[ntp:questions] Frequency Offset

unruh unruh at invalid.ca
Fri Feb 24 17:47:31 UTC 2012

On 2012-02-24, Dave Hart <hart at ntp.org> wrote:
> On Fri, Feb 24, 2012 at 02:20, Alby VA <albyva at empire.org> wrote:
>> ?Thanks. I needed that NTP 101 lesson. So is it a stretch to try and
>> obtain a quality motherboard to improve the frequency offset?
> The gross error of 25 PPM is on the good side.  It's not uncommon to
> see several times that much.  But the gross error isn't the
> interesting part -- ntpd will handle it pretty well.  The
> temperature-induced variation at around 1PPM per degree C is the major
> uncompensated error.  Finding the appropriate crystal and insulating
> it to slow its temperature change as its environment changes is
> probably the low-hanging fruit in improving your clock accuracy.  Next
> steps are more invasive, such as moving the crystal out of the
> interior to the presumably-less-variable temperatures outside the
> case, replacing it with a TCXO or OCXO.

Or running ntpd with temperature compensation built in, assuming you
have some accessible onboard temperature sensors.
chrony does have the possibility built in, but I am not sure how well
the code has ever been tested (I have not).

The tests run on nptd gave something like an order of magnitude better
behaviour as I recall than for uncompensated ntpd. 

> I recall Dr. Mills describing DEC Alpha machines has having
> particularly good oscillators because designers were willing to spend
> wee bit more than the minimum, but I don't remember what approach was
> taken to improve on the common quartz crystal approach.
> Cheers,
> Dave Hart

More information about the questions mailing list