Porting and Evaluation of a Distributed Task-driven Stencil-based Application

Abstract

Alternative programming models and runtimes are increasing in popularity and maturity. This allows porting and comparing, on competitive grounds, emerging parallel approaches against the traditional MPI+X paradigm. In this work, an implementation of distributed task-based stencil computation is compared with a traditional MPI+X implementation of the same application. The Legion task-based parallel programming system is used as an alternative to MPI, but the underlying OpenMP approach is kept at the subdomain level. Overall results are promising toward making this alternative method competitive to the traditional MPI approach. In future work, extensions to other applications will be explored, as well as the use of GPUs. CCS CONCEPTS• Computing methodologies → Parallel programming languages.

${\tt simwave}$ -- A Finite Difference Simulator for Acoustic Waves Propagation

“…The back-end supports all the concerns related to parallelism, performance, hardware specific optimizations and performance portability. Besides providing a reference implementation which is numerically correct, the baseline (serial) code can also be used as an industry proxy of seismic applications for research in high performance computing (HPC) [27,31,32,33,45]. In its first release, simwave implements three back-ends: sequential C (baseline), OpenMP, and OpenACC.…”

“…A more closely related project is Minimod [26], which implements several solvers for the acoustic wave with constant density, acoustic wave with variable density, acoustic transversely isotropic, and the elastic equation. Minimod can serve both for geophysics research, and for HPC research as in [27,31,32,33,45]. However, Minimod is currently not publicly available by the time of this writing.…”

“…The simwave's code is provided in two forms, a sequential (baseline) and an accelerated implementation for users who need to cope with large problems. The code may also be used as a representative of relevant industrial codes which can serve as benchmark for research on high-performance computing methods, such as in [27,31,32,33,45]. Finally, users and researchers can get in touch with the development team through the simwave's issue page on GitHub (https://github.com/HPCSys-Lab/simwave/issues).…”

Ookami: Deployment and Initial Experiences

“…While armgcc shows a performance improvement over armclang at low thread counts, at higher counts the advantage is mostly eliminated. We also ported [31] Minimod to use the Legion task-based parallel programming framework [2] and compared the result to MPI. We also compared the performance to Summit using only Summit's POWER9 CPUs.…”