On Tue, 2005-05-24 at 18:26, James D. Parra wrote:
I have a problem with clock drift on an opteron machine running suse 9.2. I think I've narrowed the problem down to the kernel clock. I stopped NTP so that isn't confusing the issue. I manually set the system clock and the hardware clock to match another NTP-synchronized clock that is working properly and left it for an hour. The hardware clock is still showing the correct time, but the system clock has gained 52 minutes in an hour.
I have had a similar problem and had to resort to setting a cron job restarting xntp every 30 minutes. This appears to be a Suse problem (I have a similar hardware setup as you, 4 dual Opterons running Suse 9.1). The Mandrake servers I have make use of the drift files frequently. Oddly, I have noticed that on the Suse boxes the ntp daemon dies for no apparent reason; # rcxntpd status Checking for network time protocol daemon (NTPD): dead
~James
Hi James, Is every 30 minutes not overdone? I have ntpdate just once every 24 hours (cron) and even that is too often, considering the drift: /var/log/messages.4:Apr 30 01:00:04 fw2 ntpdate[24607]: step time server 193.79.237.14 offset -0.005447 sec If you need to update it that often, one might have a hw-problem... Hans