WARNING: AS OF May 31, 2023 this has not been verified to still be correct!
The goal of this file is to walk you through installing the Anaconda Python distribution on the MSU supercomputer (managed by the Institute for Cyber-Enabled Research, as well as the mpi4py bindings for MPI (which is necessary for running parallel Python codes that use the Message-Passing Interface). The Anaconda Python distribution is an extremely convenient way to get a fully-featured Python programming environment for your computational and data science needs. It includes a huge number of useful Python packages, such as Jupyter, NumPy, SciPy, and Matplotlib. It is also easy to install new packages using conda and/or pip, which is what we will do to install mpi4py.
The primary challenges in installing Anaconda on a supercomputer are (1) the command line environment, which many people are not all that familiar with, and (2) ensuring that you are actually using the correct version of Python.
Before you do anything else, log onto the supercomputer and then onto a development node (dev-intel16, dev-intel18, dev-amd20).
MSU's supercomputer, like most supercomputers, uses a module system to make it easy for users to load and unload different software packages. A variety of them are loaded by default on the development nodes - to see what's loaded, type:
module list
You'll note that there is a Python module loaded by default. This needs to be unloaded, or else you may run into issues later! To do so, you have to do the following:
module unload Python
Now, we're going to get the Anaconda Python distribution. You can do this at the command line using the wget tool. You can find the location of the Linux distribution's downloadable file by looking for the link at the bottom of this page - in the "Linux" section there is a link that says "64-Bit (x86) Installer", and if you copy that link you will get the most current version. Then, at the command line and while you are in your home directory, type:
wget https://repo.anaconda.com/archive/Anaconda3-2022.05-Linux-x86_64.sh
(or replace that URL with an updated version, if you want.) It may take a few minutes for that file to download, but don't worry! wget will give you a progress bar. Then install the file that you have downloaded by typing the following, updating the file name if you have downloaded a newer version:
bash ./Anaconda3-2022.05-Linux-x86_64.sh
For installation it's perfectly reasonable to let Anaconda use all of the defaults, although if you know what you're doing you may wish to NOT allow conda to be your default shell environment. If you say no to that option, you'll have to use a text editor such as nano, vim, or emacs to add this line to the end of the .bashrc file, which lives in your home directory and gets run automatically every time you log in:
export PATH="$HOME/anaconda3/bin:$PATH"
Either way, you can type source ~/.bashrc to set up your Linux
environment so that you can find the Anaconda Python distribution, or
log out of the development node where you have done the installation
(using exit) and then log back in again. If you do the latter,
don't forget to unload the Python module again! (You're going to
have to do this every time you log in, unless you add a line at the
end of your .bashrc file to automatically do it.)
Congratulations! At this point you should have a new Python
distribution. Type python and it should launch a Python
interpreter! Also, if you exit that and type which python, it
should point to the file anaconda3/bin/python in your home
directory.
We'll now use the pip tool to install mpi4py. pip is Python's native package installer, and can be used to install or upgrade almost any Python tool. To install mpi4py, you type the following at the command line:
pip install mpi4py
You can then test to ensure that you have mpi4py installed correctly. At the command line, type:
mpirun -np 4 python ~oshea/MPI_example.py
and it will produce a bunch of output, starting with:
I'm the root task! 0
I'm NOT the root task! 1
I'm NOT the root task! 2
I'm NOT the root task! 3
although not necessary in that order. If you see this, congratulations! You have successfully installed and tested mpi4py. If you do NOT see this, the next step is to debug it based on the outputs you see.
While it's fine to run short computational tasks on the development
node, anything that uses significant resources (CPUs, memory, etc.)
should run using the batch system. You can easily use your new
version of Anaconda in a Slurm batch script. To do so, look at the
file mpi_examples/example_batch_script.sb in this repository, and
note the section that starts with the comment "#### THE LINES BELOW
THIS SET UP THE PYTHON ENVIRONMENT".