Author: Brian O'Shea, [email protected] (CMSE and Physics+Astronomy)
This tutorial requires use of Python 3 and mpi4py, which is the set of Python bindings to the Message Passing Interface (MPI) library. The HPCC does not have mpi4py installed, but you can load up a version of Python that includes mpi4py by first logging onto one of the development nodes of the HPCC (e.g., dev-intel16, dev-intel18, dev-amd20), and then typing the following at the command prompt:
source ~oshea/python_MPI_setup.sh
This shell script will unload the system Python modules and will load
up the Anaconda python distribution
with Python 3.8, NumPy, Matplotlib, SciPy, and many other packages, as
well as mpi4py. (You can see what this script actually does by typing
cat ~oshea/python_MPI_setup.sh .)
If you want to install the Anaconda Python distribution and mpi4py in
your own home directory on the HPCC, there is a file called
INSTALLATION.md in this repository.
Look at the examples in the directory mpi_examples, which are heavily annotated to explain what they are doing. You can run them on the development nodes with:
mpirun -np 4 python my_example_script.py
where -np 4 means to use 4 MPI tasks (this can be changed,
but people often choose to use powers of two).
- Watch this video that introduces the components of a computer.
- Read through the file
supercomputing_and_MPI_notes.pdf, which is included in this repository.
mpi4py examples (GitHub repository) -- very useful mpi4py examples!
Livermore Parallel Computing tutorial
Livermore MPI tutorial (C++/Fortran)
Introduction to Scientific Computing book by Victor Eijkhout at TACC
Parallel Programming in MPI and OpenMP, also by Victor Eijkhout