I'm still having this issue. Here is another update. I noticed that
the drift file for the system with the problem contained "0.000". On
most other systems this contains a positive number (and on two a
negative number). I deleted the drift file, resynch'ed the time with
"ntpdate <hostname>", restarted the NTP daemon, and waited for the
drift file to be recreated. It again contained "0.000" and the
output of "ntpq -np looked like this:
remote refid st t when poll reach delay
offset jitter
========================================================================
======
10.101.32.104 67.128.71.65 3 u 49 64 377 0.611
1871.40 987.132
*127.127.1.0 .LOCL. 10 l 48 64 377 0.000
0.000 0.001
I replaced the drift file with the contents of the file before the
upgrade to CentOS, resynch'ed the time, and restarted the NTP
daemon. But after a little while, the system is "bound" to itself
again.
BTW, the output of the cron job running ntpdate once an hour showed
that the system has a very steady drift of 14.9 seconds every hour.
Does that seem excessive? NTP should be able to handle this, right?
Alfred