Interesting analysis of linux kernel threading by IBM

[Robin Golden]

/* BEGIN_QUOTE
    [Sun, 23 Jan 2000 06:17:09 +0100 (CET)] Rik van Riel <riel en nl.linux.org>
wrote:
 */

On Sat, 22 Jan 2000, Brandon S. Allbery KF8NH wrote:

> Yes, but I'm one of those freaks who has 2-3 compiles running
> continuously.  And some monitoring stuff that doesn't sleep as
> much as I wish it did....  (Actually, it is mostly sleeping right
> now; I'm holding off building stuff until I get the machine
> upgraded.)

I will tell you where your overhead is. The overhead in doing
things like that is in repopulating the cache after each context
switch. When you have two gcc's running on a fairly large data
set, the cache on your processor will be useless after a context
switch (since the data of the previous process is there and not
the data of the current process).

The scheduling overhead is minute compared to the cache misses
each context swich brings you.

regards,

Rik
/* END_QUOTE */

To minimise the cache miss performance hit, while doing multiple compiles,
I modify the scheduler to reduce the frequency of scheduler calls (and
resultant context switch overhead drops). It doesn't help much on my old
486/100 though.

-----------------------------------------------------------
Please cc all replies to me.
-----------------------------------------------------------
Hi! I'm the infamous evolutionary signature virus.
Change me and put me in your .signature file.


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