[ntp:questions] best jitter on qnx always = 0.977

Николай Орехов nowhere at mail.ru
Wed May 11 04:46:44 UTC 2011


My BSP gives right nanoseconds from ClockCycles since last 1ms interrupt. So
It's far more precise than 1ms.
How can I convince ntpd that I have higher presision ? There's a define in
adjtime.c:
#define ADJUST_PERIOD       0.97  /* [s] */

May be I should lower it? Or is there another way?

2011/5/11 Dave Hart <hart at ntp.org>

> 2011/5/10 Николай Орехов <nowhere at mail.ru>:
> > Well, I'm using PPS driver which measures ticks with presision more than
> one
> > millisecond using ClockCycles and system interrupt.
> > Even with ref clock jitter less than 1ms I got 0.977 jitter.
> > I'm agree that system jitter is closely connected to presision of system
> > clock. But reference clock jitter may be even lower.
>
> If your PPS timestamps in fact are more precise than ~1 msec, the
> excess precision is wasted.  ntpd uses system timestamps extensively,
> and has no provision for using higher-precision PPS timestamps to
> discipline a lower-precision system clock any better than if the PPS
> precision were the same as the system clock precision.
>
> > Can i somehow convince ntpd that this refclock jitter can be less than
> 1ms?
> > Will it influence sync presision? For now i get offsets ranging between
> > +-0.5ms
>
> I don't believe so.  The behavior you see is as good as I would expect
> given the clock precision.
>
> Cheers,
> Dave Hart
>
>



More information about the questions mailing list