Connecting to an HPC system

Last updated on 2025-12-15 | Edit this page

Overview

Questions

How do I connect to an HPC cluster?

Objectives

Be able to use ssh to connect to an HPC cluster
Be aware that some IDEs can be configured to connect using ssh

Connecting to the HPC cluster

We learned in the previous episode that we access a cluster via the login node, but not how to. We will use an encrypted network protocol called ssh (secure shell) to connect, and the usual way to do this is via a terminal (also known as the command line or shell).

The shell is another way to interact with your computer instead of a Graphical User Interface (GUI). Instead of clicking buttons and navigating menus, you give instructions to the computer by executing commands at a prompt, which is a sequence of characters in a Command Line Interface (CLI) which indicates readiness to accept commands.

One of the advantages of a CLI is the ability to chain together commands to quickly create custom workflows, and automate repetitive tasks. A CLI will typically use fewer resources than a GUI, which is one of the reasons they’re used in HPC. A GUI can be more intuitive to use, but you’re more limited in terms of functionality.

There are pros and cons of GUIs vs CLIs (command line interface), but suffice to say that a CLI is how you interact with an HPC cluster.

We’ll first practice a little bit with the CLI by looking at the files on our laptops.

Open your terminal/Git Bash now and you’ll see that there is a prompt. This is usually something like:

BASH

[you@laptop ~ ]$

The information before the $ in the example above shows the logged in user (you) and the hostname of the machine (laptop), and the working directory (~). Your prompt may include something different (or nothing) before the $, and might finish with a different character such as # instead of $.

There is very likely a flashing cursor after your prompt – this is where you type commands. If the command takes a while to run, you’ll notice that your prompt disappears. Once the command has finished running, you’ll get your prompt back again.

Let’s run a couple of commands to view what is in our home directory:

BASH

[you@laptop ~]$ cd

The cd command changes the “working directory” which is the directory where the shell will run any commands. Without any arguments, cd will take you to your home directory. You can think of this like a default setting for the cd command.

Now we’re going to list the contents of our home directory.

BASH

[you@laptop ~]$ ls

OUTPUT

Desktop  Documents  Downloads  Music  Pictures  Public  Videos
[you@laptop ~]$ cd

Your output will look different but will show the contents of your home directory. Notice that after the output from the command is printed, you get your prompt back.

Where is your home directory? We can print the current working directory with:

BASH

[you@laptop ~]$ pwd

OUTPUT

/home/username

Again, your output will look slightly different. When using a UNIX type system such as Linux or macOS, the directory structure can be thought of as an inverse tree, with the root directory / at the top level, and everything else branching out beneath that. Git Bash (or another terminal emulator) will treat your file system similarly, with the hard drive letter showing as a directory so your home directory would be something like:

OUTPUT

/c/users/you/home

Let’s now connect to the HPC cluster, using the username and password emailed to you for this course, i.e. not your normal university username:

Callout

Forgotten your password?

Reset it here using your email address:

https://hpc-training.digitalmaterials-cdt.ac.uk:8443/reset-password/step1

BASH

[you@laptop ~ ]$ ssh yourUsername@hpc-training.digitalmaterials-cdt.ac.uk

If this is your first time connecting to the cluster, you will see a message like this:

OUTPUT

The authenticity of host 'hpc-training.digitalmaterials-cdt.ac.uk (46.62.206.78)' can't be established.
ED25519 key fingerprint is SHA256:..........
This key is not known by any other names.
Are you sure you want to continue connecting (yes/no/[fingerprint])?

This is normal and expected, and just means that your computer hasn’t connected to this server before. Type yes and then you will see something like this:

OUTPUT

Warning: Permanently added 'hpc-training.digitalmaterials-cdt.ac.uk' (ED25519) to the list of known hosts.

At this point you will be prompted for your password. As you enter it you won’t see anything you type, so type carefully.

Once you’ve logged in you should see that your prompt changes to something like:

BASH

yourUsername@login:~$

Bear in mind that other HPC systems are likely to have a slightly different prompt – some examples are given below:

BASH

[yourUsername@login1[clusterName] ~]$
[yourUsername@login1 ~]$
yourUsername@login1 [~] %

Note that while the values are different, it usually follows broadly the same pattern as the prompt on your laptop, showing your username, the hostname of the login node, and your current directory.

The command needed to connect to your local institution’s cluster will look similar to the ssh command we just used, but each HPC system should have documentation on how to get an account and connect for the first time e.g. Manchester’s CSF and the ARCHER2 national cluster.

Your home directory will usually be at a different location:

BASH

yourUsername@login:~$ pwd

OUTPUT

/some/path/yourUsername

and it will contain different files and subdirectories (but might be empty if you’ve not logged in before):

BASH

yourUsername@login:~$ ls

Callout

Git Bash isn’t the only tool you can use to connect to HPC from Windows. MobaXterm is a popular tool with a CLI and some graphical tools for file transfer.

VScode can be configured for ssh, but follow instructions for configuring VScode so it doesn’t hog resources on the login node! e.g. https://ri.itservices.manchester.ac.uk/csf3/software/tools/vscode/

Explore resources on the cluster

A typical laptop might have 2-4 cores and 8-16GB of RAM (memory). You laptop might have a bit more or a bit less, but we’ll use this as a reference to compare with the resources on the HPC cluster.

Challenge

Explore resources on the login node

We have already considered the resources of a typical laptop – how does that compare with the login node?

You should already be connected to the HPC cluster, but if not log back in using:

BASH

[you@laptop:~]$ ssh yourUsername@hpc-training.digitalmaterials-cdt.ac.uk

See how many cores and how much memory the login node has, using the commands below. nproc gives the numer of CPUs, and free -gh shows available memory.

BASH

yourUsername@login:~$ nproc
yourUsername@login:~$ free -gh

You can get more information about the processors using lscpu

Show me the solution

The output from the commands is shown below, which tells us that the login node has 4 cores and 15GB of memory, which is comparable to a typical laptop.

BASH

userName@login:~$ nproc
4
userName@login:~$ free -gh
              total        used        free      shared  buff/cache   available
Mem:           15Gi       401Mi        11Gi       5.0Mi       3.7Gi        14Gi
Swap:            0B          0B          0B

The login node on most clusters will have more CPUs and memory than your laptop, although as we will soon see, the compute nodes will have even more cores and RAM than the login nodes.

As noted in the setup section the cluster used for this course is an exception to this rule, and has very little computing resource.

Challenge

Explore the worker nodes

Run this SLURM command on the login node to display the number of CPUs, and amount of memory (MB) on each of the compute nodes.

BASH

yourUsername@login:~$ sinfo -o "%n %c %m" | column -t

Are all the compute nodes the same?

Show me the solution

Your output should show something that looks a bit like this:

OUTPUT

HOSTNAMES  CPUS  MEMORY
compute01  4     15610

On most clusters you’re likely to see a few different types of compute node with different numbers of cores and memory, e.g.

OUTPUT

HOSTNAMES  CPUS  MEMORY
node1203   168   1546944
node1204   168   1546944
node1206   168   1546944
node1207   168   1546944
node1208   168   1546944
node904    32    191520
node600    24    128280
node791    32    1546104
node868    48    514944
node870    48    514944

Challenge

Compare a laptop, the Login Node and the Compute Node

Consider the output below from a typical HPC cluster, rather than the numbers you got from the cluster used for this training course.

BASH

[username@login1[csf3] ~]$ nproc

OUTPUT

BASH

[username@login1[csf3] ~]$ free -gh

OUTPUT

               total        used        free      shared  buff/cache   available
Mem:           754Gi        50Gi       216Gi       1.2Gi       495Gi       704Gi
Swap:           15Gi          0B        15Gi

Compare your laptop’s number of processors and memory with the numbers from a typical HPC system’s login node (above) and compute nodes (below).

BASH

[username@login1[csf3] ~]$  sinfo -o "%n %c %m" | column -t

OUTPUT

HOSTNAMES  CPUS  MEMORY
node1203   168   1546944  # Units is MB, so  roughly 1.5TB
node1204   168   1546944
node1206   168   1546944
node1207   168   1546944
node1208   168   1546944
node904    32    191520
node600    24    128280
node791    32    1546104
node868    48    514944
node870    48    514944

What implications do you think the differences might have on running your research work on the different systems and nodes?

Show me the solution

Compute nodes are usually built with processors that have higher core-counts than the login node or personal computers in order to support highly parallel tasks. Compute nodes usually also have substantially more memory (RAM) installed than a personal computer. More cores tends to help jobs that depend on some work that is easy to perform in parallel, and more, faster memory is key for large or complex numerical tasks.

Key Points

The ssh protocol is used to connect to HPC clusters
The cluster should have documentation detailing how to connect