[CentOS-virt] Problem with CentOS 5.1 / Xen / Migration : domU time

Mon Jan 21 20:30:00 UTC 2008
Frederic SOULIER <frederic.soulier at univ-tlse1.fr>

Tim Verhoeven a écrit :
> On Jan 14, 2008 1:40 PM, Frederic SOULIER
> <frederic.soulier at univ-tlse1.fr> wrote:
>> I'm having issue with my Centos 5.1 /Xen installation.
>> I'm having 2  dom0 running 2.6.18-53.1.4.el5xen (x86_64).
>> These two dom0 are connected to a SAN.
>> These two dom0 hosts multiple domU like Centos 4.5 (2.6.9-55.0.9.ELxenU)
>> and Centos 5.1 (2.6.18-53.1.4.el5xen).
>> These two dom0 run ntpd to synchronize dom0 time.
>> When i start a virtual machine on the first dom0 all seems good, the
>> domU time is like the dom0 time.
>> The problem appears when i migrate the domU to another dom0.
>> After the migration, the time in domU on the new dom0 is totally false
>> (we can have more than one day of desynchronisation) and the time on the
>> domU seems to be blocked (the clock indicates always the same time ).
>> In the xend Logs we can't see any error message with the debug mode.
>> The solution for this moment is to restart the domU in the new dom0 to
>> be synchronized again.
>> We manage a lot of authentications services like CAS or Shibboleth which
>> uses timestamp for security validation and this problem is very annoying
>> for us.
>> Anyone has encountered this problem ?
>> Any idea how to solve it ?
> Hi Frederic,
> I've also seen this issue (CentOS 5.1, 64bit in Intel Xeons using
> PVM). I've also found a solution to it.
> Put this in /etc/sysctl.conf :
> xen.independent_wallclock = 1
> and then execute sysctl.
> What happens then is that the domU will do it's own timekeeping and no
> longer follow the dom0. Now this also means that you need to run a
> ntpd inside the domU to make sure it's clock keeps syncronised.
> That solved it for me. After migrations the clock kept running and
> remained in sync.
> Regards,
> Tim
Hi Tim,

Thanks for your response.
Using ntp directly in my domU seems to solve my problems :-)

My last question is about the source problem.
Before we used Centos 5.0 with Xen in the same configuration and all was ok.
The problem appears after the Centos 5.1 upgrade ( Xen was also upgraded 
from 3.0 to 3.1 in this release).  Anyone known if this problem is local 
to my configuration or if it could be a bug in the new release ?