ITR: Integrated Design of High-Performance Clustered Computing Systems and Algorithms for Large Scale Finite Element Simulation

ITR：大规模有限元仿真高性能集群计算系统与算法集成设计

• 批准号: 0081791
• 负责人: Graham Carey
• 金额: $ 45.84万
• 依托单位: University of Texas at Austin
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2000
• 资助国家: 美国
• 起止时间: 2000-09-01 至 2004-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0081791&HistoricalAwards=false
• 关键词: ITR; Integrated; Design; Performance; Clustered

For large-scale simulation problems, experience shows that, despite the rapid evolution of microprocessor and parallel computing technology, only a small fraction of peak performance from high-performance computing systems is generally realized. This is so because there are typically critical mismatches between the architecture of high-performance computing systems and the fundamental structure of the target simulations. This is clearly a problem of critical importance to information technology. It represents both an opportunity and a challenge to the disciplines of system-level design, targeting special-purpose architectures for high-performance computing, and algorithm and data structure design for large-scale engineering and software applications. Accordingly, this project will investigate systematic methodologies for designing the systems and algorithms for high performance applications. On the system design side, the project will focus on specialized billion-transistor chip multiprocessor architectures and a hierarchically distributed organization of resources tailored to the needs of the application. On the algorithm and application side, it will focus on finite element simulation applications, including representative problems from semiconductor device design and coupled flow and heat transfer processes.

对于大规模的模拟问题，经验表明，尽管微处理器和并行计算技术快速发展，但通常只实现了高性能计算系统的一小部分峰值性能。这是因为在高性能计算系统的体系结构和目标模拟的基本结构之间通常存在严重的不匹配。这显然是一个对信息技术至关重要的问题。它对系统级设计的学科既是一个机遇，也是一个挑战，目标是高性能计算的专用体系结构，以及大规模工程和软件应用的算法和数据结构设计。因此，该项目将研究为高性能应用程序设计系统和算法的系统方法。在系统设计方面，该项目将重点放在专门的十亿晶体管芯片多处理器架构和根据应用需求定制的分级分布式资源组织上。在算法和应用方面，它将侧重于有限元模拟应用，包括来自半导体器件设计和流动与传热耦合过程的代表性问题。