Generation and decomposition of finite element models for parallel computations

The utilization of compact parallel computers with distributed memory for finite element computations requires a balanced decomposition of the discretized domain, which has to be performed automatically for geometrically complex problems. The paper summarizes the basic issues related to the parallelization of finite element algorithms and introduces a systematic mesh generation procedure which encompasses the automatic decomposition of the finite element mesh and ensures a balanced distribution of the computational load of the subsequent parallel processing. The decomposition process is performed prior to the ultimate generation of the mesh so that the latter can be carried out in parallel on the created subdomains. (Author)