Analysis of Parallelization for FEM in Bio-Mechanics

摘要

Finite element analysis (FEA) plays a key-role in biomechanical research. Its application ranges from the simulation of tissue properties to the exploration of material formulations for prosthetics. Simulation by means of FEA is a time-consuming process and often requires the usage of high performance computing (HPC) to achieve acceptable running times. A crucial point in HPC is the parallelization of the finite element computations used for the physically-based simulation. As HPC resources are often shared among several facilities, a scheduler unit is often necessary to deploy the computational tasks to the computing clusters. Therefore, a prediction of the speed-up gained through the parallelization on HPC clusters with known physical properties provides a valuable hint for a cost-effective scheduling of computational tasks to an appropriate HPC resource. In the following sections we present an approach of determining the appropriate cluster for the task of solving linear/non-linear finite element problems using iterative methods by estimating the scalability of the FE solver under boundary conditions imposed by the destination computing resource.