High Performance Computing

This repo contains a guide to working in the High Performance Computing (HPC) cluster at NYUAD.

1. Main guide for NYUAD HPC.
2. Loading packages as modules
3. Setting up a Python Virtual Environment
4. Running parallel CPU and GPU jobs with slurm

Additional Guides

• Adding color to the terminal

Tools of the trade

Accessing and working with the HPC requires usage and familiarity with the linux command-line interface to:

• ssh into portal machines.
• Navigate folder directories.
• Bash scripting to launch scripts,
• Bash scripting to launch parallel CPU computing tasks/jobs.
• Bash scripting to launch neural network GPU training tasks/jobs.

The HPC is not necessarily designed for a jupyter notebook workflow, but rather the creation of .py files (and scripts in other languages) that can be launched from the command line, with results written to file.

Connecting to HPC

1. Apply for an account here.
2. Connect by VPN, and then ssh:

ssh <your NetID>@jubail.abudhabi.nyu.edu

System Overview

There are 4 directories you have access to:

1. $HOME (/home/)
   • Keep small persistent fraction of code here (e.g., source code, executables, Python packages ...).
2. $SCRATCH (/scratch/)
   • Put all data here.
   • Run jobs from here.
3. $WORK (/work/)
   • Mountable on your local workstation.
   • Useful for quick post-processing, analysis and visualization, without moving data.
4. $ARCHIVE (/archive/)
   • For long-term storage, archive work here.

	$HOME	$SCRATCH	$WORK	$ARCHIVE
Use for storing	source code / executable / perl-python-R packages	data	anything	anything
Accessible From	login / compute	login / compute	login	login
Use to Run Jobs	No	Yes	No	No
Retention Time (Days)	No Limit	90	120	No Limit
Mountable	No	No	Yes	No
Default Quota	5GB, 100K Files	5TB, 500K Files	5TB, 500K Files	5TB, 125K Files

Note: Backing up work is your responsibility. It is suggested that you always push your work onto a github repo. For larger data, we can setup a database with automatic archiving.