Scalable overset computation between a forest-of-octrees- and an arbitrary distributed parallel mesh

Researchers Hannes Brandt and Carsten Burstedde have introduced a novel algorithm for performing scalable overset computations between fundamentally different mesh types—specifically, a parallel forest of octrees and an arbitrary distributed mesh with unrelated partitioning. This addresses a critical challenge in high-performance computing where simulations often require combining meshes with different structures and resolutions, such as in fluid-structure interaction or adaptive mesh refinement scenarios. The algorithm's key innovation lies in its one-directional approach that efficiently handles the mismatch between mesh partitions while maintaining computational efficiency at extreme scales.

The technical breakthrough centers on using Morton ordering (a space-filling curve) to encode partition boundaries globally with just one Morton index per process. This enables precise, communication-free searching of query points in the partition geometry, allowing the algorithm to organize non-blocking communication only to relevant processes. In subsequent local searches, incoming queries are processed and relevant data returned to their origins. The researchers demonstrated the algorithm's performance and scalability up to 12,288 processes in both 2D and 3D example scenarios, with the approach being generalizable to applications like load balancing and adaptive refinement around intersection areas. This represents significant progress for distributed computing applications requiring mesh coupling.