Subject: Re: [linux-audio-dev] those latency numbers
From: Benno Senoner (sbenno_AT_gardena.net)
Date: Sat Mar 23 2002 - 04:07:02 EET
Hi all, got a some spare cycles to read LAD again :-)
I saw the linuxdevices.com article too, indeed quite nice results especially
given that the testing period was 15 hours (1.5msec max latency with the
combined preemt+lowlat patch).
BTW: what does the performance of jack in "plugin mode" vs "threads" mode
look like. Is the kernel performing well (in terms of latency) even with
dozen of lowlat threads running ?
Paul: kudos for your ALSA docs !
PS: I noticed that 2.4 kernels do have an rme96xx.o module for the
Hammerfall. Do consumer OSS apps see the device as a regular /dev/dsp stereo
device ? (are the spdif ins/outs the default ports ? (did non try the OSS
module))
PS2: I'm working on a video analysis project at the university (realtime
motion tracking) which runs under linux so the final version of the sw will
probably run on a lowlat kernel. I'll report about results/experiences
when the project will be completed.
cheers,
Benno
On Friday 22 March 2002 02:20 pm, you wrote:
> This from the *excellent* paper at linuxdevices on latency
> measurement. BeOS, eat your (cold dead) heart out:
>
> 2.4.17 with Andrew Morton's Low Latency patch:
>
> maximum latency: 1.3ms
> Mean: 0.0542797399957571ms
> Standard Deviation: 0.025220506601371
> 96.66250% of samples < 0.1ms
> 99.21158% of samples < 0.2ms
> 99.99592% of samples < 0.5ms
> 99.99984% of samples < 0.7ms
> 99.99992% of samples < 1ms
> 100.00000% of samples < 1.4m
>
> Yowza! I mean, I knew it was good - I didn't know it was *that* good!
>
> Even better: this is *with* IDE drives under some level stress (but no
> X Windows).
>
> --p
This archive was generated by hypermail 2b28 : Sat Mar 23 2002 - 03:55:27 EET