Large-scale vibroacoustic simulations using parallel direct solvers for high-performance clusters

The future scenario in computational sciences is the availability of abundant computational resources, thereby enabling high-performance computing. Therefore, the finite element modeling of vibroacoustic problems with large number of degrees of freedom can be efficiently realized in high-performance clusters for solving using parallel sparse solvers.  The contribution addresses the different capabilities of the in-house numerical tool, elPaSo (Elementary Parallel Solver), to solve large-scale finite element models. Parallelization is performed with the distribution of workload among different computing nodes and their threads with MPI (Message Passing Interface) and OMP (Open Multi-Processing). An investigation is performed on a generic model with fluid-structure interaction, where the complexity of the system matrix represents a wide range of practical vibroacoustic models for solving. The model is subjected to various parallel LU factorization routines of MUMPS and PARDISO using an interface from PETSc (Portable, Extensible Toolkit for Scientific Computation) suite. Finally, the performances are evaluated in terms of solver clock, speedup and the number of floating-point operations performed.