Performance

I’ve done some limited performance tests (turning a GPIO on then off, see GpioPerfTest) on a Raspberry Pi Zero, 2B, 3B and 4B using the various provider implementations. I’ve also run tests using the underlying APIs directly without going via the diozero APIs to assess the overhead of diozero.

The diozero performance test applications are all in the com.diozero.sampleapps.perf package. See PigpioPerfTest for a similar performance test application for pigpioj without diozero. The mmap-tests folder contains some basic applications for testing mmap and gpio chardev performance without diozero.

Results (with diozero 1.2.0 and pigpioj 2.5.7):

• All tests were performed with OpenJDK 11 (“11.0.9.1” 2020-11-04) using default settings.
• An LED was connected to the GPIO to verify that the GPIO was being toggled.
• The Pi 3B diozero (builtin) results have been verified with an oscilloscope.
• Operating System and libraries were all current as of April 2021 (RaspiOS 5.10.17-v7+, Armbian 5.10.21).
• Operating Systems are 32-bit unless otherwise stated.
• mmap controlled GPIO is not currently available for the BeagleBone Black using the built-in provider - it is via the BBBioLib provider.

Conclusion

• While pigpio itself is extremely fast, the JNI layer introduces significant overhead
• The builtin memory mapped GPIO implementation is extremely fast
• 64-bit O/S delivers additional performance for the memory mapped GPIO implementation

Raw Numbers

---

Table of contents

• GraalVM
• Java Flight Recorder

---