Approximate Element Computational Time for Domain Decomposition in Parallel Finite Element Code

This paper discusses a method that is able to approximate the element computational time in a finite element analysis code which can be used for domain decomposition purpose. The method models element computational time as functions of material law and model, element geometry and formulation, element integration scheme, node distribution, element contact, and other finite element approximation techniques. The method is based on multi-parameter linear model where parameters can be derived from measured CPU times of simple and small experiments. This paper only describes the computational cost approximations for solid and shell elements. The paper shows that the cost model works well and has positive impacts on parallel load balancing. Although element computational time of some exceptional material laws and element formulations can not be modeled by this method, the experimental results show that the method can predict the computational time correctly for more than 90% of material models and element formulations used in explicit finite element analysis codes.