The following contains a documentation on how to use the Benchmarks for GPI-2. These Benchmarks can be used to test bandwidth and latency for different usecases. GPI-2 is an API for asynchronous communication, which implements the GASPI specification. It provides a flexible, scalable and fault tolerant interface for parallel applications. (https://github.com/cc-hpc-itwm/GPI-2)
- GPI-2
- MPI (optional)
$ mkdir build && cd build
$ cmake -DCMAKE_INSTALL_PREFIX=<PATH> -DWITH_MPI=<yes|no> ..
$ make -j
$ make install
$ mpirun -np 2 ./gbs_write_bw
$ gaspi_run -n 2 -m <machine_file> ./gbs_write_bw
Each benchmark provides a specialized set of available parameters. use -h, --help to list them:
$ ./gbs_write_bw --help
Usage: [options]
Options:
-h [--help] Display this help message.
-w [--window-size] arg Number of messages sent per iteration. Default 64.
-s [--min-message-size] arg Minimum message size. Default 1 byte.
-e [--max-message-size] arg Maximum message size. Default (1 << 22) byte.
-b [--single-buffer] Use a single memory allocation for the measurements.
-v [--verify] Check results of the performed operation.
-i [--iterations] arg Number of iterations. Default 10.
-u [--warmup-iterations] arg Number of warmup iterations. Default 10.
--csv Print output in csv format with statistics.
--raw-csv Print the collected raw data without statistics.
If you use GPI-2 Benchmarks in your research or publications, please cite the following paper:
@INPROCEEDINGS{Bartelheimer_2025,
author={Bartelheimer, Niklas J. and Neuwirth, Sarah M.},
booktitle={2025 33rd International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS)},
title={Comprehensive Performance Analysis of Portals4 Communication Primitives on BXI Hardware},
year={2025},
volume={},
number={},
pages={1-8},
keywords={Analytical models;Computational modeling;Bandwidth;Benchmark testing;Libraries;Hardware;Performance analysis;Telecommunications;Electronics packaging;MPI;GASPI;PGAS;BXI;Portals4;Performance Study;Benchmarking;Performance Analysis},
doi={10.1109/MASCOTS67699.2025.11283317}}