# RSH 014 internal: running on a cluster

Examples:
- https://github.com/AaltoSciComp/hpc-examples/blob/master/slurm/pi-openmp.c
- https://aaltoscicomp.github.io/python-for-scicomp/parallel/#mpi

Messages to get across:
- instructions manual for script that is well defined
- "grow" the calculation: don't start immediately with 40 nodes

- Intro and purpose
- You get/have a code, you need cluster. How do you do that interface? What is your strategy?
- What kinds of things can go wrong?
- Request many processors, use few
- Request few processors, use many
- Request many processors but the code does not scale
- Ask for "wrong" queue (sometimes there are many)
- Being unsure about MPI/OMP capabilities
- Hard-coded values in the script
- safe shell settings: make script stop if there in undefined variable

- Program, how to run it
- Run code
- Run code interactively (srun)
- make script, test it
- seff
- change to '-c 2' and '--threads=2'
- change to '--threads=$SLURM_CPUS_PER_TASK' (check variable name)
- Now, this script works with any '-c' option.

- MPI code
- create script that uses it
- OpenMP code
- How does it know how many processors to use?
- OMP_NUM_THREADS
- Find example code online. How do you tell what kind it is?
- is it:
- OpenMP
- MPI
- custom thread/multiprocessing (multiprocessing, threading, parallel)
- dig through docs
- notice it doesn't use words MPI or OpenMP
- Notice some "CPUs" parameter

- How to select how many cores
- hyperthreading

- Radovan shows Slurmbrowser on Norwegian HPC
- actually does not seem to work today but i will show something else

- How to select the amount of memory