One of the backward jumps was so small that cron seems to have had two copies
of my daily cleanup/backup script running virtually simultaneously - with
the result that a complete day's worth of system logs vanished, so I
can't say anything about the events leading up to this.
I do know that my backup scripts were running twice in parallel - although
only the first one actually wrote anything to tape ...
Now, I am running xntp3.5f - and there are no unusual messages from xntp
from the logs that I do have. 3.5f was the current version a few months
ago, the last time I noticed an alert to upgrade it. Did I miss anything ?
Now there is one other piece of information that may be relevant. I
run xntp to keep gear's clock in sync, but there are two clients that
reset their clocks from gear at the end of their daily scripts, which
also run at 2:30 ... The clients use 'netdate' to correct their clocks.
I thought netdate just used the port 13 'daytime' service, but perhaps
I'm mistaken. Is it possible that these queries are disrupting the
clock on gear ?
The two machines in question logged their time jumps: +36 seconds in one
case and -39 seconds in the other. Their clocks are now (6 hours later)
19 seconds apart, and gear is dead in the middle between them.
Please remember, though, that gear does not make any calls to netdate
or anything else that could deliberately jump its clock. And that the
clock on gear was sufficiently unstable that cron ran the 2:30 jobs
three times. There is no evidence of any continuing instability problems
with gear's clock. And, this configuration has been running for months
(maybe even years) with no similar problems before today.
This is a serious problem. Does anyone have any clues ?
--------------------------------------------------------------------------
Grant R. Guenther grant@torque.net
--------------------------------------------------------------------------