[ntp:questions] Is 24PPM an Excessive Real-Time Clock Correction?

Thomas Tornblom thomas at Hax.SE
Thu Jul 5 20:46:38 UTC 2007


"David T. Ashley" <dta at e3ft.com> writes:

> "Richard B. Gilbert" <rgilbert88 at comcast.net> wrote in message 
> news:468D4A70.9030508 at comcast.net...
>> David T. Ashley wrote:
>>>
>>> Any insight into whether 24 PPM is excessive for my server?
>>
>> Consider that a computer manufacturer typically spends something like $2 
>> US on the components for the clock!  Consider, also, that they provide no 
>> way to adjust the clock hardware.  Computers are designed to compute, not 
>> to keep time.  The original IBM PC and PC/XT did not even have a clock; if 
>> you wanted a clock it was an add-on at extra cost from a third party.
>>
>> 24 PPM is pretty good.  Anything up to a hundred or two will usually work 
>> just fine.
>
> Thanks.  Interesting.  It seems to be a really steady error, i.e. seems to 
> always be between 24 PPM and 24.5 PPM.

That is the static frequency offset of your oscillator.

>
> I think I'll write a little software to plot it at a sampling frequency of 
> once per hour.  I'm wondering if it varies by time of day, temperature, etc.

It varies widely with temperature.

I once plotted the drift over a few days on an old Sun ss20 I had on
my desk, just inside a window, and it was easy to see if it had been a
sunny day or not.

I also plotted drift vs cpu temperature on an old Sun UE-450, and from
the plots I cold deduct how the office air conditioning was
programmed, even if the cpu temp only changed by about 1° between day
and night and over the weekends.

I have tuned a kernel variable in my pc running Solaris nevada so the
drift is currently -0.229. I may be able to get it closer to 0, but
I'm alright with that value. It was around -35 when I started, and the
old motherboard was around -17.

ntp does a good job to compensate for this, so only real nerds, like
me, bother with this. :-)




More information about the questions mailing list