Components

the OzSTAR docs are pretty comprehensive and good
you might even find some things that are useful outside of HPC

Nodes

• access points to the HPC

SSH: Accessing OzSTAR

• Secure SHell
• protocoll used for connecting to remote hosts

ssh user@<host>

• remember this command?

quota

• in case quota are exceeded or you need specific custom software
  • use Apptainers
    • isolate anything contained

#uploading big data to OzStar
rsync -avPxH --no-g --chmod=Dg+s <path/to/files> <username>@data-mover01.hpc.swin.edu.au:<path/to/remote/destination>

Modules

• prepackaged stacks of software
• can be loaded and used
• some modules might depend on others
  • loading order matters!

module spider [<pattern>]   #list all modules (following <pattern>)
module load <module name>   #load a module
module list                 #show loaded modules
module avail                #list available modules
module purge                #unload all modules                                                 

#relevant modules for Cn5
module purge
module load gcc/12.2.0          #compiler (required for python)
module load python/3.11.2-bare  #minimal installation of python
module load ipython/9.3.0       #for interactive computing
# module load python-scientific/3.13.1-foss-2025a #scientific python packages                    
module load openmpi/4.1.5       #message passing interface

Let's Set Up a .bashrc on OzSTAR!

• adding commonly used modules
• adding aliases

Screen Sessions

• cheat sheet
• running terminal processes in background

screen -S <name>    #start new session with name `name`                                     
screen -R <name>    #connect if existing, else create new
screen -r <name>    #connect to existing screen
screen -ls          #list active screen sessions
screen -d <name>    #force detach running session
screen -X -S <name> #force kill a session

• keybindings
  • ctrl+a+d ... detach from current session
  • ctrl+d ... quit current session
• try the example scripts: bash_slurm_example1.sh, bash_slurm_example2.sh

even if you're logged out, your screen keeps running

SLURM (Simple Linux Utility for Resource Management)

• HPC resource manager
• highly customizable (by admins)
• mangages
  • job queue
  • resource assignment
  • job monitoring

SLURM Commands

• cheat sheet

sbatch <path/to/slurm/script.sh>                                #launch a job
srun <path/to/slurm/script.sh>                                  #launch parallel job
sinfo [-s] [-N] [-l]                                            #get information about cluster components
squeue [-u <username>]                                          #check the queue
scancel [<job_id>] [-u <username>] [-t PD]                      #kill a running job `-t PD`: cancel all pending jobs
sinteractive --time=0:20:00 --mem=16g --cpus-per-task=8 --x11   #launch interactive session
jobreport <job_id>                                              #current resource usage of job

Let's Play on OzSTAR!

Resource Requests

• every job requires knowledge of
  • memory: peak memory consumption
  • time: total time required for execution
  • cores: maximum number of cores required at the same time
  • temporary memory
• estimating resources
  • memory
    • largest object + buffer
    • tools for estimation: Memory Profiler
  • time
    • longest loop + buffer
    • I/O are detremental to time

Resource Estimate

• for memory checkout Memory Profiler
• from project root run the following:

source .venv/bin/activate   #activate environment
mprof run content/session3_02_hpc_ozstar/01_resource_estimate.py
mprof plot -o "mprofile_plot.png"

• 01_resource_estimate.py contains
  • function to estimate resources
  • main function to be executed

SLURM Scripts

• check this repo for a more comprehensive template

#!/bin/bash

#SBATCH --mail-type=NONE #ALL

#SBATCH --job-name=myjob     #job name

##SBATCH --array=0-2                     #slurm array to execute multiple jobs at once
#SBATCH --output=./execlogs/%x_%a.out
#SBATCH --error=./execlogs/%x_%a.err

#SBATCH --ntasks=1                      #number of tasks per node
#SBATCH --mem=4G                        #total amount of memory per node (in case you are using slurm array)
#SBATCH --time=0-00:05:00               #time limit
##SBATCH --gres=gpu:2                    #request GPUs (amount of GPUs after colon)
##SBATCH --tmp=150GB                     #temporary memory (if large files are acessed, loads of files are read and write)

#load modules
module load python-scientific/3.13.1-foss-2025a #scientific python packages

#run your stuff
# cp /path/to/file.txt $JOBFS             #copy large files to temporary directory

source ~/<path2env>/bash/bin/activate   #activate environment
python3 <path2file.py>                  #run your files
deactivate                              #deactivate environment

# cp $JOBFS/path/to/output.txt /path/to/target/directory #don't forget to copy your results back                                      

Action!

1. Launch Your First Job!
   • you can try the example scripts: bash_slurm_example1.sh, bash_slurm_example2.sh
   • you can create your own little script