SeyedMir/nhbr-coll-benchs

Simple micro-benchmarks for MPI neighborhood collectives

Set of benchmarks to evaluate the performance of MPI neighborhood collectives

Developed these as part of my work on design and implementation of optimized communication pattern and schedules for MPI neighborhood collectives.

• Related PR in MPICH
• Related paper

Build

make

Usage

./run_bench.sh -n 16 rsg-nhbr-allgather

This will run the benckmark with 16 processes and will use a random sparse graph for the process neighborhood topology.

TODO

• Add alltoall benchmarks