Workshops Archive

Distinguished Speaker Talk: "Developing a task-based compressible Navier-Stokes solver - lessons learned and future developments"


Workshop: The 9th Annual Parallel Applications Workshop, Alternatives to MPI+X (PAW-ATM)

Authors: Mario Di Renzo (University of Salento)

Abstract: Leveraging task-based programming to efficiently execute scientific computing necessitates substantial adaptation of algorithms and software architectures in comparison to standard and established MPI implementations. In this context, the Hypersonic Task-based Research (HTR) solver is one of the examples of software for scientific computing that has adopted a task-based architecture. HTR is a compressible Navier-Stokes solver, and it is specifically designed for the study of reacting multiphysics flows at high Mach numbers. It employs finite-difference algorithms for discretizing the transport equations on curvilinear structured grids. The entire solver operates within the environment provided by the Legion runtime and is routinely utilized on hundreds of GPUs for scientific discoveries related to hypersonic turbulence, electrified combustion, and rocket burner ignition.
Over the years, the HTR solver has undergone significant updates and refactorings to enhance its capabilities and achieve improved performance while maintaining portability and maintainability. This talk aims to delve into some of the design decisions and trade-offs made during this development process. The primary objective is to shed light on the strengths and potential areas for improvement of this type of runtime from the perspective of an end user.


Back to The 9th Annual Parallel Applications Workshop, Alternatives to MPI+X (PAW-ATM) Archive Listing

Back to Full Workshop Archive Listing