Remarkably good CPU oscillators?

I have a tiny mini-ITX motherboard that has served as a stratum-1 for
several years.

It sits in the basement, so no radically wild swings in temperature or
other environmental factors.

Still, this morning when I look back at a few years of loopstats, the
oscillator in this thing is remarkably, and maybe unbelievably, stable.
It has stuck in the -2.5 to -5.5 PPM range over the past two years, and
in a typical daily swing it might go up and down by a 0.5 PPM.

Is this statistic just "too good to be true"? Or did I luck out when I
bought this box many years ago?

It has two refclocks available, wwv_audio and a Z3801A (via hpgps and a
PPS nanokernel), and cross-checking between the two I believe them
both.

Tim.

In article <1162738456.358848.206590@h48g2000cwc.googlegroups.com>,
[email]shoppa@trailing-edge.com[/email] wrote:
[color=blue]
> Is this statistic just "too good to be true"? Or did I luck out when I
> bought this box many years ago?[/color]

No. If anything it is bordering on the too big to be good. This is
one of the main reasons why I think that ntpd should be much more reluctant
to apply corrections of more than about 3 to 5 ppm, once the frequency has
been calibrated. My feeling is that it should limit the corrections to
this range unless there is an extended period indicative of a different
frequency. It might be reasonable to allow, say, 0.1ppm per day to
account for long term drift.

[email]shoppa@trailing-edge.com[/email] wrote:
[color=blue]
> I have a tiny mini-ITX motherboard that has served as a stratum-1 for
> several years.
>
> It sits in the basement, so no radically wild swings in temperature or
> other environmental factors.
>
> Still, this morning when I look back at a few years of loopstats, the
> oscillator in this thing is remarkably, and maybe unbelievably, stable.
> It has stuck in the -2.5 to -5.5 PPM range over the past two years, and
> in a typical daily swing it might go up and down by a 0.5 PPM.
>
> Is this statistic just "too good to be true"? Or did I luck out when I
> bought this box many years ago?
>
> It has two refclocks available, wwv_audio and a Z3801A (via hpgps and a
> PPS nanokernel), and cross-checking between the two I believe them
> both.
>
> Tim.
>[/color]

If this oscillator is "free running" it is, indeed, remarkable. If it
is GPS disciplined it's what you should expect!

In article <1162738456.358848.206590@h48g2000cwc.googlegroups.com>,
[email]shoppa@trailing-edge.com[/email] writes:[color=blue]
>I have a tiny mini-ITX motherboard that has served as a stratum-1 for
>several years.
>
>It sits in the basement, so no radically wild swings in temperature or
>other environmental factors.
>
>Still, this morning when I look back at a few years of loopstats, the
>oscillator in this thing is remarkably, and maybe unbelievably, stable.
>It has stuck in the -2.5 to -5.5 PPM range over the past two years, and
>in a typical daily swing it might go up and down by a 0.5 PPM.
>
>Is this statistic just "too good to be true"? Or did I luck out when I
>bought this box many years ago?[/color]

Seems fine to me. I'll bet you can back compute the temperature
from the drift.

My unheated/uncooled setup swings by 2 ppm each day. That's with
a 10 F temperature swing.

[color=blue]
>It has two refclocks available, wwv_audio and a Z3801A (via hpgps and a
>PPS nanokernel), and cross-checking between the two I believe them
>both.
>
>Tim.
>[/color]

--
The suespammers.org mail server is located in California. So are all my
other mailboxes. Please do not send unsolicited bulk e-mail or unsolicited
commercial e-mail to my suespammers.org address or any of my other addresses.
These are my opinions, not necessarily my employer's. I hate spam.

Tim,

See FreeBSD 6.1 rackety.udel.edu or Solaris 10 pogo.udel.edu; both
behave nominally as you observe with GPS and PPS. Both are near each
other in our machine room; both keep the clock generally within a few
tens of microseconds, even with a client population of sevaral hundred.

Dave

Dave

[email]shoppa@trailing-edge.com[/email] wrote:
[color=blue]
> I have a tiny mini-ITX motherboard that has served as a stratum-1 for
> several years.
>
> It sits in the basement, so no radically wild swings in temperature or
> other environmental factors.
>
> Still, this morning when I look back at a few years of loopstats, the
> oscillator in this thing is remarkably, and maybe unbelievably, stable.
> It has stuck in the -2.5 to -5.5 PPM range over the past two years, and
> in a typical daily swing it might go up and down by a 0.5 PPM.
>
> Is this statistic just "too good to be true"? Or did I luck out when I
> bought this box many years ago?
>
> It has two refclocks available, wwv_audio and a Z3801A (via hpgps and a
> PPS nanokernel), and cross-checking between the two I believe them
> both.
>
> Tim.
>[/color]

In article <T1162744621@djwhome.demon.co.uk>,
[email]david@djwhome.demon.co.uk[/email] (David Woolley) writes:[color=blue]
>In article <1162738456.358848.206590@h48g2000cwc.googlegroups.com>,
>shoppa@trailing-edge.com wrote:
>[color=green]
>> Is this statistic just "too good to be true"? Or did I luck out when I
>> bought this box many years ago?[/color]
>
>No. If anything it is bordering on the too big to be good. This is
>one of the main reasons why I think that ntpd should be much more reluctant
>to apply corrections of more than about 3 to 5 ppm, once the frequency has
>been calibrated. My feeling is that it should limit the corrections to
>this range unless there is an extended period indicative of a different
>frequency. It might be reasonable to allow, say, 0.1ppm per day to
>account for long term drift.[/color]

The short term drift depends strongly on the temperature.
[url]http://www.megapathdsl.net/~hmurray/ntp/drift.gif[/url]
The big spike on the pink at -14 and -38 is the self heating
on the CPU crystal from a big cron job.
The small blue spike is the self heating on the crystal
driving the TOY clock.

--
The suespammers.org mail server is located in California. So are all my
other mailboxes. Please do not send unsolicited bulk e-mail or unsolicited
commercial e-mail to my suespammers.org address or any of my other addresses.
These are my opinions, not necessarily my employer's. I hate spam.

David,

Very dangerous to clamp the frequency. In the normal course the time
constant ramps up to 1024 s, which makes the loop very, very stiff and
changes exceeding 0.1 PPM are very unlikely.

Dave

David Woolley wrote:
[color=blue]
> In article <1162738456.358848.206590@h48g2000cwc.googlegroups.com>,
> [email]shoppa@trailing-edge.com[/email] wrote:
>
>[color=green]
>>Is this statistic just "too good to be true"? Or did I luck out when I
>>bought this box many years ago?[/color]
>
>
> No. If anything it is bordering on the too big to be good. This is
> one of the main reasons why I think that ntpd should be much more reluctant
> to apply corrections of more than about 3 to 5 ppm, once the frequency has
> been calibrated. My feeling is that it should limit the corrections to
> this range unless there is an extended period indicative of a different
> frequency. It might be reasonable to allow, say, 0.1ppm per day to
> account for long term drift.[/color]

In article <FK6dnUAg86dZhdPYnZ2dnUVZ_oWdnZ2d@comcast.com>,
Richard B. Gilbert <rgilbert88@comcast.net> wrote:
[color=blue]
> If this oscillator is "free running" it is, indeed, remarkable. If it
> is GPS disciplined it's what you should expect![/color]

The frequency correction that ntpd reports is the difference between the
free running clock frequency and the disciplined frequency, so should have
the full variability of the undisciplined clock. The figures he quotes
are unremarkable for an undisciplined PC crystal in a temperature buffered,
but not closely controlled, environment.

One would hope that two GPS disciplined oscillators would track each other's
frequencies much much better than this.