Petascale Mesh Generation
Parallel unstructured mesh generation has been proposed as a way of overcoming memory limitations of serial mesh generation for large meshes. For example, modeling long-range electro-magnetic effects in the International Linear Collider (ILC) in SciDAC-2 will require meshes four to eight times the size of those currently generated. The use of larger meshes also routinely stresses downstream capabilities like mesh partitioning for parallel analysis and parallel IO. In SciDAC-1, ITAPS researchers developed a prototype CAD-based parallel mesh generation capability and demonstrated superior performance and scaling for SLAC's NLC accelerator design. We propose to upgrade this prototype to a production-class, scalable parallel meshing capability.
In particular, we will update the tetrahedral meshing library used by the current parallel meshing code, to take advantage of adaptive sizing control and robustness improvements (both needed for meshing the highly-curved ILC geometry). We will link parallel mesh generation with mesh partitioning and load balancing capabilities to eliminate the reading and writing of parallel mesh files. We will also extend parallel geometry IO in CGM to treat discrete or facet-based models, which will be needed to represent higher-order geometry in accelerator and front-tracking applications.
Applications such as combustion and biological modeling place extra demands on the geometric models supporting meshing, e.g. for moving boundaries and discrete (facet-based) models. Therefore we also propose to extend parallel geometry IO in CGM to support these types of models, and to access this capability through the ITAPS geometry interface.