Performance comparison between two-sided MPI, MPI-3 RMA and OpenSHMEM for a particle cluster algorithm
We present a performance benchmark of a parcel cluster algorithm for N-body simulations that uses either MPI point-to-point, MPI-3 remote memory access (RMA) or OpenSHMEM communication for the nearest-neighbour graph splitting step. The benchmarks are performed on ARCHER2 (Slingshot 10 interconnect), Cirrus (InfiniBand interconnect) and hotlum (Slingshot 11 interconnect).
Institutions
- University of Leeds (GB)
- University of St Andrews (GB)
- Mathematical Institute of the Slovak Academy of Sciences (SK)
- Hewlett Packard Enterprise (United States) (US)
- El Paso Community College (US)
- University of Edinburgh (GB)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2025-05-14
- DOI
- https://doi.org/10.5281/zenodo.15497211
- Primary Topic
- Interconnection Networks and Systems
- Type
- article
- Field-Weighted Citation Impact
- 0.00