Replying again to myself :-)
On Sunday 22 April 2007 00:58, shimi wrote:
> Replying to myself because I found the CAUSE, but not the REASON.
>
> toast ~ # date
> Sun Apr 22 01:25:13 IDT 2007
> toast ~ # date
> Sun Apr 22 00:52:25 IDT 2007
> toast ~ # date
> Sun Apr 22 01:25:14 IDT 2007
> toast ~ # date
> Sun Apr 22 00:52:26 IDT 2007
> toast ~ # date
> Sun Apr 22 01:25:15 IDT 2007
> toast ~ # date
> Sun Apr 22 00:52:27 IDT 2007
> toast ~ # date
> Sun Apr 22 01:25:17 IDT 2007
> toast ~ # date
> Sun Apr 22 00:52:29 IDT 2007
> toast ~ # date
> Sun Apr 22 01:25:18 IDT 2007
> toast ~ # date
> Sun Apr 22 00:52:32 IDT 2007
> toast ~ # date
> Sun Apr 22 01:25:21 IDT 2007
>
> Appears time is alternating 32 minutes back and forth all the time (perhaps
> every second or so?). The kernel internal timekeeping seems to work. If I
> issue a "sleep 5" at the shell, it returns after 5 seconds. However if I
> TIME that command I am receiving very weird results, like:
>
> # time sleep 5
>
> real -.m+*.253s
> user 0m0.000s
> sys 0m0.000s
>
> # time sleep 5
>
> real 33m7.761s
> user 0m0.000s
> sys 0m0.000s
>
> Obviously 'time' is looking at the same alternating value that 'date' is.
> Same goes for 'ps' - I look at the process start time, and every time I
> invoke 'ps', all the values go 32 minutes back and forth in time.
>
> I stopped ntpd and it is still happening so this is not the cause.
>
> Still looking for ideas :)
With the help of a friend's Google searching abilities, my first idea of
multiprocessors become even more probable for the reason for this behavior. I
tried recompiling the kernel, changing some RTC related options - to no
avail.
Also, I have noticed, that when booting, the drift is "small" - like - 20
seconds or so. As time passes, it gets bigger and bigger.
Then, I tried booting the machine with the kernel option maxcpus=1. Problem
Gone.
OK, so now I got one CPU less and a greatly working system. I could stay here,
but I'ld be happy to utilize the other core as well, so I am _still_ looking
for hints (especially from the kernel/SMP experts :-))
Thanks,
-- Shimi
=================================================================
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]
