Lay Summary

Running 3D fluid dynamics simulations on supercomputers becomes inefficient when physical conditions change rapidly, causing uneven workloads across processors. This paper introduces an open-source CFD simulation engine engineered specifically for unpredictable, dynamic high-performance computing loads. By combining flexible mesh generation for complex geometries with real-time dynamic load balancing, the software automatically redistributes tasks among active processors as the physical simulation evolves. This approach ensures maximum computational efficiency, shorter runtime, and better resource utilization for complex engineering and environmental fluid studies. (82 words)

Full Abstract

High-performance computing simulations of transient multi-phase or turbulent flows in irregular 3D geometries often encounter severe dynamic load imbalances as physical localized phenomena evolve. We present an open-source parallel Computational Fluid Dynamics solver built on asynchronous, message-driven parallel execution models. The solver integrates automated dynamic load balancing, unstructured adaptive meshing, and scalable linear solvers designed to handle heterogeneous computational workloads on large supercomputer clusters. Numerical verification tests demonstrate strong parallel efficiency, low communication overhead, and robust load migration under highly dynamic fluid conditions.

Metadata

• Publication Date: 2024
• Author: J. Bakosi
• Journal: SoftwareX / Computer Physics Communications