[ntp:questions] About use of PPS in NTP sync

David Woolley david at ex.djwhome.demon.invalid
Mon Jun 16 09:18:51 UTC 2014


On 16/06/14 07:08, jthulard at gmail.com wrote:
> Hi,
> the PPS signal comes from one GPS clock.
>
> At the cold start the drift could be several seconds, so we plan to perform a ntpdate before launching the ntp client.
>
> The sync accuracy we want to achieve for this particular system is below 3ms offset.

So you want less than 1ms RMS error in the time, but you will accept a 
significantly larger systematic error.

>
> After the ntpdate we assume that the offset can be several 100 ms, and we would like to optimise the time to get close to the 3 ms offset. Therefore we wonder if the use of PPS signal on the client could speed up the convergence time in this case.

ntpdate should be better than 10ms, and if 3ms RMS offsets are 
achievable, ought to be mostly below 5ms.  If it is much more than 
100ms, it is pointless as ntpd will step the time itself it if the 
offset consistently exceeds 128ms.

ntpdate tries to fully correct the measured offset, but can't make a 
frequency correction, or allow for the variability in the offset.

The subsequent difficulty in getting a fast convergence may be because 
ntpd wanders off because its remembered frequency correction is no 
longer valid and it has to build up a phase error in order to detect that.



More information about the questions mailing list