This paper discusses the application of a few parallel preconditioning techniques, which are collected in a recently developed suite of codes Parallel Algebraic Recursive Multilevel Solver (pARMS), to tackling large-scale sparse linear systems arising from real-life applications. In particular, we study the effect of different algorithmic variations and parameter choices on the overall performance of the distributed preconditioners in pARMS by means of numerical experiments related to a few realistic applications. These applications include magnetohydrodynamics, nonlinear acoustic field simulation, and tire design.
Bibliographical noteFunding Information:
This work was supported in part by NSF under grants NSF/ACI-0000443 and NSF/INT-0003274, and in part by the Minnesota Supercomputing Institute.
Copyright 2008 Elsevier B.V., All rights reserved.
- Distributed sparse linear systems
- Magnetohydrodynamic flows
- Nonlinear acoustic field simulation
- Parallel algebraic multilevel preconditioning
- Tire design