Step 1: Run basic benchmark using azhpc

[ggorman]

See installation and basic benchmark in
https://github.com/devitocodes/azhpc/blob/master/README.md

Requirements:

• Infiniband enabled MPI
• One MPI process per socket
• OMP_NUM_THREADS=number of physical sockets per socket
• Proximity placement to optimise communication with neighbours.
• Don't need parallel/POSIX file system for this POC

[mloubout]

One MPI process per socket

Depends on the nodes, if you run on IB** ones (the fancy HPC ones) these are weird AMD chips with 2 sockets but 30 NUMA(on the IB_120, 15 NUMA on IB_60) domains so lot better to run with one process per NUMA domain. and very little openmp (4 threads per numa domain).

[ggorman]

From my perspective, the point is this issue to 'knowledge capture' these as options/parameters in the first instance. We can always drop then later as specific platforms.