[mdlug] Clocks

[list1c30fe42 at bellsouth.net]

Joe,

> My system time keeps losing 3 seconds every minute ever since I 
> installed Fedora 10 a couple of weeks ago. My hardware clock works just 
> fine. I have been trying to fix this, so far without luck.

One obvious possibility, is ntpd(8) running with intermittent internet 
connection AND an automatic script running hwclock --set .



A friend's computer (a 1998 Dell) began manifesting the same problem
shortly after I upgraded it to SuSE-10 .  Visiting only twice(?) a year,
and with ssh access only a few times a year, it took two years to
determine the problem.  (It took two years until I had more than two
days in a row to examine the problem.)

The problem was that the updated version of hwclock(8) or
(corresponding kernel components) that accompanied SuSE-10 required
a more accurate hardware clock than SuSE-8.  Upon using
hwclock --systohc or hwclock --set, /etc/adjtime was set
to an unreasonable value (presumably due to rollover limit in its
internal calculations).

The workaround for my friend's computer was to disable setting
/etc/adjtime by hwclock(8) and then set /etc/adjtime manually
using measurements made during several weekend visits.
This has resulted in reducing the error to less than ~2min/yr,
which was sufficient.

You can use --noadjfile, to disable hwclock used from the command
line, setting /etc/adjtime.  SuSE-10> yast2 sysconfig  had a setting
to disable the adjustment by normal SuSE self-management.  (I cannot
find the setting on my current laptop, though it may have been
HWCLOCKSET(?))

adjtimex(8) has further details.  Please let me know if you
find a cure, though I suspect this is a limit of newer distros
with regard to older hardware.

Whether or not the hardware clock has become less accurate with more
than a decade use is uncertain.



Hopefully helpful,
--
Bob