A dynamic partition weight strategy for accelerating parallel adaptive Cartesian grid generation

The generation of meshes and the Adaptive Mesh Refinement (AMR) have presented considerable challenges in computational fluid dynamics. This paper presents a strategy for automatic adaptive Cartesian grid generation within a multicore parallel framework based on the Dynamic Partition Weight (DPW) method. It integrates the unique features of cells generated before and after each AMR and predicts the number of iterations for each cell. The partition weight of the cell is set in proportion to the number of iterations, and the grid-parallel repartition that considers the partition weight is performed before executing computations that require geometric information retrieval. A number of configurations, including a wing-body, are selected for analysis to evaluate the strategy’s effectiveness. The results indicate that the computational load imbalance is alleviated during the Cartesian grid generation process, significantly reducing time consumption, with an improvement rate exceeding 50 %. For the wing-body case, a 1.37-billion-cell grid is generated in 44.49 s by using 1024 cores with the DPW strategy, demonstrating DPW’s efficiency and strong parallel scalability for Cartesian mesh generation.