[lug] NTP with a GPS

[Gary Hodges]

I've been doing some testing, and have actually deployed at a couple 
sites, a Garmin GPS 18x to keep the time set for one computer. 
Everything was going OK, until the most recent case.  At this site I 
have the computer set up on the network, and had been keeping time set 
with NTP to another machine that uses GPS.  I don't know the physical 
location of that machine, or anything about it actually.  Due to network 
rules, that was the only time server available to me.  I hooked up a GPS 
to my machine, figuring two is better than one, but I may have invoked 
the adage "One clock is correct.  With two both are wrong."

I logged on today to see how my new set up was working, and I found that 
both time servers had an "x" preceding them when querying with  ntpq -p.

      remote           refid      st t when poll reach   delay   offset 
  jitter
==============================================================================
xSHM(0)          .GPSe.           0 l    8   16  377    0.000  -65.986 
29.232
x192.168.241.135 .GPS.            1 u  255  256  377    3.979  -11.009 
18.461

My GPS is noted by GPSe.  If I comment either one out in the ntp.conf 
file, ntp works as expected.  That is I get an "*" preceding the time 
server.  My assumption is that with a stratum 0 and 1 server available, 
but with the difference in the offsets too great, it concludes neither 
can be trusted and both are stamped with x.

I have played around with the time1 parameter in the ntp.conf file to 
bring the offsets closer together, and that seems to work.

      remote           refid      st t when poll reach   delay   offset 
  jitter
==============================================================================
*SHM(0)          .GPSe.           0 l    2   16  377    0.000  -23.996 
14.344
  SHM(1)          .GPSe.           0 l    -   16    0    0.000    0.000 
   0.001
+192.168.241.135 .GPS.            1 u  252  256  177    4.132  -27.314 
12.543

Here it has my GPS as the preferred time server, and the other is 
indicated by the + as a high quality replacement candidate.  Even though 
I seem to have made it work, going forward I'm considering commenting 
one out as there appears to be potential issues with using two time servers.

Is anyone here able to verify or refute my assumption about using two 
time servers?  Accuracy to within one second is more than sufficient for 
my needs.