Abstract
Asynchronous many-task runtime systems and MPI+X hybrid parallelism approaches have shown promise for helping manage the increasing complexity of nodes in current and emerging high performance computing (HPC) systems, including those for exascale the increasing architectural diversity of these systems, however, poses challenges for runtimes supporting more homogeneous HPC systems. Performance portability layers (PPL) have shown promise for helping manage this diversity. This paper describes a heterogeneous MPI+PPL task scheduling approach for combining these promising solutions with additional consideration for parallel third party libraries facing similar challenges to help prepare such a runtime for the diverse heterogeneous systems accompanying exascale computing. This approach is demonstrated using a heterogeneous MPI+Kokkos task scheduler and the accompanying portable abstractions [16] implemented in the Uintah Computational Framework, an asynchronous many-task runtime system, with additional consideration for hypre, a parallel third party library. Results are shown for two challenging problems executing workloads representative of typical Uintah applications these results show performance improvements up to 4.4x when using this scheduler and the accompanying portable abstractions [16] to port a previously MPI-Only problem to Kokkos::OpenMP and Kokkos::CUDA to improve complex heterogeneous node use. Good strong-scaling to 1,024 NVIDIA V100 GPUs and 512 IBM POWER9 processor are also shown using MPI+Kokkos::OpenMP+Kokkos::CUDA at scale.
| Original language | English |
|---|---|
| Title of host publication | PEARC 2021 - Practice and Experience in Advanced Research Computing 2021 |
| Subtitle of host publication | Evolution Across All Dimensions |
| Publisher | Association for Computing Machinery |
| ISBN (Electronic) | 9781450382922 |
| DOIs | |
| State | Published - Jul 17 2021 |
| Event | 5th Practice and Experience in Advanced Research Computing Conference: Evolution Across All Dimensions, PEARC 2021 - Virtual, Online, United States Duration: Jul 19 2021 → Jul 22 2021 |
Publication series
| Name | ACM International Conference Proceeding Series |
|---|
Conference
| Conference | 5th Practice and Experience in Advanced Research Computing Conference: Evolution Across All Dimensions, PEARC 2021 |
|---|---|
| Country/Territory | United States |
| City | Virtual, Online |
| Period | 07/19/21 → 07/22/21 |
Funding
This material is based upon work supported by the Department of Energy, National Nuclear Security Administration, under Award Number(s) DE-NA0002375. This research used resources of the Lawrence Livermore National Laboratory. Support for J. K. Holmen and D. Sahasrabudhe comes from the University of Texas at Austin under Award Number(s) UTA19-001215 and a gift from the Intel Parallel Computing Centers Program. We would like to thank LLNL for extended access to Lassen with special thanks to Erik Draeger for post-PSAAP II support. We would like to thank ANL for the opportunity to participate in the Aurora Early Science Program. Additionally, we would like to thank all involved with the CCMSC and Uintah, past and present, with special thanks to Brad Peter-son, Jeremy Thornock, Derek Harris, Oscar Díaz-Ibarra, and Todd Harman for Kokkos-related ARCHES efforts.
Keywords
- Asynchronous Many-Task Runtime System
- Parallelism and Concurrency
- Performance Portability
- Portability
- Software Engineering