Attention: : Confluence is not suitable for the storage of highly confidential data. Please ensure that any data classified as Highly Protected is stored using a more secure platform.
If you have any questions, please refer to the University's data classification guide or contact ict.askcyber@sydney.edu.au
How much memory, walltime and cores should I request for my jobs?
The short answer
Slightly more than what your job requires. This will ensure your job starts as soon as possible. In general, the fewer resources a job requests, the easier it is for the scheduler to squeeze it into the system.
The long answer
You have to work out what your job requires, then request slightly more than this. The rest of this section details some possible strategies to work out how much resources your job uses.
Artemis resource usage logfiles
At the end of every job on Artemis, a resource usage
CPUs or Cores
The number of CPUs you should request depends on how many CPUs your program can use at once. Most off-the-shelf programs come with an option specifying how many CPUs it will use at one time. Sometimes these may be called "threads" instead of CPUs. Whatever option you pick, you must make sure the number of CPUs requested in your program matches the number of CPUs in your PBS script, otherwise you will leave CPUs sitting idle, or you will oversubscribe CPUs, resulting in massive slowdown of your computation.
Another thing to keep in mind when requesting CPUs is that more isn't always better. Depending on how well your program is "parallelised", using more CPUs could actually slow down your code! The best way to find out the optimal number of CPUs for your problem is by trial and error.