Performance Studies with COMSOL Multiphysics via Scripting and Batch Processing

Abstract

COMSOL Multiphysics is an extremely powerful and versatile finite element package for the solution of partial differential equations. While the graphical user interface (GUI) offers a friendly environment for solving small problems, for the solution of a more demanding problem with correspondingly larger memory requirements and longer run times, it is often desirable to explore the script-based modeling capabilities, as well as the benefits of running COMSOL in parallel. This work gives step-by-step instructions on how to use m-_les in conjunction with MATLAB as scripting tool and COMSOL's own binary format for batch processing under Linux. We also investigate how to run COMSOL in parallel and report on the shared-memory parallel performance of COMSOL, that is, using all cores available on a compute node. The results show that the speedup is not in proportion to the number of cores used for any of the linear solvers. The results also show that the PARDISO linear solver outperforms all other solvers for our particular test problem.