Re: PATCH 2.3.26: kmalloc GFP_ZERO

Mikulas Patocka (mikulas@artax.karlin.mff.cuni.cz)
Tue, 9 Nov 1999 22:06:57 +0100 (CET)

> > From: Jes Sorensen <Jes.Sorensen@cern.ch>
> > Date: 09 Nov 1999 18:00:49 +0100
> >
> > now did you in fact see a 20% speedup on kernel compiles? I for
> > some reason strongly doubt that.
> >
> >I don't, it makes perfect sense since gcc is heavy on anonymous page
> >usage.
>
> Are you still talking about a cache of zeroed pages? Looks a bad idea to
> me.
>
> It can only improve latency after some idle time. If your CPU are under
> 100% of load for a long time you'll have to spend the memset(0) time
> anyway but in another place and with an additional scheduler cost. Also
> the idle task will be less fast in rescheduling itself.

If you have 100% CPU load, you don't do background zeroing and you
fallback to old zero-after-alloc.

Background zeroing would help in applications that don't consume 100% CPU
all the time but need to be fast (interactive programs, internet or
database servers etc.).
It would also help a bit when system is under i/o load.

I don't see a big problem in rescheduling. Idle task can clear only a part
of page and check for the need to reschedule.

Processor cache can be disabled (I hope).

The whole idea is that if CPU has nothing to do, it could do something
useful. Why not?

Mikulas Patocka

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