CentOS virt - Oct 2008 - Unstable VMware 1.0.5 host after 4.7 upgrade

After upgrading to 4.7 x86_64 on all our boxes (from 4.6), we have 3
VMware Server 1.0.5 hosts that are unstable. Sometimes they
spontaneously reboot, other times they freeze and require a hard
reset. Under 4.6, they were rock solid.

Failures tend to be associated with a guest start-up event, but are
often at random times as well. There is no useful evidence in either
the vmware or messages log files on the host.

The other thing that changed when we upgraded is that we upgraded our
CentOS 4.6 (x86) guests to 4.7 and instead of recompiling a new kernel
with HZ=100, we tested and implemented the newly available divider=10
kernel parameter. Initially, everything was fine with the divider
option, but now I'm not so sure.

Probably not a coincidence is the fact that the 3 hosts that are
affected do NOT have VT capable processors. Our quad-core Xeon boxes
with VT have not been a problem.

I'm in the process of going back to an HZ=100 kernel. Of course, my
first guest system reboot with the new kernel triggered a host reboot.
Arrrggghhh.

Does anyone have similar experiences? Does anyone have a *stable*
VMware Server 1.0.x host on 2.6.9-78.0.1 with non-VT processors? Is
anyone else using divider=10 with VMware 1.0.x?

-- 
Jeff

Does anyone have similar experiences? Does anyone have a
*stable*
> VMware Server 1.0.x host on 2.6.9-78.0.1 with non-VT processors? Is
> anyone else using divider=10 with VMware 1.0.x?
>
>
Jeff,

I have a stable VMware Server 1.0.7  but it is running on CentOS 5. The
server has a non-VT proccesor.

Also, instead of recompiling the kernel you can grab one already compiled
with the Hz set to 100 from here:

http://people.centos.org/tru/kernel-vm/

They are working fine on my guest systems.

Best.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.centos.org/pipermail/centos-virt/attachments/20081008/1f86dab9/attachment-0004.html>

On Wed, Oct 8, 2008 at 4:13 PM, Ramon Nieto <rnieto.linux at gmail.com>
wrote:
>> Does anyone have similar experiences? Does anyone have a *stable*
>> VMware Server 1.0.x host on 2.6.9-78.0.1 with non-VT processors? Is
>> anyone else using divider=10 with VMware 1.0.x?
>>
>
> Jeff,
>
> I have a stable VMware Server 1.0.7  but it is running on CentOS 5. The
> server has a non-VT proccesor.
Not quite apples to apples comparison, but it gives me some hope. Are
you using the tick divider option in your guests? It's still possible
that my problem is the kernel tick divider on the CentOS 4 kernel. I'm
trolling for evidence of what the source of the problem might be
instead of stabbing in the dark.
> Also, instead of recompiling the kernel you can grab one already compiled
> with the Hz set to 100 from here:
>
> http://people.centos.org/tru/kernel-vm/
>
I'm aware of those but I like doing things myself. Thanks.

-- 
Jeff