TAMU Launch

TAMU Launch uses Slurm for batch-scheduled workloads. Resource-intensive work must be submitted to compute nodes rather than run directly on a login node. Launch login nodes are intended for code development, compilation, file management, and other light work. Interactive processes on the login nodes are limited to one hour of CPU time and eight cumulative cores across all Launch login nodes.

From the Launch cluster shell in the portal, common Slurm commands include:

sbatch job.slurm          # Submit a batch job
squeue -u $USER           # List your queued and running jobs
squeue --job JOB_ID       # Check one job
scancel JOB_ID            # Cancel a job
seff JOB_ID               # Review efficiency after the job finishes

A minimal CPU job script is shown below. Save the contents in a file such as job.slurm and submit it with sbatch job.slurm.

#!/bin/bash
#SBATCH --job-name=example
#SBATCH --time=01:00:00
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --mem=4G
#SBATCH --output=example.%j.out

cd $SCRATCH
ml purge
# ml <module-name>

./my_program

GPU jobs must request the gpu partition and at least one GPU. To request any available GPU, include:

#SBATCH --partition=gpu
#SBATCH --gres=gpu:1

To request one NVIDIA A30 GPU specifically, use:

#SBATCH --partition=gpu
#SBATCH --gres=gpu:a30:1

Users who prefer a graphical workflow can open Jobs > Drona Workflow Engine in the portal. Drona generates and submits a job after the user supplies the requested resources and application commands.

For an interactive graphical session, select an application under Interactive Apps, enter the requested cores, wall time, memory, and node type, and select Launch. The interactive session is submitted as a job and can be managed under My Interactive Sessions.

See the TAMU Launch instructions for creating, submitting, monitoring, and cancelling Slurm jobs, the instructions for creating and submitting jobs with Drona, and the instructions for starting and managing interactive application sessions.