[ntp:questions] Clock jumps when refclock used

A C agcarver+ntp at acarver.net
Sat Apr 14 05:58:24 UTC 2012


I'm still having some issues with the system clock when a refclock is 
being used but I can't seem to really figure out the true cause.

The same IPX I have been using seemed to be working and then one morning 
it had a clock offset of 22 seconds.  Nothing out of the ordinary 
appeared in the log files, it just randomly started to step the clock 
0.2 to 0.3 seconds at a time.  The only thing of interest were a few 
sys_fuzz messages prior to the event.  The runaway clock was actually 
with an unmodified version of ntpd 4.2.6p270 (I had been tinkering with 
a modification but I went back to the original source).

In order to figure out if the GPS was having problems, I noselected both 
of the refclocks (SHM 28 and ATOM 22) and let the system use only 
Internet servers.  I also disabled kernel discipline and then restarted 
everything.

The system settled on the Internet servers with a nominal offset of a 
few ms.  I recorded the PPS offset in clockstats to see how it wandered 
over the course of a couple days.  The result of that test showed the 
PPS offset never really drifted beyond +/-5 ms on average with some 
impulses showing up that were about 15ms in magnitude.  I have a plot if 
anyone wants to see it.  Most of the drift is temperature driven as the 
system gets blasted by the furnace during the cold evenings (I haven't 
had an opportunity to close it up in a closet).

So I'm really confused at this point.  I don't think the short impulses 
of a few tens of ms would throw off the whole system like that but I do 
notice a lot more sys_fuzz messages when using the PPS signal.


More information about the questions mailing list