Principles and Mechanisms for Asynchronous Parallel Computing

异步并行计算原理与机制

• 批准号: 9522301
• 负责人: Robert Cypher
• 金额: $ 16.42万
• 依托单位: Johns Hopkins University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1996
• 资助国家: 美国
• 起止时间: 1996-06-01 至 1997-05-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9522301&HistoricalAwards=false
• 关键词: Principles; Mechanisms; Asynchronous; Parallel; Computing

9522301 Cypher MIMD (Multiple-Instruction-Stream, Multiple-Data-Stream) parallel computers are naturally asychronous. Many MIMD computers use a separate clock for each processor in order to avoid clock skew problems. Furthermore, even if all the processors use a common clock, the timing of commulication events is not known in advance due to factors such as data dependent processing times, processor interrupts and variable network congestion. This asynchrony has a major impact on the design of the machine's communication sub- system the semantics of the programming model, and the creation of parallel algorithms. This research studies principles for the design and use of asynchronous parallel computers. Specifically, three key areas of asynchronous parallel computing are studied. 1. specifying the semantics of message-passing, shared memory, and other, novel, programming models. 2. Creating efficient communication protocols that implement the above programming models, and 3. evaluating metrics and models for the design of algorithms for asynchronous parallel computers. This research will result in more efficient communication in MIMD machines, novel techniques for reasoning about and using asynchronous parallel computers, and a better understanding of how the theoretical community's body of knowledge in parallel algorithms can be applied to real machines. ***

9522301 Cypher MIMD（Multiple-Instruction-Stream，Multiple-Data-Stream）并行计算机是自然的异步计算机。 许多MIMD计算机为每个处理器使用单独的时钟，以避免时钟偏移问题。 此外，即使所有处理器使用公共时钟，由于诸如数据相关处理时间、处理器中断和可变网络拥塞之类的因素，通信事件的定时也是事先不知道的。 这对机器的通信子系统的设计、编程模型的语义和并行算法的创建都有重大影响。 本研究主要研究异步并行计算机的设计和使用原则。 具体而言，异步并行计算的三个关键领域进行了研究。 1.指定消息传递、共享内存和其他新颖的编程模型的语义。 2.创建实现上述编程模型的高效通信协议，以及 3.评估异步并行计算机算法设计的度量和模型。 这项研究将导致更有效的沟通MIMD机，推理和使用异步并行计算机的新技术，以及更好地了解如何在并行算法的理论界的知识体系可以应用到真实的机器。 ***