An Efficient Scheme for Parallel Parametric-Study in Finite Element Analyses

Parametric studies in parallel-computing is often considered as an "embarrassingly parallel" computation task that can be trivially parallelized and yield good parallel efficiency. However, this paper shows the parallel efficiency of such analyses can be further improved by the proposed scheme in typical cluster setups using commodity PC hardware. Numerical experiments show that the parallel efficiency of such computation task in finite element analyses can be improved by 81.6% and the peak disk usage can be reduced by 83.1%. The proposed scheme for enhancing parallel performance can be readily extended to design optimization problems, which has similar work patterns in parallel processing. Therefore, more efficient engineering analyses and designs can be achieved by employing schemes described in this work.