Quick Start Guide

Note

Start here if you are an adept HPC user and only want HAL-specific basics. If you’re new to HAL and HPC clusters, go to New User Guide.

Request Access to the System

Set Up a User Account.

Log in to HAL

The first time you log in to HAL, you must do so via SSH to initialize your account.

ssh <username>@hal.ncsa.illinois.edu

After your account is initialized, you can also log in to the HAL OnDemand dashboard with your NCSA username and password.

Queue and Resource Limits

Job Queue time limits:

• debug queue: 4 hours

• gpux<n> and cpun<n> queues: 24 hours

Resource limits:

• 5 concurrently running jobs.

• Concurrently allocated resources:

  • 5 nodes

  • 16 GPUs

• For larger/more numerous jobs, please contact admins for a special arrangement and/or a reservation.

Submit Jobs

• To submit an interactive job:

  swrun -p gpux1
  

• To submit a batch job:

  swbatch run_script.swb
  

Other Useful Commands

• To load the Open-CE module (provides PyTorch, TensorFlow and other ML tools):

  module load opence
  

• To see CLI scheduler status:

  swqueue
  

Transfer Data

The HAL endpoint collection name in Globus is “ncsa#hal”.