Re: [patch] lockless, scalable get_pid(), for_each_process() elimination, 2.5.35-BK

From: Ingo Molnar (mingo@elte.hu)
Date: Wed Sep 18 2002 - 12:24:40 EST


On Wed, 18 Sep 2002, Linus Torvalds wrote:

> And guys, if this is a performance problem for some extreme site, the
> fix is truly trivial:
>
> echo $((0x7fffffff)) > /proc/sys/max_pid
>
> and you're done.

it is a problem still. We can create/destroy 2 billion threads:

   venus:~> ./p3 -s 2000000 -t 10 -r 0 -T --sync-join
   Runtime: 19.889182138 seconds

in roughly 5 hours, on bog-standard 2-CPU x86 hardware. Or in 1.25 hours
on an 8-way box. And then we are back to step #1: trying to pass over
already allocated PIDs by destroying the contents of the L1 and L2 cache
once for each allocated PID passed. Sure, with 2 billion PIDs space that
averages out, but it's an algorithm with a very nasty worst-case behavior,
which is not so hard to trigger.

Plus the not-so-unlikely case of having to pass a couple of tens of
thousands of allocated PIDs, in a couple of minutes, with all user input
blocked, and interrupts disabled, website silent. Obviously sites will not
create threads at such a high rate as demonstrated above, so it will
'only' happen once every couple of days, or every couple of weeks.

        Ingo

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Mon Sep 23 2002 - 22:00:23 EST