Re: a joint letter on low latency and Linux

[Benno Senoner]

On Fri, 30 Jun 2000, Dan Hollis wrote:
> On Sat, 1 Jul 2000, Benno Senoner wrote:
> > It's only on audio cards where we have the privilege to set up a 64KB hw buffer
> > which means 370msecs of total buffersize which is even enough to overcome the
> > crappy 100ms latencies which Linux delivers.
> 
> I thought you had measured some 4000ms latencies...?

Yes but running the tests on IDE disks with DMA=off.

see here (look at the disk write stress test)
http://www.gardena.net/benno/linux/audio/linux-2.2.9-1000HZ-no-dma/3x256.html

note that with 4secs (!!) latencies , you can forget everything from skip-free
MP3 playing to video and other interactive stuff (even typing a text in vi
becomes a patience game)

When activating DMA the picture is MUCH smoother (although there are still
high latency peaks of around 50-100msecs)

http://www.gardena.net/benno/linux/audio/linux-2.2.9-1000HZ/3x256.html

notice the avarage case which (except the outliers) resides almost all the time
below the 3msecs.

now compare 2.2.9 (standard) to 2.4.0test2 :

http://www.gardena.net/benno/linux/audio/2.4.0-test2/3x256.html

although the 2.2.9 diagram has a bit different latency scale (15ms vs 5ms),
you can EASILY see that the average case in 2.4 really sucks compared to 2.2.9.
(I repeat  both kernels are standard without any lowlatency patch applied)

I guess the 2.4's  VM problems contribute to this bad latency behaviour.

Benno.

> 
> -Dan