This respository contains all the configuration files (modulo appropriate secrets) for deploying the computational environment of our research team across our local (i.e. campus-based) workstations.
We now use a kubernetes-based approach, replacing the pure-docker approach we used across our platforms previously (see servers repo). This retains the same containerized abstractions for the software stack (often the very same docker containers), but provides additional abstractions around the hardware, orchestration, and resource management.
Home to the configuration files for our lab jupyterhub.
See Zero to JupyterHub for excellent official documentation on everything.
See blog post for some notes on this setup, specifically for k3s and GPU.
Nvidia container toolkit setup
(not strictly necessary? install nvidia-container-toolkit and simply enable nvidia runtime in Jupyter)
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html
RAM use (etc) of active container (cgroup)
cat /sys/fs/cgroup/memory.max | awk '{printf "%.2f GB\n", $1/1024/1024/1024}'
cat /sys/fs/cgroup/memory.current | awk '{printf "%.2f GB\n", $1/1024/1024/1024}'