[ntp:questions] Slow convergence of NTP with GPS/PPS
unruh-spam at physics.ubc.ca
Tue Oct 28 19:16:38 UTC 2008
davida at pipstechnology.co.uk (David Ashmore) writes:
>An Update from what we have decided to do.
>As NTP is designed to achieve accuracy slowly and conservatively. We
>have decided not to go down the NTP route.
>We have decided to use the PPS signal from the GPS and force the system
>time into being correct.
I have no idea what this means. doyou mean that you reset the clock each
time it receives a PPS pulse? Ie, stepping the clock each time? That can be
dangerous. a) backward steps are possible confusing the filesystem.
b) GPS PPPS lines can get affected by noise. My gps PPS line every once in
a while will put out up to 10 interrupts in a second-- presumably some
weird noise on the system. Your system would be resetting the clock like
mad, creating a totally inaccurate time during that second. And if you
assumed that the time advanced one second on each interrupt you would be
out 10 seconds.
Ie, you have to sanity check the PPS to make sure that what you think is
correct, really IS correct.
>It is a great shame the NTP doesn't achieve a high accuracy quickly, as
>the method we have adopted has a low level of accuracy compared to NTP
>I am posting this mainly to help Nicola, with a solution.
>Many Thanks for all your help
More information about the questions