ON PARELLELIZATION OF LINEAR SYSTEM EQUATION SOLVER IN FINITE ELEMENT SOFTWARE

Abstract eng:
Numerical modelling using high performance computers brings in new opportunities in finite element computations. The contribution focuses on comparing the efficiency of different existing libraries to solution of large, sparse, non-symmetric systems of linear equations, based on direct or iterative algorithms. In particular, the direct solver from SuperLU library (X.S. Li, 1999), iterative solver from Portable, Extensible Toolkit for Scientific Computation (PETSc) are compared. Additionally, the performance and scalability of parallel SuperLU solver is studied, using OpenMP (Chapman, 2008, Barney, 2014) and POSIX threads (Nicols, 1996) based on shared memory model. Also, the scalability of PETSc solver on distributed memory model using Message passing interface (MPI) (Pacheco, 1997) is evaluated. The paper shows that parallelization can efficiently exploit the power of modern available hardware and significantly reduce the computation time.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket