Re: [Jack-Devel] Effects of increasing "-p, --period"

On 04/19/2014 10:19 PM, Holger Marzen wrote:
> Hi all,
> 
> I have a Qtractor project with many tracks and many instances of CPU
> hogs like Klangfalter (convolver reverb) and CALF compressor.
> To decrease realtime load and avoid xruns I had to increase the period
> size to 512.
> 
> I know that this will increase end-to-end latency, but that would'n be a
> problem in my case. What I don't have understood yet: does this
> intruduce some more latency to
> - plugins (LV2, LADSPA) running inside a DAW like Ardour, Qtractor or
>   Rosegarden?
> - jack clients like plugin hosts or other sound processing stuff?
> 
> The man page says "Specify the number of frames between JACK process()
> calls." So if plugins and jack clients only get their data as complete
> chunks after a process() call they would suffer as well from an
> increased latency.

Hi Holger,

in short: no. there is no additional latency. but, yes plugins process
the audio in larger chunks.


That assumption is wrong. jack clients are called *in* (or during) the
process call.

-=-=-
0) read N samples from the soundcard

1) call all jack-clients (incl plugins) to process N samples

2a) write the processed N sample to the soundcard's output
2b) read next N samples from the soundcard's input (same as step 0)

3) goto (1)
-=-=-


(0) never really happens but is added here to illustrate the process.
jackd just starts with an empty buffer.

(1) can include pipeline. e.g.
  "in -> client A -> client B -> client C -> out"
they are all processed in the same cycle.


in reality 2a/2b is a bit more complex. If you're interested, search the
list archive for jack2 sync/async mode, read the documentation for the
-n parameter, and feel free to ask here.

Cheers!
robin

1397939839.21311_0.ltw:2, <5352DE74.2030201 at gareus dot org>