[ntp:questions] ntp drops gps/pps - lets time drift
snews at lordynet.org
Thu Jun 9 21:37:49 UTC 2011
David Lord wrote:
> M Z wrote:
>> I am trying to get an independently clocked SuSE 11.2 machine based on
>> just GPS/PPS. I am using
>> gpsd 2.95 with garmin 17 serial port but also have tried this with
>> globalsat mr350p. gpsd/cgps shows 3D
>> signal lock. ntpq Ver. 4.2.6p2
>> *** HERE IS THE PROBLEM/Question ***
>> When I have an outside source, like 0.us.pool.ntp.org or
>> tick.usask.ca, ntp uses
>> the PPS signal perfectly. When I drop the outside source or substitute
>> with 127.127.0.1,
>> NTP x's out my PPS and then GPS (when reach is 177) even though it is
>> a good signal.
>> How can I force ntp to use the PPS signal or at least not drop it?
>> clocksource is tsc which seems to work a little better (also tried
>> hpet, same problem)
>> I also tried it with none of the 'restrict' clauses and without
>> clip from ntp.conf:
>> #Must use disable for SuSe since the internal micro-nano timekeeping
>> #conflicts with ntp and causes excess jitter ( > .010 microsec ).
>> disable kernel
>> server 127.127.28.0 minpoll 4
>> fudge 127.127.28.0 time1 0.290 refid GPS stratum 3 #garmin
>> server 127.127.28.1 minpoll 4 prefer
>> fudge 127.127.28.1 refid PPS
> Try with prefer for the refclock rather than the PPS
> server 127.127.28.0 minpoll 4 prefer
> and don't use 127.127.0.1
Is there any reason for using gpsd rather than refclock
drivers types 20 & 22?
A few minutes ago I restarted ntpd with faster poll rate
than I've been using for past couple of months.
"server 127.127.20.2 mode 18 prefer maxpoll 4"
"fudge 127.127.20.2 time2 0.350 refid GPSb"
"server 127.127.22.2 maxpoll 4"
"fudge 127.127.22.2 flag2 0 flag3 1 refid PPSb"
After a few minutes ntpd has converged
remote refid st reach delay offset jitter
*GPS_NMEA(2) .GPSb. 0 377 0.000 -39.348 14.638
oPPS(2) .PPSb. 0 377 0.000 -0.004 0.004
This for a Sure-Electronics GPS
Peer_summary gives mean=0.000 rms=0.003 max=0.030 and I hope
the faster poll will get rid of or at least reduce the 30us
blip (from load of nightly cron job).
More information about the questions