About analysis of large problems of structural mechanics on multi-core computers

Abstract

The sparse direct solvers as well as iterative methods are considered for analysis of large-scale finite element problems of structural mechanics on multi-core computers. The parallel preconditioned conjugate gradient method, based on sparse incomplete Cholesky factoring “by value” approach, is presented. The efficiency of block substructure multifrontal method BSMFM, PARFES (Parallel Finite Element Solver), proposed iterative method PSICCG (Parallel Sparse Incomplete Cholesky Conjugate Gradient) and conventional ICCG0 method are compared. PARFES is proved much more effective than multifrontal method. When the amount of RAM is insufficient, the sparse direct methods produce lots of I/O operations, therefore, an iterative method is often more preferable.

Keywords

• finite element method
• sparse direct solvers
• preconditioned conjugate gradient method
• incomplete Cholesky factorization
• multi-core computers