Re: [iiwusynth-devel] Latency measurement results

[Peter Hanappe]

On Fri, 2002-02-22 at 19:22, M. Nentwig wrote:
> Hello again,
> 
> as I mentioned before, I'm really interested in getting the latency down
> as much as possible (and keep it as constant as possible), because I'm
> keen on using iiwusynth 'realtime' in a band.

Are you using a low-latency patches kernel. I have installed it but
haven't done any real testing on it yet. Josh Green (the third person on
the iiwusynth mailing list ;) has done more testing. (BTW have you
checked Josh's Smurf Soundfont Editor?)

> Therefore I went to the lab and did a couple of measurements.
> What I have measured is the time between the audio output of my MIDI
> controller and the audio output of iiwusynth - that is the latency. I
> repeated this for different fragment settings: for 16 fragments (the
> only setting that works without tricks on my own PC) and 3 fragments
> (soundcard set to 16 fragments, and a loop delays the synth until there
> are 13 or more buffers empty).

Can you send the code of that loop. I'm interested.

> The problem with 16 frags is, that the latency varies a lot.
> For 3 fragments, it is stable and short (8 ms of latency is equivalent
> with the monitor speaker being two-and-a-half meters away, that's
> tolerable).
> 
> A typical printout of the oscilloscope:
> http://millisampler.sourceforge.net/3frag.gif (see also 1frag.gif and
> 2frag.gif)
> 
> All results summed up in a plot:
> http://millisampler.sourceforge.net/plot.gif

That's nice. (Love the scanned notes!)

> It is obvious, that using 16 fragments without additional 'tricks'
> results in quite sloppy timing. But I have a feeling, that the
> performance with a proper '3-fragment-capable' sound card can compete
> with any commercial product. 

I suppose that with a professional sound card and a LL-patched kernel it
should be possible to have latencies in the order of 3 msec. I'm not too
worried about that. What worries me more is that the synthesizer still
takes up too much CPU. 

Can anyone recommend some profiling tools for Linux?


> For example, in the keyboard review of
> Native instrument's latest B4 softsynth, 'Keyboards' magazine wrote,
> that the latency is 'tolerable'.
> Now concerning iiwusynth, 8 +/- 2 ms is better than 'tolerable', if you
> ask me :-)

You friends might start arguing you're playing too fast ;)

Peter

> Regards
> 
> Markus
> 
> 
> _______________________________________________
> iiwusynth-devel mailing list
> address@hidden
> http://mail.freesoftware.fsf.org/mailman/listinfo/iiwusynth-devel
>