[linux-audio-user] Re: [Jackit-devel] 2.6.10

Hi,

Just some notes about my jack_test3 suite I've posted a few days back,
which I find worth regarding.

Always check the source on jack_test3_run.sh and try to adapt it to your
own needs or environment. This is highly recommended if you're trying to
run some definitive tests.

The results obtained only make sense for your own hardware and software
setup. In my case, it has been barely useful about comparing _between_
kernels and/or jackd versions. It was not (never) meant to benchmark your
specific hardware, nor it makes sense when comparing with someone else
results.

The main objectives have been to check and confirm the performance
benefits of the Ingo's RT patch, specially regarding the jackd realtime
audio work. It may also fit to stress your audio station into such a
(un)practical scenario that it may fall apart. I think we all be
discussing those scenarios very soon :)

For example, my laptop starts behaving in a seriously compromised state if
I reach above the 24 test clients. It runs fairly well for a couple of
minutes, topping 90%CPU, then it lockups badly with no determined cause.
Guess this behavior will be my next subject onto the kernel folks ;)

Matthew Allen wrote:
> using 2.6.10-rc3-mm with Ingo's latest Real-Time Preempt on a pentium
> M 1.6ghz laptop with a gig of ram. The test was done as a normal user
> using the realtime-lsm patch and I have Rui's rtirq script up and
> running. One of the three tests was completed while compiling gnuplot
> in the background :).
>

Please always try with latest JACK CVS, where some additional statistics
are available [0.99.41+]. Read on.

> *********** CONSOLIDATED RESULTS ************
> Total seconds ran . . . . . . : 1200
> Number of clients . . . . . . : 20
> Ports per client . . . . . . : 4
> Frames per buffer . . . . . . : 64
> *********************************************
> Summary Result Count . . . . : 3 / 3
> *********************************************
> Timeout Rate . . . . . . . . :( 0.0)/hour
> XRUN Rate . . . . . . . . . . : 4.0 /hour
> Delay Rate (>spare time) . . : 85477.0 /hour

These delays (>spare time) are pretty bad, quite as bad as XRUNs, if not
worst. Beware.

> Delay Rate (>1000 usecs) . . : 0.0 /hour

This one only gets a non-zero result if you have the
max_delayed_usecs-histogram patch applied. I sure you're not, but this
isn't important either. Just ignore it.

> Delay Maximum . . . . . . . . : 0 usecs

Similarly, this line will be only operational if you have the
max_delayed_usecs functionality on jackd. This has been made available
after JACK 0.99.41cvs. Check it out.

> Cycle Maximum . . . . . . . . : 0 usecs

Zero process cycle? This is no good news at all. Most probably your test
run didn't make it. I guess you'll have to try it again, with less
stressful arguments (the default of 20 clients might be too much for your
setup).

Don't consider for consolidation any of those log results that have either
non-zero timeouts or a zero cycle maximum time. They're bogus and just
damn plain noisy.

> Average DSP Load. . . . . . . : 34.0 %
> Average CPU System Load . . . : 25.5 %
> Average CPU User Load . . . . : 27.7 %
> Average CPU Nice Load . . . . : 0.0 %
> Average CPU I/O Wait Load . . : 0.1 %
> Average CPU IRQ Load . . . . : 0.0 %
> Average CPU Soft-IRQ Load . . : 0.0 %
> Average Interrupt Rate . . . : 1881.0 /sec
> Average Context-Switch Rate . : 42065.8 /sec

Your machine reached this context-switch rate? Geez. I wonder how it could
make anything else besides context-switching processes :) Definitively
this run was too much your machine could handle. IMHO.

> *********************************************
>
> I ran it 3 times. twice at 300s and once for 600 seconds. However I
> dont think It actually ran for 5 minutes or 10 minutes. Every time I
> ran it I got a similar warning/error (maybe I am supposed to get
> this?)
>
> $ ./jack_test3_run.sh 600
> Launching jack_test3_nmeter...done.
> Launching 20 ./jack_test3_client(s) instance(s)...done.
> Running
> jackd...1...2...3...4...5...6...7...8...9...10...11...12...13...14...15...16..../jack_test3_run.sh:
> line 48: 32325 Killed ${CMD} >>${LOG} 2>&1
> done.

OK. That's it. Your test runs were being terminated prematurely. I think
your setup can't handle the stress of more than 16 clients. This maybe
caused by an old jackd version. Please, retry with 14 clients only, and be
back after then ;)

Cheers
and a
happy new year.

-- 
rncbc aka Rui Nuno Capela
rncbc@email-addr-hidden