A multi-level parallel solver for rarefied gas flows in porous media

TL;DR

This paper introduces a high-performance, multi-level parallel solver for simulating rarefied gas flows in porous media, enabling direct pore-scale analysis using digital rock images with high computational efficiency.

Contribution

The paper presents a novel multi-level MPI/OpenMP parallelization method for solving Boltzmann model equations in porous media, achieving high efficiency and scalability.

Findings

Parallel efficiency of 94% on 1536 cores in 2D simulations.

Parallel efficiency of 81% on 12288 cores in 3D simulations.

Multi-level parallelization outperforms traditional MPI-only approaches.

Abstract

A high-performance gas kinetic solver using multi-level parallelization is developed to enable pore-scale simulations of rarefied flows in porous media. The Boltzmann model equation is solved by the discrete velocity method with an iterative scheme. The multi-level MPI/OpenMP parallelization is implemented with the aim to efficiently utilise the computational resources to allow direct simulation of rarefied gas flows in porous media based on digital rock images for the first time. The multi-level parallel approach is analyzed in details confirming its better performance than the commonly-used MPI processing alone for an iterative scheme. With high communication efficiency and appropriate load balancing among CPU processes, parallel efficiency of 94% is achieved for 1536 cores in the 2D simulations, and 81% for 12288 cores in the 3D simulations. While decomposition in the spatial space does not affect the simulation results, one additional benefit of this approach is that the number of subdomains can be kept minimal to avoid deterioration of the convergence rate of the iteration process. This multi-level parallel approach can be readily extended to solve other Boltzmann model equations.