General-Purpose Parallel Architectures and Routing Schemes

通用并行架构和路由方案

• 批准号: 9321388
• 负责人: Ronald Greenberg
• 金额: $ 16.95万
• 依托单位: University of Maryland, College Park
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1994
• 资助国家: 美国
• 起止时间: 1994-09-15 至 1997-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9321388&HistoricalAwards=false
• 关键词: General; Purpose; Parallel; Architectures; Routing

9321388 Greenberg This research seeks to determine good interconnection networks and routing schemes for massively parallel computers of the future. Such networks must have modestly increasing hardware requirements as the number of processors increases but also must efficiently support general patterns of communication. Hardware cost will be measured by using models that account for real-world constraints such as pin-out and wiring area rather than more naive component counting. Particular attention will be given to ensuring that general-purpose communication capabilities hold up even when interprocessor distances are large and when faults are introduced. This work will employ both mathematical proof and simulation of proposed networks. The identification of a superior interconnection network can increase the programmer's flexibility. On one hand, machine-independent programs should be fairly efficiently executable. On the other hand, programs that choose to exploit specifics of the interconnection network will be more portable and scalable given a robust interconnection scheme that can be retained as the number of processors grows. ***

9321388格林伯格这项研究旨在确定良好的互连网络和路由方案的大规模并行计算机的未来。 这样的网络必须有适度增加的硬件要求，随着处理器的数量增加，但也必须有效地支持一般模式的通信。 硬件成本将通过使用考虑实际约束（如引脚和布线面积）的模型来衡量，而不是通过更简单的元件计数来衡量。 将特别注意确保通用通信能力，即使当处理器间的距离很大，当故障被引入。 这项工作将采用数学证明和模拟拟议的网络。 上级互连网络的识别可以增加程序员的灵活性。 一方面，独立于机器的程序应该是相当有效地可执行的。 另一方面，选择利用互连网络的细节的程序将更加可移植和可扩展，给出了一个鲁棒的互连方案，该方案可以随着处理器数量的增长而保留。 ***