Access to the supercomputer

Access Ngarrgu Tindebeek

Before connecting, set up Two-Factor Authentication (2FA) in the accounts portal.

Access to the supercomputer is available through Secure Shell (SSH) to nt.swin.edu.au for all users. i.e:

ssh [your-username]@nt.swin.edu.au

Important

After logging in you will be assigned to one of the two login nodes: tooarrana1 or tooarrana2.

From these nodes you can submit jobs to the queue nodes using Slurm. You may also use the login nodes as interactive nodes to run short jobs directly, compile code, or test your application.

Note

Do not use these login nodes to run long jobs or jobs with big computational requirements. Running such jobs in interactive mode can be done by requesting interactive nodes from the queue.

Each login node is also available directly

ssh [your-username]@tooarrana1.hpc.swin.edu.au
ssh [your-username]@tooarrana2.hpc.swin.edu.au

Access OzSTAR

The login nodes for the previous, older, supercomputer OzSTAR (farnarkle1/2) are also still available via the round-robin address

ssh [your-username]@ozstar.swin.edu.au

or directly

ssh [your-username]@farnarkle1.hpc.swin.edu.au
ssh [your-username]@farnarkle2.hpc.swin.edu.au

It is also possible to jump from any login node to any other login node using the hostnames t1, t2, f1, and f2. For example,

# From tooarrana1 to tooarrana2
[tooarrana1]$ ssh t2
[tooarrana2]$

# From tooarrana2 to farnarkle1
[tooarrana2]$ ssh f1
[farnarkle1]$

Access Trevor

The login node for the cloud-based trevor cluster is only accessible by jumping from one of the other login nodes. e.g.

# From tooarrana1 to trevor
[tooarrana1]$ ssh trevor
[trevor]$

Note

Jobs submitted on tooarrana1/2 will run on the new AMD Milan nodes, whereas jobs submitted on farnarkle1/2 will run on the older Intel Skylake nodes. Jobs submitted on trevor will run on the cloud-based trevor nodes.

Access data-mover nodes

The data-mover nodes can be accessed directly or via the login nodes. e.g.

# From local machine to data-mover01
ssh [your-username]@data-mover01.hpc.swin.edu.au

# From tooarrana1 to data-mover02
[tooarrana1]$ ssh data-mover02
[data-mover02]$

# From farnarkle1 to data-mover03
[farnarkle1]$ ssh dm3
[data-mover03]$

Tip

If you want to avoid having to type the rather long [your-username]@ozstar.swin.edu.au every time you want to connect to the supercomputer, you can set a shortcut in your SSH config.

For example, you can set the following in ~/.ssh/config:

Host nt
    HostName nt.swin.edu.au
    User your-username

Once set, you can use this shortcut to SSH into the supercomputer. The following commands will produce the same outcome:

Connect to supercomputer:

ssh [your-username]@nt.swin.edu.au

ssh nt

Copy a local file to supercomputer:

scp myfile.txt [your-username]@nt.swin.edu.au:path/to/location/

scp myfile.txt nt:path/to/location/

Reusing SSH connections (multiplexing)

Enter your 2FA code once, then reuse that connection for other terminals and transfers to the same destination without entering another code.

This works on Linux, macOS, and Windows with the Windows Subsystem for Linux (WSL). On Windows, run every command below inside WSL. Native Windows OpenSSH and PuTTY do not support connection multiplexing.

Set up multiplexing

  1. On your computer, create a folder for shared connections:

    mkdir -p ~/.ssh/sockets
chmod 700 ~/.ssh/sockets

  2. Add the block for your system to ~/.ssh/config.

    Linux with a graphical desktop (recommended). This setting asks you to approve every use of the connection. First install the pop-up helper. On Ubuntu or Debian:

    sudo apt install ssh-askpass

    Then add:

    Host nt ozstar nt.swin.edu.au ozstar.swin.edu.au *.hpc.swin.edu.au
    ControlMaster autoask
    ControlPath ~/.ssh/sockets/%C
    ControlPersist 300

    macOS, Windows (WSL), or Linux without a desktop. Add:

    Host nt ozstar nt.swin.edu.au ozstar.swin.edu.au *.hpc.swin.edu.au
    ControlMaster auto
    ControlPath ~/.ssh/sockets/%C
    ControlPersist 300

Using a shared connection

Connect using your usual SSH command and enter your 2FA code. Later connections using the same SSH client, destination, username, and port reuse that connection instantly. Different login nodes and data-movers each establish their own connection.

With autoask, a small window asks whether to allow the shared connection each time something uses it, including your own first login:

  • Click OK if you have just started that connection yourself.

  • Click Cancel if you have not. That program then has to log in from scratch and will be asked for a new 2FA code.

Warning

While a shared connection is open, any program running under your local user account can use it without a 2FA code, including scripts and AI coding assistants. With auto this happens without any prompt. Lock your computer when leaving it unattended.

The connection stays open for 300 seconds (5 minutes) after the last session closes. If you connect again within 5 minutes, you will not be asked for a 2FA code.

Tip

You can increase ControlPersist to 10m (10 minutes) or 1h (1 hour) if you want the connection to stay open longer between tasks.

If your login fails with Failed to connect to new control master, the pop-up could not be shown. Change autoask to auto in ~/.ssh/config and connect again.

Checking and closing a shared connection

To check if a connection is currently active:

ssh -O check nt

To close the shared connection (note: this will close all active sessions sharing that connection):

ssh -O exit nt

With autoask, you are asked to confirm before it closes.

Here nt is the shortcut host name configured above. If you connected using a full hostname instead, specify the full destination (e.g. ssh -O exit your-username@nt.swin.edu.au). If a connection stops responding after a network change, close it and reconnect.

Client Requirements

Terminal

A terminal is required to access OzSTAR via Secure Shell (SSH) and issue command line instructions.

Operating System

Terminal

Mac

Available by default via Applications > Utilities > Terminal

Linux

Available by default

Windows

The recommended method is to use the Windows Subsystem for Linux (WSL). More information available at the Microsoft documentation. If you are using an older version of Windows that does not provide WSL, you will need to install PuTTY.

X11 Windows Forwarding

If you need to use a graphical screen of your application rather than a terminal, the ability to forward a graphical screen from the application back to your home computer is available via X11 Windows Forwarding.

Note

Graphical workflows are inherently unsuitable the queue-based job submission model of the supercomputer. However, having a graphical window can still be useful for examining the results of your jobs.

With Linux and MacOS this can be done by login from the command line Secure Shell with the -Y option:

ssh -Y [username]@ozstar.swin.edu.au

For MacOS you will first need to install XQuartz.

For modern Windows installations, you can use the Windows Subsystem for Linux (WSL), which provides a Linux command prompt. This is the recommended method. From there, the steps are the same as for Linux. Please refer to Microsoft's documentation on how to configure your system to run Linux GUI applications: https://learn.microsoft.com/en-us/windows/wsl/tutorials/gui-apps

For older Windows installations that do not provide WSL, you will need to install an X11 Server implementation. The following options can be used:

If you are using PuTTY, you will also need to enable X11 forwarding before connecting, This can be done through connection > SSH > X11 by selecting “Enable X11 Forwarding”.

VNC Sessions

In some instances, you may need to run a graphical application that requires a full desktop environment. In these cases, you can use a Virtual Network Computing (VNC) session.

VNC is also more efficient than X11 forwarding, so it can be useful for X11 applications that are slow to forward, especially over high-latency network connections -- any applications that show detailed 2D images or 3D graphics will benefit from it.

1. Setup and launch a VNC server on OzSTAR

Choose a login node to run the VNC server on. You cannot rely on the generic ozstar or nt logins as a destination for remote persistent processes, since they are both round-robin addresses. You need to pick one of tooarrana1/2 or farnarkle1/2 and stick to it.

# Connect via ssh to a login node
ssh [username]@tooarrana1.hpc.swin.edu.au

# Ensure you have no modules loaded
ml purge

# Select up a vnc password
vncpasswd

# Start a vnc server
vncserver

The VNC server will start on the first available port, e.g. :1. It should say something like:

New 'tooarrana1:1 (username)' desktop is tooarrana1:1

Starting applications specified in /home/username/.vnc/xstartup
Log file is /home/username/.vnc/tooarrana1:1.log

You only need to do this once. Running vncserver again will start a new session on the next available port, e.g. :2.

2. Use SSH tunneling to forward a port from your local machine to your VNC server

From your local machine, setup up an SSH tunnel:

ssh -L 5901:localhost:5901 -N [username]@tooarrana1.hpc.swin.edu.au

This will forward port 5901 on your local machine to port 5901 on the login node. If you started the VNC server on a different port, adjust the port number accordingly, e.g. for port :2 use 5902.

3. Connect to your VNC server from your local machine with a VNC client

You can use any VNC client to connect to the VNC server on the login node. Some popular clients include:

  • TigerVNC

  • RealVNC

  • The inbuilt VNC client in MacOS. (You can make use of it via the "Screen Sharing" app, or via Finder -> Go -> Connect to Server, and enter vnc://localhost:5901)

In each client, you just simply connect to localhost:5901 (or whichever port you forwarded) and enter the password you set up with vncpasswd.

You should then have a desktop window from the login node on your local machine. You can run graphical applications from the terminal in this window.

4. Tidying up

When you are finished, remember to stop the VNC server on the login node

vncserver -kill :X

where X is the number of the VNC server you started, e.g. 1 or 2. You can see what servers are running with:

vncserver -list