[ntp:questions] SHM refclock not being selected

John Ackermann N8UR jra at febo.com
Sat Mar 12 13:21:44 UTC 2005


In order to use a PPS source with Linux 2.6.x, I've been running David 
Schwartz' shm_linux_clock driver with an HP3801A GPSDO as the external 
refclock.  I am getting very good reported results and the SHM clock 
consistently has the lowest jitter of any of my servers, but for some 
reason ntpd doesn't select it as the source.  Here's the ntpq -p output 
after a week of operation:

     remote           refid      st t when poll reach   delay   offset  
jitter
==============================================================================
+GPS_HP(0)       .GPS.            0 l   18   64  377    0.000   -2.045   
3.867
+SHM(0)          .PPS.            0 l   54   64  377    0.000    0.882   
0.033
*tock.febo.com   .PPS.            1 u  411 1024  377   10.362    5.926   
0.192
+time-B.timefreq .ACTS.           1 u  411 1024  377   74.705   -5.855   
0.759
-ntp1.usno.navy. .USNO.           1 u  451 1024  175   79.246   23.077   
0.253
-ntp2.usno.navy. .USNO.           1 u  452 1024  377   79.070   26.126   
4.196
 
Selecting "tock" as the survivor isn't a bad thing, as it is a FreeBSD 
machine sitting on the local subnet, and has two stratum 0 refclocks, 
but I would think that since the SHM clock *always* has lower delay and 
jitter (as well as sane offset values), it should end up being the survivor.

Should I be using the "prefer" keyword to force selection of the SHM 
clock?  Or is there some other trick I should be playing with this 
configuration?

By the way -- I was getting absolutely horrendous, unusable time from 
this machine until I recompiled the 2.6.9 kernel with HZ=100, and 
selected "clock=pit" at boot time (not sure which of the two changes is 
making the most significant impact; I wasn't in a position to experiment 
through multiple reboots).

John




More information about the questions mailing list