Re: HZ and real-time performance

Benno Senoner (benno@gardena.net)
Thu, 01 Jul 1999 20:48:19 +0200

>
> On Thu, Jul 01, 1999 at 02:24:21AM -0400, nunca wrote:
> > I've been playing with writing real-time guitar effects boxes, and I've
> > gotten the latency down to 11ms with stock linux. Would changing the HZ
> > constant to 1024 help me reduce the latency (ie, does it cause a task to
> > get scheduled more often)? Any other/better suggestions for decreasing
> > latency? Thanks in advance.
> > -Nolan
>
> I don't know what you are dpoing excatly, so I can't answer your question.
> Latencies can not only be introduced by scheduling latency, but also by
> interrupts holding spinlocks and similar.
>
> I'd suggest you try my HZ=400 patch from http://www.garloff.de/kurt/linux/
> (It not only sets the HZ value to 400, but also fixes the times reported to
> userspace.) and test it out.
>
> Regards,
> --
> Kurt Garloff <garloff@suse.de> SuSE GmbH, Nürnberg, FRG
> Linux kernel development; SCSI driver: DC390 (tmscsim/AM53C974)

I found your patch some time ago, and modified it for 1000HZ ,
(changing all 4's to 10's :-) ) , and it works quite well
but what are the implications (are these serious) of wrong times reported to userspace ?
do you know which syscalls will report wrong times ?

PS: does anyone a reasonable upper limit to HZ on PII hardware ?
will increase HZ too much lead to cache-performance drops,
du too much context switching ?

I think 2000 would not degrade much the performance on a PII class system.
(If i remember BeOS has 250usec scheduling latencies HZ=4000 ? )

regards,
Benno.

-
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/