Recommendations on using the shell on shared HPC systems

Last updated on 2026-03-31 | Edit this page

Estimated time: 5 minutes

Overview

Questions

  • What actions I take might interfere with what other people are doing on the system?

Objectives

  • Not cause problems to other users

HPC systems are powerful machines composed of many parts.

The uses on a HPC system share resources. Part of these resources are managed by a resource manager like SLURM (for computing nodes), or by a system of quotas (for disk space), but other parts are (almost) left to the sense of the users.

Working on the Login Node

Problems happen from time to time when users, on the login node, abuse the system by invoking commands that take too many resources.

While the login nodes of HPC clusters tend to be more powerful than a typical laptop, there are potentially many users working at the same time.

When working with a shell on a login node, it is deceptively easy to run a command that might require too many resources and make it harder for other people connected to the login node to work normally. They might experience: - simple commands taking much longer to execute than normal; - even difficulties typing correctly and receiving feedback while typing

If you run a command that takes more than a couple of minutes or uses many cores, please require a SLURM allocation.

Callout

How can I know if my command is requiring too much resources?

  • Read the documentation of the tools you are using, searching for terms like “multithreading”, “multiprocessing” and “parallelism”.
  • Do a quick test, using top/htop to see how many cores and memory your commands are using.

Also, watch your mailbox: someone might write to you pointing out the problem and a possible cause.

Working with Global Filesystems

Global filesystems (which are typically not the only filesystems on a HPC cluster) has limited capacity, both in terms of bandwidth (amount of data that can be transferred/copied) and in terms of metadata operations (like deletion/creation/moving of files).

For this reason, typically operating on many tiny files on the global filesystem should be avoided.

Typically, the home directories of users are in a global filesystem.

Callout

How to learn more

Attend a course for the HPC system you intend to use, where all the limitations and risks will be explained in concrete, and what you can do to cope with those limits.

Key Points
  • When working on the login node of a HPC system, make sure your commands are not using too many computational resources and do not run for more than a couple of minutes.
  • If you need more power, create an interactive (or batch) allocation with the resource manager of the HPC cluster (e.g., SLURM)
  • Performing too many file operations might as well cause slowdowns for other users attempting to use the filesystem.
  • Performing too many file operations might be inefficient