Benchmarking audio performance

Pisound Projects
1

There has been some discussion of how build and configure an image, balancing tools, convenience, and performance. It raised questions like:

Does installing X on a system that will be used mostlly headless, cost much? Is it worth turning the service off by default?

I wrote a benchmarking tool to answer such questions. From the README.md:

Benchmarking Audio System Performance

These files benchmark the audio processing capacity of a particular set up.
The benchmark runs a test to find the maximum number of test voices it can
play simulateously without getting buffer underruns (“xruns”). The test is
repeated at multiple buffer sizes (-p in jackd).

You can find it here:

3 Likes
2

Results from my Pisound headless setup:

  • Distro: Raspian Stretch Lite
  • Kernel: Linux 4.14.34-v7+ #1110 SMP Mon Apr 16 15:18:51 BST 2018 armv7l GNU/Linux
  • RaspberryPi 3b
  • Pisound: server 1.02, firmware 1.01
  • SuperCollider: 3.9.3 (Built from branch ‘3.9’ [f61c21d3d], w/o GUI or IDE)

Basic

Fresh boot, ssh'd in:

64 ==> 42 voices, [ 43, 45, 39 ]
128 ==> 47 voices, [ 49, 48, 45 ]
256 ==> 50 voices, [ 50, 47, 52 ]
512 ==> 52 voices, [ 53, 51, 52 ]
1024 ==> 54 voices, [ 53, 54, 55 ]

Performance Governor

Ran:

& echo performance | sudo tee /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
# the Pi has only one governor, changing one changes them all

results:

64 ==> 79 voices, [ 78, 78, 80 ]
128 ==> 86 voices, [ 83, 88, 87 ]
256 ==> 94 voices, [ 95, 92, 96 ]
512 ==> 103 voices, [ 103, 103, 103 ]
1024 ==> 107 voices, [ 107, 106, 108 ]

Disabling X & VNC

Ran:

& sudo systemctl stop vncserver-x11-serviced.service  lightdm.service

results:

64 ==> 83 voices, [ 80, 85, 83 ]
128 ==> 86 voices, [ 84, 88, 87 ]
256 ==> 94 voices, [ 93, 95, 93 ]
512 ==> 101 voices, [ 103, 101, 100 ]
1024 ==> 106 voices, [ 107, 107, 105 ]

Not really all that different than with them running.

2 Likes
3

This is a great tool! Will definitely help when putting together a basic image for use with Pisound!

I think another good metric to keep track of would be the RAM usage of the tested image - I also kind of expected unused services which are running in the background not to cause much of a difference in terms of audio performance, but knowing what’s the expense of RAM used by background software might be useful in making decisions whether some software’s usage is not too big to include into the image, or for documentation purposes, so users may choose to remove or disable some of the included software manually, especially if he/she intends to do RAM intensive audio processing.

1 Like