Nvargus-daemon cpu usage

[jfk214]

This forum post mentions that when running 6 cameras at 1 megapixel at 60 Hz, it is expected behavior that the nvargus-daemon will use 3 out of the 6 cores on the Xavier NX. What is happening inside the navargus-daemon that is using this amount of CPU?

High CPU usage of nvargus-daemon

Hi, We are seeing the following CPU usage for nvargus-daemon when capturing with nvarguscamerasrc on Xavier NX / Jetpack 4.6 / 1280x720@60 fps: 2 cameras: 94% 4 cameras: 170% 6 cameras: 256% Q1) This CPU usage seems high and increases considerably with the number of cameras. Is this the expected CPU load and behavior for nvargus-daemon? Q2) How can we reduce the CPU usage of nvargus-daemon? Thanks

Also, this forum post mentions that the source code for nvargus-daemon is available, but I couldn’t find it in the public_sources link provided in the answer. Do you know where it is?

Where is the source code for the process which name "/usr/sbin/nvargus-daemon"?

Where is the source code for the process which name “/usr/sbin/nvargus-daemon”? And if use the nvarguscamerasrc plugin,does it get its image source data from this process?

[JerryChang]

hello jfk214,

may I know what’s your approach to evaluate CPU usage, are you using tegrastats then average the CPU usage? if you’re using top utility, please do remember to switch-off irix mode (i.e. shift+i) to get average CPU usage.

number of cameras and also camera frame rate that influences CPU utilization.
there’re some CPU usage for the camera use-case, it’s camera application and EGL streams consume CPU resources, mostly CPU usage was taken by argus_camera application due to buffer transmit. besides, camera pipeline still need CPU to run lots of algorithms like 3A.

please access Jetson Linux | NVIDIA Developer page and download the [L4T Driver Package (BSP) Sources], you may also download MMAPI via APT server,
for example, $ sudo apt install nvidia-l4t-jetson-multimedia-api.

[jfk214]

Hi Jerry, thanks for the quick reply. I’ll keep disabling irix mode in mind. I’m assuming the cpu usage numbers in the post I linked had it enabled since the percentages are over 100. I also have some follow up questions regarding your answer.

1. Since you mentioned the EGL streams, I also found this post which mentions that using IBufferOutputStream instead of IFrameConsumer will avoid copying to the EGL stream. Would you expect doing this to save a significant amount of cpu?
   IFrameConsumer vs IBufferOutputStream - #7 by ShaneCCC

2. Does 3A algorithms refer to auto-focus, auto-exposure, and auto-whitebalance? If so would you also expect disabling them to save a significant amount of cpu? I think I can disable them with the Argus::IAutoControlSettings. Are there any other algorithms besides 3A that I could disable?

3. I downloaded the BSP sources you linked but couldn’t find the nvargus-daemon source code. Do you know the filename and path where it is located?

[JerryChang]

hello jfk214,

Q1
it should be, you may have quick try to compare the CPU usage by disable the camera preview frames rendering.
for example, $ gst-launch-1.0 nvarguscamerasrc sensor-id=0 ! 'video/x-raw(memory:NVMM),width=1920, height=1080, framerate=30/1, format=NV12' ! nvvidconv ! 'video/x-raw(memory:NVMM),format=I420' ! fpsdisplaysink text-overlay=0 video-sink=fakesink sync=0 -v

Q2
yes, 3A means auto-focus, auto-exposure, auto-white balance… such camera controls. you may try enable camera stream with argus_camera application, there’s user-interface for you toggle the AE lock, please try enable AE lock and you shall see slightly CPU usage drop.

Q3
sorry, we don’t public the sources code of /usr/sbin/nvargus-daemon.
you may download MMAPI via APT server, $ sudo apt install nvidia-l4t-jetson-multimedia-api to get the sources of argus_camera application.