Raspberry Pi 4B: LinuxCNC Max Jitter or Latency Test

I conducted the LinuxCNC Max Jitter or Latency Test on my Raspberry Pi 4 Model B Rev 1.5, equipped with 8GB of RAM and running Debian GNU/Linux 12 (Bookworm) 6.12.11. The results are rather erratic, and I’m not sure how to interpret them. In this post, I present the results and describe the tests.

🐧 Index of the Complete Series.


Raspberry Pi 4B: LinuxCNC Max Jitter or Latency Test

I use the following official image: https://linuxcnc.org/iso/rpi-4-debian-bookworm-6.12.11-arm64-ext4-2025-01-27-0404.img.xz. Then, I enabled WiFi, installed Samba, and set up remote desktop access. For more information on these official images, please refer to this forum post: Linuxcnc & the Rpasberry Pi (4 & 5).

The command:

C:\Users\behai>ssh cnc@192.168.0.45

reports the following information:

cnc@192.168.0.45's password:
   Debian GNU/Linux 12 (bookworm) 6.12.11 #1_RT Mon Jan 27 02:47:25 AEDT 2025
                                    aarch64

                                                 Raspberry Pi 4 Model B Rev 1.5

── DISK
Filesystem      Size  Used Avail Use% Mounted on
/dev/mmcblk0p2   29G  4.8G   23G  18% /
/dev/mmcblk0p1  508M   97M  411M  20% /boot/broadcom

── NETWORK
Hostname:    picnc
Wired:       eth0 DOWN 10.10.10.100
Wireless:    wlan0 UP 192.168.0.45

── SYSTEM
Processor:   Cortex-A72 @ 1800MHz 43°C
Frequency:   1800MHz
Online:      0-3
Governor:    performance
Memory:      7.6G 370M
Entropy:     256
Uptime:      14:08:00 up 4 min, 0 user, load average: 0.03, 0.18, 0.09

Last login: Fri Feb  7 01:22:37 2025 from 192.168.0.2

And the kernel information:

$ uname -a

reports:

Linux picnc 6.12.11 #1 SMP PREEMPT_RT Mon Jan 27 02:47:25 AEDT 2025 aarch64 GNU/Linux

❶ The Test Environment

The Raspberry Pi 4B was enclosed in a heat sink, but the fans were not running.
WiFi is enabled.
A USB wireless keyboard and mouse were used.
A USB stick was plugged in, but it was not accessed during any tests.
The Geany editor was running and used during the tests.

❷ Test Results and Description

⓵ Test 1, Duration: 111 Minutes

Max Jitter recorded at the end of this test:

Servo Thread: 164,463 nanoseconds (164.463 microseconds). This was reached at the 41st minute.
Base Thread: 210,589 nanoseconds (210.589 microseconds). This was reached at the 66th minute.

While the test was running, I carried out the following activities:

From the start to the end of the test, Firefox was trying to load a 350MB MP4 file, but it never loaded and played.
From the 1st to the 17th minutes: ⓵ Used the file explorer application to copy and delete the 350MB MP4 once. ⓶ Opened a terminal window. ⓷ Switched windows and edited a text file.
From the 17th to the 41st minutes: ⓵ Used the file explorer application. ⓶ Opened another Firefox tab to download a 28MB file. ⓷ Opened a third Firefox tab to search. ⓸ Ran glxgears. ⓹ Switched windows and edited a text file.
👉 Please note: The Max Jitter for the Servo Thread reached 164,463 nanoseconds and remained constant until the end of the test.

🙏 At this point, the heat sink felt really hot, so I wrapped a wet cloth around it. Eventually, the heat sink temperature came down, and it just felt warm to the touch.
From the 41st to the 66th minutes: ⓵ Had 3 glxgears instances running. ⓶ Used the terminal windows. ⓷ Used an existing Firefox tab to search. ⓸ Did a screen capture. ⓹ Switched windows and edited a text file.
👉 Please note: The Max Jitter for the Base Thread reached 210,589 nanoseconds and remained constant until the end of the test.
From the 66th to the 93rd minutes: ⓵ Progressively had 6 glxgears instances running. ⓶ Used Ristretto Image Viewer to view an image, then terminated it. ⓷ Switched windows and edited a text file.
From the 93rd to the 111th minutes: Progressively stopped applications. ⓵ Closed the Firefox tab which was still trying to load a 350MB MP4 file. ⓶ Then terminated all 6 glxgears instances at some minute intervals. ⓷ Switched windows, edited a text file, etc.
Shut down the HAL Latency Test application.

The screenshot below shows the result of this test:

⓶ Test 2, Duration: 109 Minutes

Max Jitter recorded at the end of this test:

Servo Thread: 175,368 nanoseconds (175.368 microseconds). This was reached at the 60th minute.
Base Thread: 127,312 nanoseconds (127.312 microseconds). This was reached at the 60th minute.

🙏 Throughout this test, the temperature of the heat sink felt warm to the touch. It was not too hot.

For this test, only the Geany editor was running initially. Then, I started the HAL Latency Test application. To start off, the Servo Thread jitter was 68,997 nanoseconds, and the Base Thread jitter was 67,054 nanoseconds.

While the test was running, I carried out the following activities:

From the 1st to the 43rd minutes: ⓵ Started Firefox and used Google to search for various things. ⓶ From the 26th minute, a total of 6 glxgears instances were running. ⓷ Switched windows and edited a text file.
From the 43rd to the 60th minutes: No activities other than switching windows and editing a text file. Both the Servo Thread and the Base Thread jitters increased.
👉 At the 60th minute, the Max Jitter for the Servo Thread reached 175,368 nanoseconds, and for the Base Thread reached 127,312 nanoseconds. These values remained constant until the end of the test.
From the 60th to the 109th minutes: ⓵ I progressively started a total of 12 glxgears instances. ⓶ Then, I progressively shut them down 3 at a time until there were none. ⓷ Switched windows and edited a text file in between.
The Max Jitter for both threads remained unchanged from the 60th minute.

The screenshot below shows the result of this test:

⓷ Test 3, Duration: 30 Minutes

Continuing from Test 02 above, except for the Geany editor, all other applications were terminated. With only the Geany editor running, I clicked the Reset Statistics button and let the test run.

Max Jitter recorded at the end of this test:

Servo Thread: 67,333 nanoseconds (67.333 microseconds).
Base Thread: 94,072 nanoseconds (94.072 microseconds).

❸ According to the official page on Latency Testing, the results reported by the first two tests indicate that this computer is not a good candidate for running LinuxCNC. However, people have successfully run LinuxCNC on similar hardware and software, so I might conduct some more tests.

I am using the 7I96S STEP/IO Step & dir plus I/O card for LinuxCNC. Since the setup uses hardware step generation, kernel latency is somewhat less critical. However, I still want to determine the latency as accurately as possible. I will revisit this issue in the future as needed.

If you happen to read this post, thank you for your time, and I hope I did not waste it. Stay safe, as always.

✿✿✿

Feature image source: