I’m currently using python programs to train neural networks in lxc containers with Nvidia GPUs.
To find and address performance bottleneck, I was planing to do profiling with Nvidia Nsight Systems tool.
However, I wasn’t able to get backtrace information maybe due to security restrictions.
I’ve tried to find relavant approach to get over the restrictions with no good:
- security.syscalls.allow = perf_event_open (failed to start containers)
- security.privileged = true (can’t be used when nvidia.runtime=true)
Nvidia officially give instructions for Docker users: User Guide :: Nsight Systems Documentation (nvidia.com).
I was wondering if there is any way to use this tool with full functions inside lxc containers?