appearing. The system feels to be very, very, very
>> slow.
>>
>
> Hmm.. I wonder if this is the same issue I'm sometimes seeing on my laptop..
> suddenly it starts becoming slower, and after a while it's *very* slow..
> not dead, but unusably slow..
>
> I haven't had time to (try to) debug it..
>
> -- Pasi
>

I'm seeing slowness issues on our systems as well. As in some code
starts running really, really slowly. Local TCP 'heartbeat' like
mechanism from Dom0 to a DomU timing out. Code which should execute
quickly becoming orders of magnitude slower for no obvious reason.
Typically we see evidence of this in logging.

I felt there was a connection between this slowness and the
'unresponsive Dom0' but I haven't been able to confirm this. Normally
we see weird things in our logs, but on the unresponsive systems I
didn't see anything strange in the logs yet. Most likely the logs
weren't synced to disk yet.

After more investigation it seems that in my case all issues, seem to
happen after the DomU is up and we somehow 'start' using it. During
startup of our own software, both DomU and Dom0 would at least see a
spike in cpu/disk activity before things settle down a bit. My feeling
is (based on some logs) that this is when Dom0 sometimes becomes
unresponsive.

I'm still running tests on a number of machines, but it takes ages to
get results.

Roderick

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