ENHANCING THE PERFORMANCE OF ITERATIVE SOLVERS IN THE CONTEXT OF STRUCTURAL DYNAMICS

Abstract eng:
Spatial and temporal discretization of partial differential equations, such as the ones occurring from continuum mechanics theory, produce sparse matrices which are dominated by zero elements. These matrices are usually involved in the solution of large linear systems which describe the behaviour of complex structural systems which may exhibit dynamic and/or non-linear properties. For the solution of these systems in a parallel environment, domain decomposition methods are usually employed ([2], [3]) which utilize iterative solvers such as the Krylov subspace solvers. These solvers involve the execution of three time-consuming kernels [1], namely sparse-matrix vector products (SpMV), vector-vector operations (AXPY) and dot products, while for domain decomposition methods, forward and backward substitution is also included. In this work, an acceleration scheme for the SpMV is presented, taking into consideration the specific properties of the sparse matrices that are produced from the spatial and temporal discretization of structural dynamics problems and its numerical performance is showcased. .

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