clock drift in Xen 4.0.0

clock drift in Xen 4.0.0

[James Harper]

Are there any known problems with clock drift in Xen 4.0.0? One DomU
with an uptime of 49 days is now 7 minutes in front. I never noticed the
problem before I upgraded.

/proc/sys/xen/independent_wallclock=0

James

Re: clock drift in Xen 4.0.0

[Keir Fraser]

On 25/05/2010 07:50, "James Harper" <james.harper@bendigoit.com.au> wrote:

> Are there any known problems with clock drift in Xen 4.0.0? One DomU
> with an uptime of 49 days is now 7 minutes in front. I never noticed the
> problem before I upgraded.
> 
> /proc/sys/xen/independent_wallclock=0

Are you running NTP? 100ppm drift isn't unbelievable without external
synchronisation.

 -- Keir

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

RE: clock drift in Xen 4.0.0

[James Harper]

> On 25/05/2010 07:50, "James Harper" <james.harper@bendigoit.com.au>
wrote:
> 
> > Are there any known problems with clock drift in Xen 4.0.0? One DomU
> > with an uptime of 49 days is now 7 minutes in front. I never noticed
the
> > problem before I upgraded.
> >
> > /proc/sys/xen/independent_wallclock=0
> 
> Are you running NTP? 100ppm drift isn't unbelievable without external
> synchronisation.
> 

That's 7 minutes in front of the Dom0 that it's sitting on. The Dom0
runs ntp. Another DomU on the same system is ahead by a similar amount
(9 minutes over 74 days). I thought that with independent_wallclock=0
that they should stay in step.

I actually just noticed that I'm running 4.0.0-rc6, not 4.0.0 proper...
d'oh. I think I should probably update before thinking about it much
more.

James

Re: clock drift in Xen 4.0.0

[Keir Fraser]

On 25/05/2010 08:01, "James Harper" <james.harper@bendigoit.com.au> wrote:

>> Are you running NTP? 100ppm drift isn't unbelievable without external
>> synchronisation.
>> 
> 
> That's 7 minutes in front of the Dom0 that it's sitting on. The Dom0
> runs ntp. Another DomU on the same system is ahead by a similar amount
> (9 minutes over 74 days). I thought that with independent_wallclock=0
> that they should stay in step.
> 
> I actually just noticed that I'm running 4.0.0-rc6, not 4.0.0 proper...
> d'oh. I think I should probably update before thinking about it much
> more.

It probably depends on your dom0 and domU kernels also. Have they both been
observed to behave 'properly' before? In particular I'm not sure pv_ops
behaves same as ealrier kernels.

 -- Keir

RE: clock drift in Xen 4.0.0

[Dan Magenheimer]

> From: Keir Fraser [mailto:keir.fraser@eu.citrix.com]
> Subject: Re: [Xen-devel] clock drift in Xen 4.0.0
> 
> >> Are you running NTP? 100ppm drift isn't unbelievable without
> external
> >> synchronisation.
> >
> > That's 7 minutes in front of the Dom0 that it's sitting on. The Dom0
> > runs ntp. Another DomU on the same system is ahead by a similar
> amount
> > (9 minutes over 74 days). I thought that with independent_wallclock=0
> > that they should stay in step.
> >
> > I actually just noticed that I'm running 4.0.0-rc6, not 4.0.0
> proper...
> > d'oh. I think I should probably update before thinking about it much
> > more.
> 
> It probably depends on your dom0 and domU kernels also. Have they both
> been
> observed to behave 'properly' before? In particular I'm not sure pv_ops
> behaves same as ealrier kernels.

James --

"DomU" may imply PV but just to verify is this a PV domain or HV?

If an HV, Keir, didn't the default timer_mode change in 4.0?

Also, if HV, isn't independent_wallclock ignored?

Re: clock drift in Xen 4.0.0

[Keir Fraser]

On 25/05/2010 15:34, "Dan Magenheimer" <dan.magenheimer@oracle.com> wrote:

> "DomU" may imply PV but just to verify is this a PV domain or HV?
> 
> If an HV, Keir, didn't the default timer_mode change in 4.0?

Default is still 1 afaiaa.

> Also, if HV, isn't independent_wallclock ignored?

RE: clock drift in Xen 4.0.0

[Dan Magenheimer]

> From: Keir Fraser [mailto:keir.fraser@eu.citrix.com]
> Subject: Re: [Xen-devel] clock drift in Xen 4.0.0
> 
> On 25/05/2010 15:34, "Dan Magenheimer" <dan.magenheimer@oracle.com>
> wrote:
> 
> > "DomU" may imply PV but just to verify is this a PV domain or HV?
> >
> > If an HV, Keir, didn't the default timer_mode change in 4.0?
> 
> Default is still 1 afaiaa.

It was changed in cs 18554, September 2008, so I think that
was actually between 3.3 and 3.4.

Re: clock drift in Xen 4.0.0

[Jeremy Fitzhardinge]

On 05/25/2010 12:01 AM, James Harper wrote:
> That's 7 minutes in front of the Dom0 that it's sitting on. The Dom0
> runs ntp. Another DomU on the same system is ahead by a similar amount
> (9 minutes over 74 days). I thought that with independent_wallclock=0
> that they should stay in step.
>
> I actually just noticed that I'm running 4.0.0-rc6, not 4.0.0 proper...
> d'oh. I think I should probably update before thinking about it much
> more.
>   

What domU and dom0 kernel are you running?  pvops always has independent
wallclock.

    J

RE: clock drift in Xen 4.0.0

[James Harper]

> 
> On 05/25/2010 12:01 AM, James Harper wrote:
> > That's 7 minutes in front of the Dom0 that it's sitting on. The Dom0
> > runs ntp. Another DomU on the same system is ahead by a similar
amount
> > (9 minutes over 74 days). I thought that with
independent_wallclock=0
> > that they should stay in step.
> >
> > I actually just noticed that I'm running 4.0.0-rc6, not 4.0.0
proper...
> > d'oh. I think I should probably update before thinking about it much
> > more.
> >
> 
> What domU and dom0 kernel are you running?  pvops always has
independent
> wallclock.
> 

Dom0 is 2.6.31.12 (pvops), domU is 2.6.18.8 (not pvops). Which one
matters here?

James

Re: clock drift in Xen 4.0.0

[Keir Fraser]

On 26/05/2010 10:33, "James Harper" <james.harper@bendigoit.com.au> wrote:

>> What domU and dom0 kernel are you running?  pvops always has
> independent
>> wallclock.
> 
> Dom0 is 2.6.31.12 (pvops), domU is 2.6.18.8 (not pvops). Which one
> matters here?

I think pv_ops doesn't play with dependent-wallclock either way round: It
won't resync its wallclock with Xen when running as domU; nor will it
periodically push its wallclock down to Xen when running as dom0. So if you
have pv_ops as the dom0 or the domU, you'd better run NTP in the domU.

 -- Keir

RE: clock drift in Xen 4.0.0

[James Harper]

> 
> On 26/05/2010 10:33, "James Harper" <james.harper@bendigoit.com.au>
wrote:
> 
> >> What domU and dom0 kernel are you running?  pvops always has
> > independent
> >> wallclock.
> >
> > Dom0 is 2.6.31.12 (pvops), domU is 2.6.18.8 (not pvops). Which one
> > matters here?
> 
> I think pv_ops doesn't play with dependent-wallclock either way round:
It
> won't resync its wallclock with Xen when running as domU; nor will it
> periodically push its wallclock down to Xen when running as dom0. So
if you
> have pv_ops as the dom0 or the domU, you'd better run NTP in the domU.
> 

Thanks. Will it ever support dependent-wallclock?

James

Re: clock drift in Xen 4.0.0

[Keir Fraser]

On 26/05/2010 10:53, "James Harper" <james.harper@bendigoit.com.au> wrote:

>> I think pv_ops doesn't play with dependent-wallclock either way round:
> It
>> won't resync its wallclock with Xen when running as domU; nor will it
>> periodically push its wallclock down to Xen when running as dom0. So
> if you
>> have pv_ops as the dom0 or the domU, you'd better run NTP in the domU.
>> 
> 
> Thanks. Will it ever support dependent-wallclock?

I don't think Jeremy likes dependent-wallclock. :-)

 K.

Re: clock drift in Xen 4.0.0

[Stefano Stabellini]

On Wed, 26 May 2010, Keir Fraser wrote:
> On 26/05/2010 10:33, "James Harper" <james.harper@bendigoit.com.au> wrote:
> 
> >> What domU and dom0 kernel are you running?  pvops always has
> > independent
> >> wallclock.
> > 
> > Dom0 is 2.6.31.12 (pvops), domU is 2.6.18.8 (not pvops). Which one
> > matters here?
> 
> I think pv_ops doesn't play with dependent-wallclock either way round: It
> won't resync its wallclock with Xen when running as domU; nor will it
> periodically push its wallclock down to Xen when running as dom0. So if you
> have pv_ops as the dom0 or the domU, you'd better run NTP in the domU.

A while ago I sent a patch to pvops to fix this problem (look for
"[PATCH] [PVOPS] [UPDATE] dom0 sync xen wallclock" in the
archives), but Jeremy didn't like it :)

RE: clock drift in Xen 4.0.0

[James Harper]

> 
> >> What domU and dom0 kernel are you running?  pvops always has
> > independent
> >> wallclock.
> >
> > Dom0 is 2.6.31.12 (pvops), domU is 2.6.18.8 (not pvops). Which one
> > matters here?
> 
> I think pv_ops doesn't play with dependent-wallclock either way round:
It
> won't resync its wallclock with Xen when running as domU; nor will it
> periodically push its wallclock down to Xen when running as dom0. So
if you
> have pv_ops as the dom0 or the domU, you'd better run NTP in the domU.
> 

Just for the record, with a pvops dom0 and a non-pvops domu, you still
need to set independent_wallclock=1 for ntp to be able to modify the
clock. Without that change, ntp tells me:

time reset -409.699016 s
...
time reset -409.776693 s
...
time reset -409.854404 s
...
time reset -409.930385 s
...
time reset -410.009187 s

but never actually adjusts the time!

James