Re: [LAU] jack performance in ubuntustudio

From: Len Ovens <len@email-addr-hidden>
Date: Sat Jul 26 2014 - 03:20:31 EEST

On Sat, 26 Jul 2014, Joakim Hernberg wrote:

> On Fri, 25 Jul 2014 06:27:54 -0700 (PDT)
> Len Ovens <len@email-addr-hidden> wrote:
>
>> Running ubuntustudio 14.04 here. jackdbus is set 2/32 and I turned
>> pulse->jack bridging on. qjackctl shows DSP at 6-12%, Gnome system
>> monitor shows all cores at around 4%, Top shows pulse at 11%, the
>> sound card at 6%, jack at 3.5%, a2jmidid at 2%.... turning pulse off
>> has top showing the sound card at 8%, jack at 6% and a2j at 2%, but
>> jack DSP is 3.5%. All of which brings into question the tools we use
>> for measuring this stuff and what those numbers really mean.
>>
>> Setting my system to desktop mode (2048 frames, pulse on) gives jack
>> dsp of .88% (or less) and makes them harder to follow in top because
>> they are now at least fifth down and all less than 1%.
>
> You are comparing apples and oranges :)
>
> Most of those measurements are cpu use, qjackctl's dsp load is a
> measure of latency (execution time). That is to say there is a deadline
> given by the sample rate divided by the buffer size. If DSP load is
> 50%, the audio processing has used up half of the time available. If
> 100% it has used up all the time. Note that there is overhead
> involved, the conventional wisdom is to not go over 50%, but ymmw..

I think that was my point. Except I was also pointing out that looking at
cpu usage is not the greatest number to use because every program that
shows it seems to say something different, so knowing what that number is
made up of is important. Probably DSP load gives the best idea of what is
left in the system for audio. CPU usage does not take into acount that
some things that jack does are all on one thread/core.

--
Len Ovens
www.ovenwerks.net
_______________________________________________
Linux-audio-user mailing list
Linux-audio-user@email-addr-hidden
http://lists.linuxaudio.org/listinfo/linux-audio-user
Received on Sat Jul 26 04:15:03 2014

This archive was generated by hypermail 2.1.8 : Sat Jul 26 2014 - 04:15:03 EEST