Host Setup

After the Holoscan sensor bridge board is set up, follow the directions on the appropriate page below to configure your host system.

• IGX running IGX OS 1.0 DP, with CX7 SmartNIC

• Jetson AGX Orin, L4T JP6.0, with Linux Sockets

Configure a few prerequisites in order to access and run the sensor bridge container:

1. Grant your user permission to the docker subsystem:

   Copy

   Copied!

               
   
               
   $ sudo usermod -aG docker $USER
           

   Reboot the computer to activate this setting.

2. Log in to Nvidia GPU Cloud (NGC) with your developer account:

   • If you don’t have a developer account for NGC please register at https://catalog.ngc.nvidia.com/

   • Create an API key for your account: https://ngc.nvidia.com/setup/api-key

   • Use your API key to log in to nvcr.io:

     Copy

     Copied!

                 
     
                 
     $ docker login nvcr.io
     Username: $oauthtoken
     Password: <Your token key to NGC>
     WARNING! Your password will be stored unencrypted in /home/<user>/.docker/config.json.
     Configure a credential helper to remove this warning. See
     https://docs.docker.com/engine/reference/commandline/login/#credentials-store
     
     Login Succeeded
             

Holoscan sensor bridge host software includes instructions for building a demo container. This container is used to run all holoscan tests and examples.

Build the demo container

1. Fetch the sensor bridge source code from https://github.com/nvidia-holoscan/holoscan-sensor-bridge

   Copy

   Copied!

               
   
               
   $ git clone https://github.com/nvidia-holoscan/holoscan-sensor-bridge
           

2. Build the sensor bridge demonstration container. For systems with dGPU,

   Copy

   Copied!

               
   
               
   $ cd holoscan-sensor-bridge
   $ sh docker/build.sh --dgpu
           

   For systems with iGPU,

   Copy

   Copied!

               
   
               
   $ cd holoscan-sensor-bridge
   $ sh docker/build.sh --igpu
           

   Notes:

   • --dgpu requires a system with a dGPU installed (e.g. IGX with A6000 dGPU) and an OS installed with appropriate dGPU support (e.g. IGX OS 1.0 DP with dGPU).

   • --igpu is appropriate for systems running on a system with iGPU (e.g. AGX or IGX without a dGPU). This requires an OS installed with iGPU support (e.g. for AGX: JetPack 6.0; for IGX: IGX OS with iGPU configuration).

Run the demo container

To run the sensor bridge demonstration container, from a terminal in the GUI,

            

            
xhost +
sh docker/demo.sh
        

This brings you to a shell prompt inside the Holoscan sensor bridge demo container. (Note that iGPU configurations, when starting the demo container, will display the message “Failed to detect NVIDIA driver version”: this can be ignored.) Now you’re ready to run sensor bridge applcations.

Sensor bridge host software includes a test fixture that runs in loopback mode, where no sensor bridge equipment is necessary. This test works by generating UDP messages and sending them over the Linux loopback interface.

In the shell in the demo container:

            

            
pytest
        

Note that the test fixture intentionally introduces errors into the software stack. As long as pytest indicates that all test have passed, any error messages published by individual tests can be ignored.

For systems with a sensor bridge device and IMX274, the test fixture can execute additional tests that prove that the device and network connections are working as expected.

First, ensure that the sensor bridge firmware is up to date.

For IGX configurations, connect both SFP+ connections on the sensor bridge device to the QSFP connectors, then

            

            
pytest --udp-server
        

For AGX configurations, only one camera is supported, so only SFP+ 0 is to be connected. Run the device test on AGX this way:

            

            
pytest --udp-server --unaccelerated-only
        

If things are not working as expected, check the troubleshooting page.