[ntp:questions] Sudden change in precision and jitter

A C agcarver+ntp at acarver.net
Sat Jun 1 23:47:30 UTC 2013


I restarted ntpd early this morning (no configuration changes, just 
rolling a log and it's been running for over 12 hours now), but for some 
reason, for the first time, it recomputed the precision to -12 (used to 
be -13) and the jitter on PPS is now reaching a floor value of 
0.000244141 which is exactly double (with a slight rounding error) of 
its previous floor value 0.000122070.  I've seen this doubling once 
before prior to switching to 4.2.7 (the original value was 0.000061035 
so another exact doubling) but I've never seen the precision change.

Nothing has changed in the hardware and loopstats indicates that it's 
doing quite well for offset and dispersion, achieving the same 
dispersions as it used to prior to the restart.  I don't quite 
understand why it recalculated these values and why they are exactly 
double the old values when nothing has changed.  I've restarted many 
times in the past few months and it always returned to -13/122070 (and 
usually did so in a few hours).  The stability value in the loopstats 
file is also reasonable and currently is about the same as before the 
restart.

Here's an example from today's loopstats file before and after the restart:

[prior to restart about 0100 UTC]
56444   615.967  0.000005475 -73.823  0.000122070  0.000133 3
56444   623.967  0.000005310 -73.823  0.000122070  0.000137 3
56444   631.984  0.000005350 -73.823  0.000122070  0.000141 3
56444   639.971  0.000004189 -73.823  0.000122070  0.000139 3
56444   647.968  0.000004221 -73.823  0.000122070  0.000138 3
56444   655.985  0.000004861 -73.823  0.000122070  0.000139 3
56444   663.969  0.000004501 -73.823  0.000122070  0.000139 3

[after restart about 2300 UTC]
56444 84759.530  0.000002948 -74.118  0.000244141  0.000134 3
56444 84767.533  0.000005786 -74.118  0.000244141  0.000140 3
56444 84775.551  0.000001827 -74.118  0.000244141  0.000133 3
56444 84783.530  0.000004656 -74.118  0.000244141  0.000134 3
56444 84791.532  0.000002302 -74.118  0.000244141  0.000128 3
56444 84799.534  0.000005531 -74.118  0.000244141  0.000134 3
56444 84807.521  0.000001769 -74.118  0.000244141  0.000126 3
56444 84815.537  0.000002605 -74.118  0.000244141  0.000122 3


The PPM changes because there's a bit of a temperature swing but that 
hasn't affected the system before.  During the last three days the 
temperature swing (I record the room temperature) was very similar with 
no ill effects.

The machine's sole purpose in life is ntpd.  No major cron jobs (two log 
rotations each month), no other applications running save for ntpd and 
gpsd (to feed an SHM, PPS is by ATOM).

It's very odd that the new floor is now double the previous floor.  It 
hasn't shown signs of changing below the new value for the past six 
hours.  I'll see what happens for the next 12 hours just in case it 
regains sanity.


More information about the questions mailing list