Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
Computational Experiments and Testing of a New Distributed Memory Unstructured CFD Solver Developed with Chapel
Submission Author:
Fábio Malacco Moreira , SP
Co-Authors:
Fábio Malacco Moreira, João Luiz F. Azevedo
Presenter: Fábio Malacco Moreira
doi://10.26678/ABCM.COBEM2023.COB2023-2001
Abstract
The Flux Reconstruction (FR) method is a recent approach to compact high-order spatial discretizations that unifies schemes such as the Discontinuous Galerkin (DG), Spectral Volume (SFV), and Spectral Difference (SD) methods into a single mathematical framework. It also allows for a wide range of new scheme variants to be developed inside this framework, and it is generally of simpler computational implementation than its predecessors. The Chapel programming language is also a recent development that integrates parallel programming directives and controls natively into the language. It uses a Partitioned Global Address Space (PGAS) approach in which the data communication in a distributed memory cluster is hidden from the programmer, while still allowing fine control over data locality. The objective of this effort is to build a competitively performant, shared-memory parallel, 3D Euler flow solver in Chapel using the FR method. While 2nd-order Finite Volume (FV) solvers are common and sufficient for many applications, especially steady-state flows, high-order methods are particularly well suited for problems in which spatial resolution is critical, such as Large Eddy Simulations (LES) and vortex-dominated flows. The Chapel programming language offers an opportunity to simplify software development, reducing entry barriers for new researchers, increasing productivity, and, therefore, reducing the time spent on computational optimization. Such a programming environment should also allow for easier scaling of the resulting code to a distributed memory solver. The solver developed in this work has been validated through comparisons of flow solutions to some standard CFD test cases, such as the Ringleb flow and subsonic flows over common research airfoil geometries. This work demonstrates both the capability of the FR method in accurately resolving critical 2D benchmark problems and of the Chapel programming language as a viable alternative to the traditional MPI approach. These experiments served as a proof-of-concept for the current effort to expand the solver's capabilities to 3D flow and distributed memory parallelism.
Keywords
CFD, Chapel, High-order Method, HPC, Flux Reconstruction, Parallel Programming, Aerodynamics, Compressible flow
