On Mon, Apr 23, 2007 at 08:26:25AM +0300, shimi wrote: > This is a server box. Windows never touched it since it went out of its' > package... (nor do I see a reason for it to do that :)). I am also unsure if > Windows would "see" this problem, because I think their timekeeping is > different (live sync with the RTC if I am not wrong). The fact is that if I > used 'hwclock' WITH the SMP running, I always got the RIGHT time. So the > board's RTC is going well. It just appears that every CPU runs its own > timekeeping (and I might be saying something very silly here, but I never did > read the relevant kernel source code, so I can only guess). Ticks, Jiffies, > CPUs synchronization, alternating clock speeds (Intel SpeedStep, Frequency > Scaling) - can all cause this - and what I am seeing might just be that > "date" was assigned to a different CPU every time (perhaps I should have > forced an assignment to each CPU individually and try?)
This is a good idea. Use 'taskset' (apt-get install schedutils) and try that. Also, as a temporary solution, if you say that at boot the drift is small, you can run in cron (assuming it will work :-) ) a small loop over the cpus that once every some short time will hwclock --hctosys, or ntpdate etc. -- Didi ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
