MPI-Partitioned-Microbenchmarks

These are the MPI Partitioned Microbenchmarks used in the ICCP confrence paper titled "Micro-Benchmarking MPI Partitioned Point-to-Point Communication". This contains a benchmark suite to measure the following point-to-point benchmarks:

• availability.c
• early_bird.c
• overhead.c
• perceived_bandwidth.c

And the following communication patterns:

• Halo3D
• Sweep3D

Build Instructions

The results in the ICPP paper were collected using MPIPCL. As were are not the authors of that library, it is not included in this repository, to recreate the results you can add your copy of that library to src/MPIPCL.

This benchmark has not been tested with MPI native implemenations of MPI Partitioned as they were not sufficiantly mature at the time of writing of this paper. There is currently a branch mpi-native where ongoing work to port these benchmarks to MPI native libraries is currently being conducted, but they have not been fully tested.

The benchmarks can be built like so:

./autogen.sh
./configure CC=<mpicc_path> --prefix=<prefix_path>
make
make install

Run Instructions

We have the following runtime parameters to use our benchmarks

--disable-hot-cache        # Invalidate the CPU cache with each iteration .
-threads <n>               # The number of threads to use, currently this is
                           # equal to the number of partitions.
-mmessage-size <min>:<max> # The message range for the benchmark to use.
-iterations <n>            # The number of iterations
-x <n>                     # The number of warmup iterations for hot-cache
-compute-time <n>          # The amount of compute time
-percent-noise <n>         # The percent noise [0, 100]
-noise-type <noise>        # The noise type [Single, Uniform, Gaussian]

You can run the tests like so:

mpirun -np 2 ./build/bin/perceived_bandwidth --iterations 100
                                             --threads 16
                                             --message-size 1024:4096
                                             --compute-time 100
                                             --percent-noise 4
                                             --noise-type Uniform

Reference

@inproceedings{2022_TEMUCIN_ICPP,
  author = {Yıltan Hassan Temuçin, and Ryan E. Grant, and Ahmad Afsahi}
  title = {{Micro-Benchmarking MPI Partitioned Point-to-Point Communication}},
  year = {2022},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  url = {},
  doi = {},
  booktitle = {{51st International Conference on Parallel Processing}},
  articleno = {},
  numpages = {},
  location = {},
  series = {ICPP '22}
}