patran nastran学习第九届dyna国际用户大会计算代码技术.pdfVIP

The Advantages of HP-MPI for MPP LS-DYNA Yih-Yih Lin Hewlett-Packard 3000 Watervie rkway Richardson, TX 75080 ephone: 1 (978) 841-7650 : The standard portable message-passing library MPI is the software tool that drives the parallelism in MPP LS- DYNA. MPI is required to operate in a complex environment: Currently, the major compu rchitectures lude X86, X86_64, and In Itanium 2; the major operating systems lude Linux, Window, and UNIX; diverse interconnects and switches, using different protocols, are offered by various vendors; and furthermore, in recent years most compu rchitectures have been evolved into multiple-core from single-core architecture. A well- implemented MPI should achieve the following goals in such a complex environment: (1) supporting all major compu rchitectures, operating systems, interconnects and switches; (2) being user friendly; (3) being optimized for the performance of the application. In this paper, an in-depth demonstration, using MPP LS-DYNA, on how HP- MPI achieved these goals is presented. Introduction Parallel applications, such as MPP LS-DYNA, run on a collection of computers that are connected with shared media or switches [1], which are commonly referred as interconnects and switches. Currently, the ma ompu rchitectures lude X86, X86_64, and In Itanium 2, and the major operating systems lude Linux and Windows. Shared memory is a well-known shared media, and a variety of switches, using different communication protocols, are availab