A publication of the Office of Advanced Simulation & Computing, NA-114, NNSA Defense Programs
March 2008
NA-ASC-500-07—Issue 6
Return to this issue’s stories
New Scalable Solver for SIERRA Mechanics
Linear solvers are the central building block of many applications. These solvers must be robust, accurate, scalable, and fast. The complexity arising from the growing number of constraint equations is of particular concern in this regard. These constraints make model building easier, but have been extremely challenging to iterative solvers.
Working with the ASC structural dynamics application team, a new solver has been developed that will address growing problem size and complexity of applications. The new solver is based on an overlapping Schwarz preconditioner (Figure 1), and its computer implementation was facilitated in large part by the parallel linear algebra services available in Trilinos.
On performance test suite problems, such as the one shown in Figure 2, this solver is often faster than previous ones, with speedups in excess of five times on some problems. The solver is more robust than its predecessor CLOP and handles large numbers of constraints quite efficiently, which has been an issue with other solvers. This new solver has been run on a wide suite of test problems with up to 3.5 M degrees of freedom.
The solver has been integrated into the ASC software development cycle and will be included in the SIERRA Mechanics Structural Dynamics module software release planned for April 2008. The SIERRA Mechanics capabilities are widely used at Sandia for thermal and mechanical response predictions on weapon component design and qualification programs, as well as for analyses to assist with SFI closures. The SIERRA Mechanics package is also the Engineering Code cornerstone of the ASC National Code strategy for consolidating and focusing code development efforts within the three NNSA laboratories.
Figure 1. Overlapping grids used in the Schwarz preconditioner
Figure 2. The new solver achieves over 5x improvement for some test problems.
DOE Privacy Disclaimer | Sandia Privacy Disclaimer | SAND 2008-2191 W
Developed and maintained by Sandia National Laboratories for NA.114