I have 3 virtual SLES 10 32-bit servers running on one ESxi 4.0 host.
Further there is one SLES 64-bit server and one CENTOS 64-bit server
running on the same host, which are perfectly syncing time, with very
little differences (below 1 ms), but the 32-bit servers are really

If I do not restart them earlier all 3 servers show the same behaviour.
After boot ntpd syncs fine and needs app. 2 hours to come to a smooth
operation without time resets. It then runs fine for app. 5 days -
after 5 days there are 2 to 4 time resets per day necessary to keep
time in sync. After app. 9 days after bootup the frequency error starts
to exceed 500ppm and the time resets are increasing in frequency and
time offset. One day later the time is not reset for app 8 hours and
during these 8 hours time gets app 600 to 750 seconds ahead of real
time. After this 8 hours time is reset back to the correct time. Then
the frequency error goes back to normal within 1 hour and after that
all is the same as after a reboot - that means the same cycle starts
over and over till a reboot occurs.

I have used all suggestions for timekeeping from Vmware without any
better results. What I'd like to know is: Is there any possibility for
restarting ntpd and related services, which have the same effect as a
reboot? If I stop ntpd and even if I delete
/var/lib/ntp/drift/ntp.drift any start of the ntpd gets it back into
the same behaviour, which is only dependent on the time, when the last
bootup or the last reset for 600 - 750 secs occured; so a restart of
ntpd is useless to cure this problem. I suspect, that's a kernel
problem as ntpd directly manages kernel time parameters.

It would be nice to not have to reboot the OES 32-bit servers every 8-9

W. Prindl