[pulseaudio-discuss] best benchmark tool for pulseaudio ?

[Tanu Kaskinen]

On Tue, 2014-07-22 at 16:50 +0900, JoonCheol Park wrote:
> Hello,
> 
> 
> I'd like to optimize pulseaudio settings on my system (arm based).
> and I'm planning to try test with several change settings to find best
> settings on my system.
> To do this, I want to know what is the best way to measure the
> performance of pulseaudio for each settings. Especially I'd like to
> check the cpu consumption or latency.
> 
> 
> I tried to googling for this and reviewed code briefly to find some
> unit tests for performance checking..
> * oprofile - looks profile each function in point of developer view
> * top or ptop or perf top - difficult to check exact start/end time of
> pulse

Why would you need to check the exact start and end time of pulseaudio?
If you do cpu usage benchmarking, you typically start pulseaudio, and
then start one or more clients that do streaming with parameters that
you've decided for the benchmark. When the streams are running, you
check the cpu usage with some tool, e.g. top (remember to lock the cpu
speed first so that cpu speed scaling doesn't mess up the benchmarking
results). It doesn't matter when pulseaudio started.

You mentioned that you'd like to "check the latency". There's no such
thing as "the latency". Maybe you mean the minimum latency that
pulseaudio is able to provide without drop-outs on specific hardware for
a single stream with specific parameters? I don't know what would be the
best way to find that out, but running e.g. "pacat --verbose
--latency-msec=something /dev/zero" is one way to try to get some idea
of the minimum latency. pacat will print a message whenever there's an
underrun in the client buffer, so you can try different values for
--latency-msec to find the lowest value that doesn't cause client
underruns. For server-side underruns, you can watch the pulseaudio log
for messages like this:

I: [alsa-sink-ALC269VC Analog] alsa-sink.c: Underrun!
I: [alsa-sink-ALC269VC Analog] alsa-sink.c: Increasing minimal latency to 4.00 ms

-- 
Tanu