Skip to content

Home

Jump to bottom
rpguiteras edited this page Sep 15, 2025 · 13 revisions

Welcome to the ARE-HPC wiki. This page collects basic information on using the NCSU HPC cluster.

Getting started

The four most important things to note when getting started are

  1. only faculty / PIs have automatic access. Students will need to request access through their advisor / PI, and this can take a little time. See https://hpc.ncsu.edu/Accounts/GetAccess.php

  2. base storage on the cluster is fairly limited (1GB). See https://hpc.ncsu.edu/Documents/Storage.php for information on storage. There is a lot of "scratch" space (10TB per project), but this is not backed up. You can link to OIT Research Storage (2TB for faculty by default), but this takes a little time and effort to set up.

  3. ARE users have access to an ARE-only node, in addition to the general resources provided by HPC. After you have set up your HPC account (see item 1 above), you can use the HPC (ARE) Access Form to submit your request for access to the are node.

    Note -- the are node has 64 cores, and users on the are node are limited to 16 cores at a time. If you require or would benefit from more cores, use the general HPC resources (limit of 1,024 cores at a time, although you may never hit this limit since your jobs will typically not all be given highest priority at the same time unless overall HPC use is low).

  4. To make most effective use of the cluster, it's important to understand (a) the processors on the cluster are not all that powerful or fast, the main benefit of using the cluster is that you can use a lot of them and (b) it is easier to access the cluster's resources by requesting many small (in terms of time, memory and especially number of cores) jobs rather than a few large jobs. If your software supports parallelization / distributed memory (for example, R with the Rmpi package), use the HPC's built-in parallel job features described here and here. If not (for example, Stata), you can replicate some features of distributed memory using LSF arrays. See Using arrays to speed up multiple jobs for an example in Stata. Check the program-specific HPC page for more information on parallelization with your software.

Users unfamiliar with cluster computing

If you are not so familiar with cluster computing, we recommend starting by working through OIT's excellent Quick Start Tutorial: https://hpc.ncsu.edu/Guide/

Before you do the tutorial, you will need to know some basic Linux. Here are some recommendations from OIT:

This tutorial may be sufficient for you to get up to speed. For a more thorough training module, see the HPC Workshop on REPORTER. This workshop starts with a review of basic Linux and continues through submitting parallel jobs.

OIT also has some nice video tutorials on YouTube

Users with some experience with cluster computing

If you are already familiar with cluster computing, linux, etc., you may want to skip to the "Links to more detailed information" section below.

However, the OIT Quick Start Tutorial and the REPORTER HPC Workshop referenced above are nice refreshers.

The important thing to note is that you do not run jobs directly at your login prompt. Rather, you write an LSF batch script (a.k.a. .csh) to handle running your job and then submit that script to the cluster. The script contains certain parameters affect the amount of memory you are requesting from the cluster, etc.

See Running jobs for more information on LSF / .csh scripts and submitting jobs.

Links to more detailed information

ARE-specific information: https://areit.wordpress.ncsu.edu/hpc-high-performance-computing/

Overview from OIT: Getting Started on Hazel

Specific topics

  1. Basics (access, tutorials)

  2. Managing users

  3. Logging in

  4. Storage

  5. Managing files

  6. Software

  7. Running jobs

  8. Stata-specific material

  9. Using arrays to speed up multiple jobs

Clone this wiki locally