lkml.org 
[lkml]   [2004]   [Oct]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [Fwd: Re: [patch] Real-Time Preemption, -RT-2.6.9-mm1-V0.4]
Date
From
>Can anyone answer Ingo's questions about JACK & the best way to profile
>it? This is a bit over my head.
>
>Lee
>
>
>-------- Forwarded Message --------
>Subject: Re: [patch] Real-Time Preemption, -RT-2.6.9-mm1-V0.4
>Date: Thu, 28 Oct 2004 21:16:05 +0200
>* Rui Nuno Capela <rncbc@rncbc.org> wrote:
>
>> OK. Here are my early consolidated results. Feel free to comment.
>>
>> 2.6.9 RT-U3 RT-V0.4.3
>> --------- --------- ---------
>> XRUN Rate . . . . . . . . . . . 424 8 4 /hour
>> Delay Rate (>spare time) . . . 496 0 0 /hour
>> Delay Rate (>1000 usecs) . . . 940 8 4 /hour
>> Maximum Delay . . . . . . . . . 6904 921 721 usecs
>> Maximum Process Cycle . . . . . 1449 1469 1590 usecs
>> Average DSP CPU Load . . . . . 38 39 40 %
>> Average Context-Switch Rate . . 7480 8929 9726 /sec
>
>looks pretty good, doesnt it?

yes and no. its troubling that we're using an extra 100usecs of time
for the max process cycle, if the statistics make that number
meaningful. and why a 30% increase in the context switch rate? is that
an artifact or a real behavioural change? the xrun rate is not bad,
although i'd need to know the period size. 4 clicks per hour would
actually be unacceptable to most professionals, but this may have been
with significant loading outside of JACK - i don't know.

>how is the 'maximum delay' calculated? Could you put in a tracing hook
>into jackd whenever such a ~720 usecs maximum is hit? I'd _love_ to see
>how such a latency path looks like, it seems a bit long.

"maximum delay" is computed like this:

a) we know that jackd should be woken at regular, known
intervals based on the interrupt of the audio interface.

b) jackd is (often) blocked in poll(2), waiting for the
audio interface to become readable+writable.

c) on return from poll, we check the time we are returning
and compare it to the last wakeup. the difference should
be almost exactly the interrupt interval.

d) any amount larger than this is a delay caused by
kernel scheduling. we track the largest such delay.

>also, i looked at the sound/pci/ali5451/ali5451.c driver code and it has
>one weird piece of code on line 988:
>
> udelay(100);

yeah, some ALSA drivers have broken stuff like this in them. its a
definite problem. this is why we know that some audio interfaces are
not able to provide acceptable performance for JACK - that is OK, the
same is true under Win/Mac where some h/w just can't be used for low
latency work.

>plus i've also got questions about how Jackd interfaces with ALSA: does
>it use SIGIO, or some direct driver ioctl? If SIGIO is used then how is
>it done precisely - is an 'RT' queued signal used or ordinary SIGIO?
>Also, how is the 'channel' information established upon getting a SIGIO,
>is it in the siginfo structure?

we absolutely do not use SIGIO - i am one of the most vocal critics of
this interface even be present. POSIX says that signal handling
context is essentially crippled, so there is very little you can do
from there. in addition, measurements of signal delivery time on Linux
suggest that it is really quite slow. i considered using signals to
wake jack clients, but it was dramatically slower than using a FIFO
plus poll(2).

we block in poll(2), waiting for the interrupt from the audio
interface to mark the relevant jackd thread as runnable.

i don't know what you mean by "channel" information in this context.

--p
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 14:07    [W:0.135 / U:0.140 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site