Parallel Solid Mechanics Codes at Sandia National Laboratories

Abstract Computational physicists at Sandia National Laboratories have moved their production codes to distributed memory parallel computers. The codes include the multi-material CTH Eulerian code (McGlaun et al., 1988, McGlaun et. al. 1990), the ALEGRA arbitrary-Lagrangian-Eulerian (ALE) code (Peery et. al. 1993) and the PRONTO (Flanagan and Taylor, 1987 and Flanagan and Taylor, 1988) structural mechanics code. This presentation discusses our experiences moving the codes to parallel computers and experiences running the codes. Moving large production codes onto parallel computers require developing parallel algorithms, parallel data bases and parallel support tools. We rewrote the Eulerian CTH code for parallel computers. We were able to move both ALEGRA and PRONTO to parallel computers with only a modest number of modifications. We restructured the restart and graphics data bases to make them parallel and minimize the I/O to the parallel computer. We developed mesh decomposition tools to divide a rectangular or arbitrary connectivity mesh into sub-meshes. The sub-meshes map to processors and minimize the communication between processors. We developed new visualization tools to process the very large, parallel data bases. This presentation also discusses our experiences running these codes on Sandia’s 1840 compute node Intel Paragon, 1024 processor nCUBE and networked workstations. The parallel version of CTH uses the Paragon and nCUBE for production calculations. The ALEGRA and PRONTO codes are moving off networked workstations onto the Paragon and nCUBE massively parallel computers.