【作者】 李于锋； 张亚林； 赵强；

【Author】 Li Yu-feng Zhang Ya-lin Zhao qiang(Institute of computer application,China Academy of Engineering Physics,mianyang 621900,china)

【机构】 中国工程物理研究院计算机应用研究所；

【摘要】 随着大量高性能计算机的出现,特别是集群系统的广泛应用,使得大规模复杂数值模拟能得以实现,进而必须使用并行技术来开发程序。为了降低高复杂度数值模拟的开发难度,设计了大型有限元并行计算框架PANDA。基于框架开发有限元模拟程序给开发人员屏蔽了复杂的并行实现,比单独开发大大缩短了开发周期。框架的可扩展性依赖于数据结构和并行实现机制,本文描述了大型有限元数值模拟支撑软件框架PANDA在并行方面的数据结构,阐述了框架的并行实现策略,通过算例表明并行实现的正确性和可靠性。更多还原

【Abstract】 The large scale complex numerical simulation could be taken effect along with the appearance of many high performance computer systems,especially with the various clusters application.Thus parallel programming is necessay to do this job.In order to decrease difficultis of the software development,a scalable computing framework PANDA using finite element method is designed.The development period based on the framwork is largly shorter than that starting from scratch,because the coder can development in spite of parallelism.Scalability of the framework is dependent on its data structure and its parallel strategies,this paper describes the data structures which support parallel computing and the parallel method used by this framwork.Numeric experiment illustrates the correctness and credibility.更多还原