A hierarchical parallel implementation model for algebra-based CFD simulations on hybrid supercomputers
(English) Continuous enhancement in hardware technologies enables scientific computing to advance incessantly and reach further aims. Since the start of the global race for exascale high-performance computing (HPC), massively-parallel devices of various architectures have been incorporated into the...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2022 |
| País: | España |
| Institución: | CBUC, CESCA |
| Repositorio: | TDR. Tesis Doctorales en Red |
| OAI Identifier: | oai:www.tdx.cat:10803/687793 |
| Acceso en línea: | http://hdl.handle.net/10803/687793 https://dx.doi.org/10.5821/dissertation-2117-384828 |
| Access Level: | acceso abierto |
| Palabra clave: | Àrees temàtiques de la UPC::Física 004 531/534 536 |
| Sumario: | (English) Continuous enhancement in hardware technologies enables scientific computing to advance incessantly and reach further aims. Since the start of the global race for exascale high-performance computing (HPC), massively-parallel devices of various architectures have been incorporated into the newest supercomputers, leading to an increasing hybridization of HPC systems. In this context of accelerated innovation, software portability and efficiency become crucial. Traditionally, scientific computing software development is based on calculations in iterative stencil loops (ISL) over a discretized geometry—the mesh. Despite being intuitive and versatile, the interdependency between algorithms and their computational implementations in stencil applications usually results in a large number of subroutines and introduces an inevitable complexity when it comes to portability and sustainability. An alternative is to break the interdependency between algorithm and implementation to cast the calculations into a minimalist set of kernels. The portable implementation model that is the object of this thesis is not restricted to a particular numerical method or problem. However, owing to the CTTC's long tradition in computational fluid dynamics (CFD) and without loss of generality, this work is targeted to solve transient CFD simulations. By casting discrete operators and mesh functions into (sparse) matrices and vectors, it is shown that all the calculations in a typical CFD algorithm boil down to the following basic linear algebra subroutines: the sparse matrix-vector product, the linear combination of vectors, and the dot product. The proposed formulation eases the deployment of scientific computing software in massively parallel hybrid computing systems and is demonstrated in the large-scale, direct numerical simulation of transient turbulent flows. |
|---|