Re: Analysis for Linux-2.5 fix/improve get_pid(), comparing various approaches

From: Hubertus Franke (frankeh@watson.ibm.com)
Date: Fri Aug 09 2002 - 13:18:07 EST


On Friday 09 August 2002 12:05 pm, Linus Torvalds wrote:
> On Fri, 9 Aug 2002, Hubertus Franke wrote:
> > I dragged the various algorithms into a userlevel test program to figure
> > out where the cut off points are with PID_MAX=32768. In this testprogram
> > I maintain A tasks, and for 10 rounds (delete D random tasks and
> > reallocate D tasks again) resulting in T=10*D total measured allocations.
>
> Mind re-doing that with PID_MAX=999999 or similar? The whole point of the
> current simple algorithm is that the common case (nay, done right, the
> _only_ case) is where the number of threads << PID_MAX.
>

Don't have time right now...
Simply look at the numbers for the ratio you are expected.
I would be very surprise if the relative curves would change
when moving to 132K tasks and also populate the pid space only by
let's say 25%.

Otherwise, Paul can you run this....

> That certainly used to be true with PID_MAX=32768 (not many people may
> realize it, but in 1991 the maximum number of tasks in the system was
> limited to 63, simply because of how the VM carved out the 4GB address
> space).
>
> Things have changed, but considering that some people think that 32k
> threads are a limitation already, and that the current code should work
> fine (and be pretty much optimal) with a larger PID_MAX, I really think
> it's unfair to not even benchmark that case..
>
> Linus

-- 
-- Hubertus Franke  (frankeh@watson.ibm.com)
-
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 : Thu Aug 15 2002 - 22:00:20 EST