Scaling Large Special Purpose Computers

扩展大型专用计算机

• 批准号: 9201484
• 负责人: Kenneth Steiglitz
• 金额: $ 24.45万
• 依托单位: Princeton University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1992
• 资助国家: 美国
• 起止时间: 1992-06-01 至 1995-11-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9201484&HistoricalAwards=false
• 关键词: Scaling; Large; Special; Purpose; Computers

Steiglitz The research is on the limits of scalable special-purpose architectures for highly parallel computation, especially computation on regular arrays. Important applications of array- based computation can be found in digital signal processing, and in many areas of scientific computation, including many-body problems with gravitational or coulombic interactions, fluid dynamics, and wave propagation. These applications require highly parallel computation either because of the need for real- time operation, or because the total number of operations required is very large. The basic problems being addressed are: 1. The design and reliability analysis of large fault-tolerant arrays, and on-line reconfiguration and error-detection algorithms for these special-purpose architectures; 2. The design and analysis of reliable clocking schemes for very large arrays of processors, especially the comparison of synchronous, self-timed, and hybrid clocking; 3. A study of the general issue of cost-effective scaling of large, special-purpose computer architectures for regular computations; and, 4. Extension of the work to Hierarchical Data Computations, such as those in the fast monopole and multipole methods for many-body problems, where particles are distributed in space, and are clustered by decomposing space into hierarchical domains.

施泰格利茨 研究的是高度并行计算，特别是计算规则数组的可扩展的专用架构的限制。 基于阵列的计算的重要应用可以在数字信号处理和科学计算的许多领域中找到，包括具有引力或库仑相互作用的多体问题、流体动力学和波传播。 这些应用程序需要高度并行计算，因为需要真实的时间操作，或者因为所需的操作总数非常大。 所要解决的基本问题是：1。 大型容错阵列的设计和可靠性分析，以及这些专用结构的在线重构和错误检测算法; 2。 大型处理器阵列的可靠时钟方案的设计和分析，特别是同步，自定时和混合时钟的比较; 3。 一项研究的一般问题的成本效益的规模的大型，专用计算机体系结构的定期计算;和，4。 将工作扩展到分层数据计算，例如用于多体问题的快速多极和多极方法，其中粒子分布在空间中，并通过将空间分解为分层域来聚类。