Re: Is the Optiplex hwclock junk?

John Peacock <[email protected]> writes:

> Andrew A. Raines wrote:
>
>> So is the only recourse in my situation to cron (er, runwhen)
>> clockadd and manually force the clock into submission?
>
> I have a couple of cheap machines which have this same
> problem.  I found that if I just reduced the maximum time
> between adjustments to a week (604800 sec), they stay
> perfectly synced:

[...]

Thanks.  I've set it to 604800.  I guess if I still see problems
I could reduce it to a couple hours.

I'm still a little confused though.  Until clockspeed was left
to its own devices for a couple weeks (before that, remember, it
had been getting measurements at most every few hours or other
day), the clock didn't seem skewed.  I even set my wristwatch to
it and found my watch to be losing a second every couple weeks.

Notice now that the clock can't even be trusted for a few hours:

2003-12-02 21:41:42.473442500 Sleeping 38400 seconds ...
2003-12-03 08:22:01.741170500 before: 2003-12-03 08:22:01.741018000000000000
2003-12-03 08:22:01.772756500 after:  2003-12-03 08:23:37.767401673950195312
2003-12-03 08:22:01.772874500 before: 2003-12-03 08:22:01.772682000000000000
2003-12-03 08:22:01.772891500 after:  2003-12-03 08:22:01.772682000418202732
2003-12-03 08:22:01.773345500 ----------------------
2003-12-03 08:22:01.786059500 Sleeping 76800 seconds ...

Does the system clock develop a memory which can only be reset
when the machine reboots, or is the actual hardware clock going
bad?  If it's the former, do I need to reboot before John's
suggestion will work?

-Drew