[ntp:questions] What is the NTP recovery time from 16s step in GPS server?

unruh unruh at invalid.ca
Tue Oct 30 19:39:51 UTC 2012


On 2012-10-30, David Taylor <david-taylor at blueyonder.co.uk.invalid> wrote:
> I have discovered that on a cold start my Resolution SMT GPS receiver 
> outputs the time in GPS time, until it has downloaded enough information 
> to determine the GSP offset, when it switches to UTC.  This particular 
> receiver has no battery backup (it's an evaluation board).
>
> I'm feeding this into gpsd, from which NTP gets its coarse seconds. 
> This is no problem when I have other Internet or LAN servers telling NTP 
> what the time actually is - NTP isn't fooled and sticks with UTC.
>
> However, suppose I had /only/ the GPS receiver?  NTP has the GPS time 
> and the PPS signal for the exact second, syncs, and then a few minutes 
> in the time suddenly changes by 16 seconds.  I would hope that then 
> causes NTP to step the clock onto UTC rather than GPS time.  I realise 
> that the time for the GPS receiver to be sending UTC rather than GPS 
> time varies, but how long might it then take NTP to react to the 16 
> second change, and alter the system clock?  Both the PPS and the gpsd 
> are being polled at fixed 16 second intervals.

ntpd jumps if the time is out more than .128 sec. But it waits a while
to make sure that the time is not some weird outlier. I think it may
wait for 8 clock reads ( the clock filter buffer) so that would be 128
sec. 

>
> I'm hoping that it wouldn't take NTP too long to fix the clock.  I see 
> the situation as someone being out with some portable kit - how long 
> should they wait before assuming the time is correct?

They should perhaps get a better gps.

>
> It strikes me that a receiver with no battery - like mine - is a poor 
> choice for out portable!  I have a similar receiver with battery backup 
> on order, so I hope it's better  You know where: 251164946899.



More information about the questions mailing list