A benchmark code to study the impact of CPU caches on MPI communication routines
Run make to compile the cache_bench executable.
Set the CXX environment variable to point to mpicxx.
The mpi-cache-bench supports only 2 MPI processes. Therefore in order to run the benchmark you need to run the following command:
mpirun -np 2 ./cache_bench 1 8 6M
Where the parameters are respectively:
- number of sockets in the system
- total number of cores in the system
- amount of last level cache for a single CPU