Welcome to HPC Helper’s documentation!

This documentation summarises my experience mainly with Sunbird, a HPC cluster located at the Computational Foundry, Swansea University, Wales. One can follow this basic tutorial to understand the standard operations on any HPC cluster.

These tutorials are fairly advanced and is meant to increase the productivity. A comprehensive but non-intuitive guide can be found here.

Basics Tutorials

SSH basics

• Automatic SSH
  • Create a config file
  • Generate SSH keys
  • Copy the Public key
  • The need of a SSH agent

Remote VS Code

• VS Code
  • Remote-SSH extension
  • Once you are connected.
  • Accessing the terminal
  • Accessing the file explorer
  • Terminating the ssh session

Submit Interactive Jobs

• Use GPU on interactive session
• List of partitions
• Start an interactive session
• Loading Anaconda
• Run a python file
• Two GPUs

Submit Batch Jobs

• Job submission with GPU: Running batch jobs
• Run bash file

MIG partition Issues

• Interactive GPU best practices
• See only allocated GPUs in nvidia-smi
• Problem with MIG partition

Intermediate Tutorials

Jupyter Lab port forward on local network

• Remote Jupyter Lab
  • Setup the Windows server
  • Setup the Linux Client
  • Configure Jupyter Lab

Jupyter Lab port forward from the Login node of Sunbird

• Jupyter lab on Sunbird using port forwarding
• Installing Jupyter lab
• Tinkering configuration file
• Port forwarding
• Running Jupyter lab on login node
• Wrapping ssh command in Bash file

Jupyter Lab port forward from a compute node of Sunbird

• Jupyter lab server on a GPU node
• Job Script
• Node number
• What if someone is already using port 8888 on Sunbird or your local PC?
• Running Jupyter lab on local PC usign ssh tunnel
  • Killing a process that is using port 8888 on local PC
• Some screenshots

Advanced Tutorials

Remote VNC session

• Running GUIs on Sunbird via VNC server
  • Start VNC server
  • Check log file for errors
  • Error: cannot open display
  • Port Forwarding
  • Setup a VNC client
  • Running an app
  • Closing VNC server
  • Killing VNC server on Sunbird
• Quick VNC server on sunbird
  • Start VNC server
  • Port forward
  • Using GPUs in VNC session
  • GPU job forwarding to VNC server
  • Checking GPUs

Swansea University VPN Tunnel

• Setting up the VPN
• Setting up the VPN software
• Setting up the host (Ubuntu only)
• Setting up for SSH
• Setting up the RDP server
• Enable connection after logout from the host
• Getting the Local IP
• Using SSH
• SSH
• Adding an alias for the HOST PC
• Passwordless login
• RDP connection

Docker Tutorials

Basic commands

• Docker containers
• What is a Container? What is Docker?
• Virtualisation
• Potential applications
• Docker command line
• Install using the repository
• Copy paste
• Docker is working correctly ?
• Getting help
• Docker Command Line Interface (CLI) syntax
• Running containers interactively
• Installing Ubuntu
• Summary
• Cleaning up containers
• Finding Containers on Docker Hub
• Exploring Container image versions
• Docker Hub is a repository
• Create a new Docker image
• Share your new container image on Docker hub

Advanced commands

• Creating more complex container image
• Using scripts and files from outside the container
• Including your scripts and data within a container image
• If the files are on a GITHUB repo
• More fancy Dockerfile options
• Add the sum.py script to the PATH so you can run it directly

Port Mapping of Jupyter server

• Jupyter lab inside Docker container
• Run a Docker image as container interactively
• Running the Jupyter-lab server
• Opening Jupyter-lab in browser
• Some screenshots

Singularity Tutorials

Singularity Basics

• Singularity: software carpentry
• Images and containers
• Pulling image
• Running an image
• Singularity’s image cache
• Running specific command with a Singularity container
• Singularity container: interactive shell

Volume Binding

• Files and directories within a Singularity container
• Binding additional host system directories to the container

Basic Builds

• Using Docker images with Singularity
• Running pulled image

NVIDIA Modulus Singularity container (Failed)

• Nvidia Modulus 22.03.1 Docker images on Sunbird

NVIDIA Modulus Setup

v22.03

Interactive GPU session

• Setup Nvidia Modulus v22.03 on Sunbird using interactive GPU session
• Installation
• Running an interactive GPU session

Multi-GPU

• Parallel Processing using Multi-GPU Configurations

With PySDF

• Nvidia Modulus with tesselated geometry module on Sunbird
• Installing Python 3.8
• Installing Pytorch
• Installing Modulus dependencies
• Installing Modulus
• Trying an example
• Running Nvidia Modulus Docker image
• Final testing

v22.03 Summary

• Efficiently train model

v22.07 - Apptainer Image

Constructing the Apptainer script

• Getting the base image ready to use
• Getting the Modulus source code
• The base image
• Getting the base image ready
• Converting the recipe
• Build the image
• Testing the image

Overiding Defaults of Apptainer

• Problems with Apptainer
• Running the Modulus Apptainer image
• Using the GPU
• Downloading Modulus examples
• Testing the Modulus Apptainer image
• Apptainer supports guest pull from NVIDIA NGC

Multi-GPU trainign using mpirun

• Running jobs using multiple GPUs
• Problem with mpirun (SKIP IF NEEDED)
• Use mpirun
• Best Practice

v22.09 - Apptainer Image

The Apptainer script

• Building an Apptainer image of Modulus v22.09
• Getting the examples using the SSH key
• Port forwarding the Jupyter-lab
• Checking if the port is already occupied
• A robust way to execute jobs