Parallel computation methods for large-scale nonlinear CSM

摘要

The analysis of structures undergoing complex inelastic responses to loads, such as those resulting from explosive detonations or high-speed impact, are challenging mechanics problems, which can typically require significant computational resources. The analyses presented here involve large models (up to several million elements) and different types of material models with varying levels of complexity and computational expense. The parallel computational strategy is first described, including an overlapping computation/message passing algorithm and a material-weighting mesh partitioning scheme. These procedures were implemented into a parallel finite element code, ParaAble, developed by the authors and then used for several large-scale applications. Analyses were performed on as many as 1024 processors of Cray T3E, Compaq AlphaServer, IBM SP, and SGI Origin platforms, which showed excellent speedups. The performance demonstrates the ability to efficiently perform such large complex analyses by the use of parallel computing. The analyses also show that for certain multiple material analyses, the material-weighting scheme can greatly reduce parallel load imbalances.