Ignite Parallel Cluster Goals

[MrA2K2]

1. • Add user to docker in add_new_user.bash (and post_install.bash) (Adds user to docker group if docker is present #21) (outdated) => we removed docker and reverted this feature
2. • Packer Github Actions

3. ... removed ...

4. • Create our custom AMI on ubuntu 20.04 + nvidia drivers and docker (outdated)
   5. • Create our custom AMI on ubuntu 20.04 + nvidia drivers and enroot/pyxis
5. • Create playground cluster using custom AMI (playground configuration updates #20)
6. • Change Default location of Docker Images using /var/docker/daemon.json, (https://stackoverflow.com/a/24312133 ; https://docs.docker.com/engine/reference/commandline/dockerd/#daemon-configuration-file) => this wont work as expected
7. • Let's check https://github.com/NVIDIA/pyxis instead and see how it would work => using enroot and pyxis
8. • Download 1-2 Docker Images as sqsh to /shared/enroot_data
9. • Test with a new user: launching cifar10 training on 2 nodes with 1 GPU per node
     • using conda
     • using enroot with docker image: pytorchignite/apex-vision:latest
       • expose /shared and no root mapping inside container: Mount shared and no root mapping inside container #26
10. • Download pascal voc training dataset and launch training on 2 nodes with 1 or 4 GPU(s)
      • using pytorch_ignite_vision conda env
      • using enroot as cifar10, run_docker.sbatch
11. • Create the aws-parallel-cluster tutorial and a presentation for the team