Subject: Re: [linux-audio-dev] Re: Low Latency Kernel Combos
From: Benno Senoner (sbenno_AT_gardena.net)
Date: Thu Jun 21 2001 - 12:40:55 EEST
Notice that the output of latencytest is stereo thus a framentsize of 128byte
means 32 samples/fragment (32 * 16bit * 2 channels = 128bytes)
each fragment has a latency of 0.7msec therefore 3 x 0.7 = 2.1msec latency.
Josh, although your tests look very good, it seems that the soundcard uses
only 2 fragments rather than 3. (that would mean that the effective latency
would be 1.4msec :-) )
The strange thing on your tests is that in the 3 x 256 test you posted,
3 fragments get allocated. (see the red line).
If the test with fragsize=256 used 2 fragments too, then I'd say that your
card is setting wrong fragsizes all the time but in your case it seems that
it's ok for 3x256 but not for 3x128.
Notice that latencytest was developed testing it with a 2.2.x kernel and
as we know you need to pass to the (2.2.x) OSS/Free driver
desided_fragments - 1 as the num_fragments parameter.
I'm not sure if that got fixed in the 2.4.x kernels ?
I recall a discussion where ALSA produced somewhat different results.
(eg you needed to pass the real # of fragments and not numfrags-1).
I'm a bit confused here, any OSS/Free experts (2.2.x vs 2.4.x differences?),
ALSA experts out here ?
Most "consumer" apps to not care much about latencies to a difference of
one fragment remains unnoticed in most cased, but with lowlatency apps
it is quite important to know the real latency, especially if you need to sync
with other apps/audio sources.
Thoughts ?
Josh: regarding the latencytest database:
I was thinking about the following: use the phpnuke engine, make a topic
"low latency" and then users can simply submit their results.
That way people interested in low latency simply click to the lowlatency icon
and see all latest news on the matter (new tests like yours, new patches, rpms
etc).
Would this be ok folks ?
PS: I made a small cosmetic change on
http://www.linuxaudiodev.org
At the beginning it got redirected to http://www.linuxdj.com/audio/lad/
Now the redirected URL looks is the following:
http://www2.linuxaudiodev.org/audio/lad/
(basically linuxdj.com aliased to www2.linuxaudiodev.org)
That way the linuxaudiodev.org part remains, helping us to build
a strong branding for our linuxaudiodev.org domain :-) )
It's only a temporary solution, but LinuxTag will be a good occasion to discuss
about how the LAD websites should look like in future.
cheers,
Benno.
On Thu, 21 Jun 2001, Josh Green wrote:
> >
>
> Nope. Haven't tried that :) I guess the 3x128 means there are 3
> fragments of 128 bytes (64 samples for 16 bit mono data)? So this would
> mean that total delay of sound buffer at 44.1khz would be (3 * 64) /
> 44100 = 4.35 ms. Whereas the maximum allowed latency to insure no
> underruns would be 64 / 44100 = 1.45 ms. Please correct me if I'm wrong.
> It would seem like having 2 fragments versus 3 would be just the same
> for the hardware, and you would get the lower audio delay. I'd like to
> try it with 2 fragments to see what happens.
>
> Anybody writing a web form driven database for latencytest results?
> Lates..
> Josh Green
-- http://www.linuxaudiodev.org The Home of Linux Audio Development
This archive was generated by hypermail 2b28 : Thu Jun 21 2001 - 11:57:30 EEST