GNU/Linux ◆ xterm-256color ◆ bash 1913 views

This builds on the previous demo (https://asciinema.org/a/251773), but instead of running stand-alone, we run inside of a SLURM job. We ask for 23 MPI tasks, each of which needs only one core and 512 MB of RAM. We say that the job can complete within five minutes. Finally, we ask for at least three nodes, but we are OK with 23 (in other words, having one task run on each of 23 nodes) if that will get the job done faster.

The Python code lives on shared storage, and SLURM does the work of allocating nodes and running jobs.

The code used in this example is the exact same code that was used in the previous examples. No changes were required to make the code work within SLURM.

The code demonstrated in this asciicast is available here: https://github.com/akkornel/mpi4py