MTSU Introduction to HPC

Welcome to the Middle Tennessee State University (MTSU) Introduction to High-Performance Computing (HPC) resources. This repository is designed to assist Computational and Data Science (CDS) faculty and students in effectively utilizing the HPC clusters available at MTSU.

Table of Contents

• Prerequisites
• Cluster Overview
• Getting Started
• File Transfers
• SLURM Job Scripts
• Example Code
• Getting Help
• HPC Frequently Asked Questions (FAQ)

Prerequisites

Before using the HPC clusters, ensure you have the following:

1. MTSU Account:

   • You must have an active MTSU account and be added to the appropriate user groups to access the clusters.

2. SSH Client:

   • Install an SSH client on your local system. Examples:
     • Linux/macOS: Use the built-in ssh command.
     • Windows: Use PuTTY or Windows Terminal.

3. At least one of the following if off campus:

   • Jumphost Access:
     • This must be requested via ITD Help Desk. More information at https://help.mtsu.edu/sp
   • VPN Access:
     • This will likely only be available to faculty/staff.
   • Note: If on campus, you can directly ssh to cluster resources. WiFi still requires use of VPN or jumphost.

4. Basic Linux Knowledge:

   • Familiarity with the Linux command line is essential for effective cluster usage.

Cluster Overview

For a detailed overview of the HPC clusters, including hardware specifications, available software modules, and storage resources, please refer to the Cluster Overview document.

Getting Started

To begin using the HPC resources:

1. Accessing the Clusters:

   • Use SSH to connect to the cluster head nodes.
   • Ensure you have the necessary credentials and VPN access if required.

2. Loading Modules:

   • The clusters utilize the Environment Modules system.
   • Load the required modules for your workflow using the module load command.
   • For Python environments, it is recommended to use the miniconda module to manage dependencies.

3. Submitting Jobs:

   • Write SLURM job scripts to define your computational tasks.
   • Submit jobs using the sbatch command.

File Transfer

Detailed commands/guide for remote file transfer are located here.

SLURM Job Scripts

Pre-configured SLURM job scripts are available to assist you in submitting jobs efficiently. These scripts are located in the /projects/examples/scripts directory on the cluster. The available scripts include:

• basic_cpu.slurm: Template for basic CPU jobs.
• array_job.slurm: Template for job arrays.
• gpu_job.slurm: Template for GPU-accelerated jobs.
• high_mem.slurm: Template for high-memory jobs. (*Babbage only)

To use these scripts:

1. Copy the desired script to your working directory:

   cp /projects/examples/scripts/basic_cpu.slurm ~/my_jobs/
   

2. Modify the script parameters as needed.

3. Submit the job:

   sbatch ~/my_jobs/basic_cpu.slurm
   

SLURM Job Submission Guide

Detailed examples for submitting SLURM jobs can be found in SLURM Job Submission Examples.

Example Code

Example code is provided to demonstrate various computational tasks:

• Python:

  • CPU-based matrix multiplication.
  • GPU-accelerated TensorFlow model training.

• R:

  • Parallel processing using the foreach and doParallel packages.

These examples are located in the /projects/examples directory on the cluster:

• Python scripts: /projects/examples/python/
• R scripts: /projects/examples/r/

Feel free to copy and modify these scripts to suit your research needs.

Getting Help

For assistance with HPC resources:

• Email: [email protected]
• Documentation: Additional documentation is available at help.mtsu.edu/kb under the Computational Science Systems Knowledge Base.

---

Additional Documentation

For more details on HPC cluster usage, SLURM job management, and troubleshooting, see the HPC FAQ:

• HPC Frequently Asked Questions (FAQ)

--