[Optimize] nsight systems remote profile python in container on Orin

Please provide the following info (tick the boxes after creating this topic):
Software Version
[*] DRIVE OS 6.0.6

Target Operating System
[*] Linux

Hardware Platform
[*] DRIVE AGX Orin Developer Kit

Host Machine Version
[*] Linux Containers on DRIVE AGX Orin

Desc:
I want to profile python code, which in docker container which on DRIVE AGX Orin, using nsight systems on my Linux host remotely.
Currently I can profile the docker run shell script on DRIVE AGX Orin. But that can’t provide more information in my python program(eg, python code cache hit rate, cpu usage by process/thread id).

Dear @ChinSun,
I notice this issue is filed via NVBug as well. We continue this issue support via Nvbug.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.

Profiling in Docker is possible in two ways:

A) Use the CLI inside the container, and then open the resulting report file using nsys-ui. To do that, please follow these steps:

  1. Run Nsight Systems GUI on the host Ubuntu system (x86_64), and connect to DRIVE AGX Orin over SSH. First time this is done, the target binaries will be copied to /opt/nvidia/nsight_systems/ on the target. All next steps will happen on the target (DRIVE AGX Orin).
  2. Run “/opt/nvidia/nsight_systems/nsys --version” to verify that the Nsys CLI works correctly.
  3. Mount the nsys directory into the container, and also mount the output directory where report files will be created, in this example /tmp/nsys_out:
    mkdir -p /tmp/nsys_out
    docker run -v /opt/nvidia/nsight_systems:/nsys:ro -v /tmp/nsys_out /nsys_out …
  4. Inside the container, run a profiling session:
    /nsys/nsys profile -o /nsys_out/report.nsys-rep …
  5. This will create file /nsys_out/report.nsys-rep in the container, and therefore /tmp/nsys_out/report.nsys-rep outside of the container. Copy this file over to the host, and open it in the Nsys GUI.

B) Install and configure SSH server in the container, and connect to the container directly from the GUI