A new row-wise parallel finite element analysis algorithm with dynamic load balancing

Abstract

A parallel scheme is devised to efficiently parallelise all steps of parallel finite element analysis in this study. In addition, this scheme is based on a row-wise matrix distribution. A new row-wise parallel finite element analysis algorithm that exploits the nature of distributed compressed row sparse matrices and multivectors to improve concurrency is developed. A new dynamic load balancing technique has also been devised. The dynamic load balancing technique has been designed specifically to balance the computational workload among processors suitable for the analysis of nonlinear structures. This new algorithm has been implemented in ParaStruc, which is a parallel structural analysis system. Trilinos, a set of parallel numerical libraries developed by researchers in the Sandia National Laboratory has been used to build this algorithm. ParaStruc is a lightweight fully parallelised parallel finite element analysis system, which contains only three classes and a pre-processor. It is shown that this approach produces superior performance in terms of speedup, efficiency, and isoefficiency in the analysis of nonlinear structure response ranges when compared to parallel ABAQUS. The performance and efficiency of this algorithm has been verified with numerical simulations of a 200-metre 50-story 10-frame 10-bay 3D structure subjected to various load levels.