ntpd Frequency (drift) Jumps

I just started using ntpd. It seems to be most accurate in terms of the
offset remaining in the range [-10, 10] ms when the frequency drift
is -41.xxx. However, every now and then, for no apparent reason, the
frequency will jump +- 50 to 100 PPM. When it does, the offset increases
greatly in magnitude, becomes wildly erratic, and ntpd adjustes the clock
frequently. In a day or so ntpd will slowly reduce the drift to -4y.xxx,
then
it will start the cycle again. I am using ntpd 4.2.0b from
[email]ntp-4.2.0b@1.1436mbg-xmas-o-win32-setup.exe[/email]. Has anyone else
noticed this behavior? Do you know of a remedy?

C Elliott

Charles Elliott wrote:[color=blue]
> I just started using ntpd. It seems to be most accurate in terms of
> the offset remaining in the range [-10, 10] ms when the frequency
> drift is -41.xxx. However, every now and then, for no apparent reason,
> the
> frequency will jump +- 50 to 100 PPM. When it does, the offset
> increases greatly in magnitude, becomes wildly erratic, and ntpd
> adjustes the clock frequently. In a day or so ntpd will slowly
> reduce the drift to -4y.xxx, then
> it will start the cycle again. I am using ntpd 4.2.0b from
> [email]ntp-4.2.0b@1.1436mbg-xmas-o-win32-setup.exe[/email]. Has anyone else
> noticed this behavior? Do you know of a remedy?
>
> C Elliott[/color]

Do you have the Multi-Media timer option set? It was critical to getting
good stability with my systems:

[url]http://www.david-taylor.myby.co.uk/ntp/ntp-events.htm#glitch[/url]

David

Yes, the Multi-Media timer option (-M) is set.

CHE

"David J Taylor" <david-taylor@blueyonder.co.not-this-bit.nor-this-part.uk>
wrote in message news:9uNpg.94762$wl.82500@text.news.blueyonder.co.uk...[color=blue]
> Charles Elliott wrote:[color=green]
>> I just started using ntpd. It seems to be most accurate in terms of
>> the offset remaining in the range [-10, 10] ms when the frequency
>> drift is -41.xxx. However, every now and then, for no apparent reason,
>> the
>> frequency will jump +- 50 to 100 PPM. When it does, the offset
>> increases greatly in magnitude, becomes wildly erratic, and ntpd
>> adjustes the clock frequently. In a day or so ntpd will slowly
>> reduce the drift to -4y.xxx, then
>> it will start the cycle again. I am using ntpd 4.2.0b from
>> [email]ntp-4.2.0b@1.1436mbg-xmas-o-win32-setup.exe[/email]. Has anyone else
>> noticed this behavior? Do you know of a remedy?
>>
>> C Elliott[/color]
>
> Do you have the Multi-Media timer option set? It was critical to getting
> good stability with my systems:
>
> [url]http://www.david-taylor.myby.co.uk/ntp/ntp-events.htm#glitch[/url]
>
> David
>[/color]

Charles Elliott wrote:[color=blue]
> Yes, the Multi-Media timer option (-M) is set.
>
> CHE[/color]

Well, that leaves drivers taking too much time time at interrupt level
(PIO mode on the disks), or something else. It could be bad hardware on
your motherboard. Do you have any spread-spectrum or power-saving
functions set in your BIOS?

Cheers,
David

No, this motherboard does not seem to have any spread-spectrum functions; I
looked thru the BIOS
setup screen and in the manual. AFIK power-saving is disabled. I do run
Seti@Home on both CPUs, which can make the system seem sluggish at times.
Also hyperthreading is enabled so
it appears as 2 CPUs. I was looking thru the ntpd code and in at least one
place they claim to
have made a fix that will work on single processor systems, but might not
work on multiprocessor
systems. I am considering moving it over to a Server 2003 single-cpu AMD 64
machine to see if
the same problem appears.

The problem always seems to happen when the time offset is very small (< 5
ms) and is about to change sign. I wrote a Java routine to call ntpq every
5 seconds and I have been watching it for the last half hour. The offset
just did change from -0.979 ms to 2.711 ms with no untoward result.

It always happens late in the evening or in the very early morning. Verizon
is making changes to its DSL system, and so there have been some outages and
slow downs at these times. Also the 50 to 100 and one 500 PPM frequency
jumps are always accompanid by a large time adjustment (on the order of .180
seconds).

Do you have any more ideas?

CHE

"David J Taylor" <david-taylor@blueyonder.co.not-this-bit.nor-this-part.uk>
wrote in message news:tROpg.94815$wl.28713@text.news.blueyonder.co.uk...[color=blue]
> Charles Elliott wrote:[color=green]
>> Yes, the Multi-Media timer option (-M) is set.
>>
>> CHE[/color]
>
> Well, that leaves drivers taking too much time time at interrupt level
> (PIO mode on the disks), or something else. It could be bad hardware on
> your motherboard. Do you have any spread-spectrum or power-saving
> functions set in your BIOS?
>
> Cheers,
> David
>[/color]

Charles Elliott wrote:
[][color=blue]
> Do you have any more ideas?
>
> CHE[/color]

Sorry, I don't. My own measurements with Windows are on a +/- 100ms
timescale rather than anything finer, and I now have a simple stratum 1
source rather than relying purely on Internet sources.

Cheers,
David