Multi-frontal based approach for concurrent finite element analysis

Abstract

With the increased popularity of parallel and vector processors, much research has recently been devoted to the development of efficient algorithms for finite element analysis which exploit the capabilities of these new machines. This paper presents an approach for concurrent finite element analysis employing a ‘multi-frontal scheme’, a concurrent version of the popular frontal solver. The frontal solver coupled with mesh decomposition is well suited for implementation on parallel processors with ease of algorithmic portability across different classes of machines. Unlike the well-known substructure based concurrent approaches, the strategy proposed in this paper does not require the formation of the complete global stiffness matrix of the assigned submesh. Hence, this can be applied more efficiently for very large and complex structures. This multi-frontal based finite element algorithm has been implemented on Magnum IV, a shared memory parallel processing platform. A plate bending problem is taken as a numerical example. The numerical studies in this paper to study the performance of the concurrent code indicate that the proposed method is highly adaptive for concurrent processing.