Parallel Meshing
This panel will address several questions regarding the Catch-22 of modern simulation: how the meshing community can catch up to solvers that already leverage massive parallelism and GPU support. With the exception of smoothing, most meshing operations require dynamic data structures that do not lend themselves to traditional “solver-style” parallel speedups. Do we need better computer science or different meshing algorithms that are more “parallel friendly?” Parallel mesh generation is hard and requires a different mindset for serial programming. What can we do to help bridge the programming gap? Many aspects of parallel coding mean that adding new functionality can require a major rewrite of the codebase. What are the options to proceed, for example scalability-first designs and the adaptation of functionality-first legacy codes. We will discuss the algorithmic shifts needed to move from 1-billion to 1-trillion cell meshes by 2028 as envisioned by the NASA CFD Vision 2030 roadmap.
Invited Panelists:
- John Verdicchio (Chair), Siemens
- Rao Garimella, Los Alamos National Labs
Metric-Based Anisotropic Mesh Adaptation
This panel will address several questions regarding the application of anisotropic mesh adaptation (AMA) to advance Computational Fluid Dynamics (CFD). We will detail the scientific benefits of metric-based AMA in enabling the certification of numerical solutions and its role as a critical tool for improving turbulence modeling. We will explore the industrial benefits, specifically emphasizing how automation can handle the extreme-scale requirements of the NASA CFD Vision 2030 roadmap. The discussion will clarify the cost of this technology compared to traditional single-simulation best practices. To conclude, we will explore future extensions toward high-order methods and turbulence scale-resolving modeling.
Invited Panelists
- Frederic Alauzet (Chair), Inria Saclay
- Lucien Rochery, FlexCompute
- Xevi Roca, Barcelona Supercomputing Center
- John Verdicchio, Siemens
