Example Benchmarks

In this repository, you find source code examples accompanying our paper submission. We provide complete and executable source codes for our allgatherv, sample sort, and breadth-first search (BFS) examples using:

• Boost.MPI
• KaMPIng
• MPI
• MPL
• RWTH-MPI

Building

Requirements

To compile this project you need:

• A C++17-ready compiler such as g++ version 9 or higher or clang version 11 or higher.
• OpenMPI or Intel MPI
• (Boost) disabled by default, use -DKAMPING_EXAMPLES_USE_BOOST=ON to enable, or fetch from cmake by providing -DKAMPING_EXAMPLES_USE_BOOST_AS_SUBMODULE=ON
• CMake 3.26

For generating job files from experiments suites from experiment_suites/:

• python3
• PyYAML (pip install pyyaml or use the Pipfile provided in kaval/)

Compiling

cmake --preset experiments
cmake --build --preset experiments --parallel

(Note: in the anonymized version, it is quite likely that the source code in this repository will not compile out of the box)

Example Codes/Benchmarks

Note that the parts of the source code which counted towards the reported LOC are marked by //> START ... and //> END (excluding blank and comment lines). Run

cd evaluation
./run_LOC_counting

to obtain the reported lines of code.

1 Allgatherv

The main executable file is exectuables/vector_allgather.cpp. The vector allgather implementation for each binding can be found in include/vector_allgather/.

2 Sample Sort

The main executable file is exectuables/sorting.cpp. The sample sort implementation for each binding can be found in include/sorting/bindings/.

Running

For reproducing our experiments run

python ./kaval/run-experiments.py sorting                 \
              --machine             generic-job-file

this will create a directory containing generic MPI jobfiles for all experiment configurations.

3 BFS

The main executable file is exectuables/bfs.cpp. The core BFS algorithm shared by all implementations can be found in include/bfs/bfs_algorithm. The frontier exchange functionality which is different for each binding/variant can be found in the accordingly named files in include/bfs/bindings/.

Running

For reproducing our experiments run

python ./kaval/run-experiments.py bfs                     \ 
              --machine             generic-job-file

this will create a directory containing generic MPI jobfiles for all experiment configurations.

Making plots

Scripts for parsing log files and generating plots can be found in the evaluation/ subdirectory