FAQ

Who is eligible for an account on the facility?

All Swinburne staff and students are eligible for accounts as are researchers in the field of astronomy and space sciences based at publicly funded institutions within Australia. International collaborators may also apply for accounts but cannot be the lead on a research project.

How do I access the facility?

To access OzSTAR, all access is down through the head node ozstar.swin.edu.au. Upon login, you will be assigned to one of two nodes, namely farnarkle1 and farnarkle2, in a round robin fashion. From there you can submit jobs to the queue. These nodes can be used to run small tasks interactively. However, for larger jobs that require interaction, you should request an interactive node via Slurm (for more information, see Interactive jobs).

For more details, see the page Access to the supercomputer.

What operating system is used?

All nodes of OzSTAR are running CentOS 7 as the main operating system. As such, we advise that you get yourself familiar with the Linux operating system and the use of command line before using the supercomputer.

How do I change my login password?

How can I change my login shell?

You can change your login shell, just type changeShell in your terminal. This information will take a few minutes to propagate, and you will have to login again for it to take effect.

How can I avoid session timeouts?

There are a few ways to do this but a simple method is to go to the .ssh directory on your laptop or desktop machine and create a file called “config”. In that file place the line

ServerAliveInterval 120

and that should do it, although on some machines you may need to also run the command

chmod 600 ~/.ssh/*

to ensure the file has the correct permissions.

How can I find out more about using GPUs?

If you would like to learn more about using graphics programming units (GPUs) in your research a good starting point is the NVIDIA Developer Zone or our HPC/GPU webinar series.

You can also look through slides on Getting Started with the CUDA programming language for GPUs, an introduction to the OpenACC directives for accelerating code with GPUs and the Thrust acceleration library.

These were all presented at a recent CUDA Easy workshop at Swinburne (thanks to Michael Wang, Paul Mignone, Amr Hassan and Luke Hodkinson).

Another great starting point is to go to the GPU Technology Conference website and search through past presentations using their On-Demand tool. You can search by field and/or topic, for example, and most likely someone in your field has already tackled what you are hoping to do.

Why am I getting “Disk quota exceeded” message?

Type the following command on any node

quota

and it will highlight which of your project or home quotas you are exceeding.

Why don't some nvidia and slurm commands, or srun/sinteractive gpu jobs work from my screen session?

screen is old and weird and setgid. Linux unsets LD_LIBRARY_PATH for security reasons when running setgid executables, which breaks our pre-loaded slurm and nvidia modules. Interactive slurm jobs started from screen sessions inherit this broken environment.

The simple workaround is to run bash -l or tcsh -l in each screen window you open, or to use tmux instead.

What's with the weird machine names?

All components of the OzSTAR cluster are named in memory of the late satirist, actor, comedian, and writer John Clarke.

Login nodes are farnarkle. login node cgroups are grommet. The main filesysem is Dagg mounted at Fred. Lustre servers are arkle, warble, umlaut. The majority of compute nodes are called John, with high memory nodes being Bryan, and the KNL nodes Gina. Unfortunately the mighty Dave Sorenson does not get a guernsey - he might be out injured.