Three-Dimensional, Parallel, Hybrid Grid Generation for The ASCI Program

2nd Symposium on Trends in Unstructured Mesh Generation, University of Colorado, Boulder, August 1999

MESHING
RESEARCH
CORNER

Hydrodynamics Methods(XHM) and Computer Methods(XCM), Development Groups, Los Alamos National Laboratory, Los Alamos, NM 87544
het@lanl.gov

Abstract
The DOE Accelerated Strategic Computing Initiative (ASCI) supports the development of a suite of new 3-D computational physics codes to model phenomena related to the maintenance of the US nuclear weapons stockpile.

One of the critical areas that had to be addressed very quickly was the specifications for grid generation and setup capabilities for these new physics codes. The specifications that we came up with covered a lot of territory considering the target physics codes use 3-D multi-material complex moving geometries and every type of grid topology that applies to the multi-material finite volume integration method (Adaptive Mesh Refinement elements, Voronoi/Delaunay elements, unstructured degenerate

Finite Element elements, and general polyhedral elements). In addition the grids are very large (100 million to several billion elements) and distributed across tens of thousands of parallel processors.

In this paper we will describe the grid generation and setup system that has been developed to meet the needs of the ASCI program. The basic description of our system is a distributed relational database management system that services the code's data management needs, a 3-D computational geometry engine, and the hybrid grid generation algorithms. We will focus our description on our grid generation algorithms and their interactions with each other in a distributed computing environment. Our system uses a variety of grid generation algorithms including: 3-D Voronoi/Delaunay, Unstructured Finite Element, Adaptive Mesh Refinement (AMR), arbitrary polyhedral and advancing front. The fundamental concept that allows us to do hybrid grid generation on arbitrarily complex geometries is the interaction of these algorithms within a single parallel grid generation run.

We will also describe in some detail our grid generation parallelization strategy. The goal of our parallelization strategy is to generate very large meshes very quickly. Some grid generation algorithms such as our Voronoi and AMR algorithms are almost perfectly parallel. Some others like the Delaunay algorithm and the advancing front algorithm require very subtle algorithm tuning to work efficiently.

The important bottom line to all of this development is are real application physics codes using our grids for engineering calculations. Our examples will attempt to demonstrate this.