Unofficial UCT High Performance Computing Resources

Example scripts and resources for using UCT's High Performance Cluster

Windows Login

SSH using putty

• open putty and enter hpc.uct.ac.za
• login with username and password
• YOU ARE NOW ON THE MASTER NODE! CAREFUL!

Linux Login

SSH using terminal

• open terminal and enter ssh studentno@hpc.uct.ac.za
• login with username and password
• YOU ARE NOW ON THE MASTER NODE! CAREFUL!

Example Usage

Install R/Julia packages

• Open an interactive session on a worker node by running srun --pty --account=stats --time=60:00 bash -l (tip: right click to paste in PuTTY)
• Load R/Julia by running module load software/R-3.x.x or module load compilers/julia-1.1.1
• Open R/Julia Terminal by running R or Julia
• Install R packages using install.packages() function
• Install Julia packages by using ] add

Once the packages have been installed, press CTRL D to terminate the interactive bash session on the worker node.

To reserve more than one worker node, in the first step use 'srun --pty --account=stats --time=08:00:00 --cpus-per-task=40 bash -l'

Prepare Job Script

Create a .sh file with SBATCH directives which will contain all the settings and commands that we want the worker to run. For example:

#!/bin/sh
#SBATCH --account=stats
#SBATCH --partition=ada
#SBATCH --time=24:00:00
#SBATCH --nodes=1
#SBATCH --job-name="surface_plot_alpha_nu"
#SBATCH --mail-user=gntmic002@myuct.ac.za
#SBATCH --mail-type=ALL
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=40

module load compilers/julia-1.1.1
julia -p40 ./hpc_run.jl > out.txt

The first line is the shebang which tells the computer that this is a bash script. The SBATCH lines are arguments for the HPC's job scheduler. We are using the Statistics account, the ada partition (40 cores per node) for 24 hours, only 1 node, my UCT email will be emailed for status updates, we are running only 1 task and reserving 40 cores for this task.

The lines after the #SBATCH lines are the actual bash commands that are run. The first line gets the worker node to load Julia and then the second line opens 40 Julia processes and executes the hpc-run.jl file. Any output that would be printed to the command line is instead saved to out.txt

We save this .sh file in our working directory somewhere. E.g. I saved the .sh file as test-surface.sh in my /home/gntmic002/masters_dissertation/test_scripts folder. The Julia script file I want to run, run_hpc.jl will also be in this folder.

Run Job Script

Now to get the HPC to run our job, we run sbatch test-surface.sh in the test_scripts folder. You should then see a message saying Submitted batch job xxxxxx.

Monitor Job Script

To monitor this job, run squeue -u gntmic002, just replace gntmic002 with your student number. This should list all the running jobs that you have submitted and which node they are running on. You can then go to http://hpc.uct.ac.za/db/ and visually monitor how your assigned node is doing.