[ntp:questions] how do I lock in average frequency correction

unruh unruh at invalid.ca
Tue Feb 14 00:53:09 UTC 2012


On 2012-02-13, Dave Hart <davehart at gmail.com> wrote:
> On Sun, Feb 12, 2012 at 17:36, unruh <unruh at invalid.ca> wrote:
>> Uh, that is what ntp tries to do, and is why it shifts up its poll
>> interval as time goes along ( and since it throws away 80% of
>> measurments, the effective poll interval is about 8 times as long as the
>> actual poll interval).
>
> s/effective poll interval/time constant/
>
> That up to 8 x poll time constant is accounted for in the design.
> There is no bug here.  In fact, while you harp endlessly on this

I never claimed there was a bug. It is just incredibly profligate of
precious measurements, but it exactly as designed. 

> point, it's due to the IMHO brilliant minimum delay clock filter which

I guess brilliance is in the eye of the beholder. I consider it an
incredibly crude kludge, but I guess to each his own. 

> very effectively sifts signal out of noise.  ntpd is not going to rip
> out the clock filter so it can whip around tracking noise as quickly
> as possible.  Please find somewhere else to relieve your bowels.

Yes, I know, ntp has sacrificed speed of response for other concerns.
The question is whether or not that sacrifice was the minimum needed and
what the cost is.  You and I clearly differ on that evaluation.


>
> Cheers,
> Dave Hart



More information about the questions mailing list