Re: [linux-audio-dev] latest low latency patch

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: Re: [linux-audio-dev] latest low latency patch
From: Benno Senoner (sbenno_AT_gardena.net)
Date: Fri May 05 2000 - 16:41:34 EEST


On Fri, 05 May 2000, Maarten de Boer wrote:
> Hi,
>
> Can somebody tell me where to find the latest
> low latency patch (and if it will work ok on SMP)?
> Benno's page mentions 2.2.13, but in the kernel
> mailing list there is mention of 2.2.14. And how
> about 2.2.15?
>
> Maarten

http://people.redhat.com/mingo/lowlatency-patches/

I think the 2.2.15 version is for a pre2.2.15 kernel, and I don't
know if it applies cleanly.

Plus I am not sure if th SMP issues are solved.
Can anyone check this out ? (if the patch applies)

Anyway there are now several firms interested in this topic,
because the embedded applications benefit too from this
enhancement.
Therefore I suspect that it will not take too long until
we have such stuff in the official kernel.

(at least if we are able to convince Linus, since he fears
a 0.001% raw performance decrease when a few reschedules
more are added to the kernel, but that is unfounded, since
my tests using lmbench
( http://www.bitmover.com/lmbench/ ) and bonnie
has indicated that there is no performance loss
compared to an unpatched kernel.

The only performance loss I saw, was the less
disk write throughput when using latencytest with 80%
CPU usage (that means my UP box had only 20% of CPU left for
disk writing).
And it's easy to figure out why this happens:
the unpatched kernel simply causes big peaks during the test,
where the kernel has "plenty" (100msec) of time to write buffers,
while the lowlatency kernel doesn't jitter more than 1msec.

Even a BeOs developer said that BeOS may face the same problems
simply because the CPU is only one, and you have to share it among
the disk writing and the realtime cpu-waster which runs with higher priority.

This was maybe what confused Linus and made it sceptical.
(a stupid slashdotter even referenced this kernel-post
(Linus' reply to my disk IO tests) in a posting a few
weeks ago on the the linux and audio article)

Benno.


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : Fri May 05 2000 - 18:00:11 EEST