On Tue, 4 Oct 2005 08:25:34 +1000, you wrote:
Hi all,
I've got a weird problem with a 9.3 box. After running for a few hours, the system clock seems to start running extremely slowly - about 1 second for every real 5 seconds.
Next time it happens I'm going to check on the hardware clock with hwclock, but I would be very surprised if it is out - I think this is a software problem. The system is fairly basic - it's a firewall using Shorewall and a mail server using Dovecot and sendmail, no GUI at all. Kernel version installed (with the non GPL stuff as well) is 2.6.11.4-21.9, SMP version (it's a hyperthreading box).
Does anyone have any ideas, or suggestions for how best to troubleshoot this one?
You don't mention your hardware details but there was a design decision made back in the stone age that causes the hardware clock to drop a bit under heavy disk IO (like a mail server with possible insufficient cache ram). The technical explanation of the problem is "Critical interrupt masking during disk IO". I don't know if the linux system clock driver compensates for that properly or not. My usual practice (since DOS days, as a matter of fact) has been to have a NIST clock update (like ntpdate) run every 12 or 24 hours. Mike- -- Mornings: Evolution in action. Only the grumpy will survive. -- Please note - Due to the intense volume of spam, we have installed site-wide spam filters at catherders.com. If email from you bounces, try non-HTML, non-encoded, non-attachments.