Design And Implementation Of Calculix-based Parallel Computing Platform For Finite Element

With the development of multi-core processors and cloud computing systems,parallel computing is not unpopular research area neglected in the computation field.Finite element analysis is a kind of numerical analyses widely used in modern engineeringfields like physics, aviation, ship design, architecture, etc. Finite element solvers, e.g.,ANSYS, ADINA, ABAQUS and MSC, have become powerful numerical analysis toolsfor numerous applications. High performance computing (HPC), as one of the frontiers ofthe development of computer architecture, have played an important role in major stateissues such as national economy, social construction and researches on big data analyses.It has become an important issue for engineering domain, academic community as well ascomputer science field to study how HPC can be applied to all fields, how the research onfinite element can bring new insights to problem solutions and how to improve theefficiency and accuracy of complicated finite element computing in a context of big data.For the issues mentioned above, we have designed and implemented a parallelcomputing platform based on a finite element solver named Calculix. We have realized atwo-level parallelism, which implement multi-core programming using Calculix’smultithread scheme and construct distributed platform using MPI. Our platform enablesmodeling, analysis and computing of the data entity in real engineering fields. By takingworkload balance among the nodes and task scheduling into consideration, we split gridsthrough regional partition, reduce workload imbalance, realize distributed computingusing MPI paradigm. Further, considering the influences among nodes during computationand their information delivery activities, we manage to conduct interactions among nodes’status information in a timely manner. Thus, a parallel finite-element computing platformis established. Considering its overall design scheme and implementation, such a platformpresents certain pioneering nature and a prospect for development.The experimental results indicate that the proposed parallel FEM framework ispracticable and has important development prospects.