CAREER: Online Measurement, Evaluation, and Adaptation of Parallel Applications

职业：并行应用的在线测量、评估和适应

• 批准号: 9703212
• 负责人: Jeff Hollingsworth
• 金额: $ 24.28万
• 依托单位: University of Maryland, College Park
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1997
• 资助国家: 美国
• 起止时间: 1997-03-01 至 2002-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9703212&HistoricalAwards=false
• 关键词: CAREER; Online; Measurement; Evaluation; Adaptation

The goal of this research program is to fundamentally change the activity of tuning parallel and distributed applications, from a manual, user intensive process, to an automatic one. The project will investigate four specific steps towards this goal. First, it will seek ways to not only identify the place where a bottleneck occurs, but also the cause of the bottleneck. Second, it will develop techniques to permit programmers to evaluate different tuning alternatives by predicting the impact specific tuning options will have on a program's execution time. Third, it will expand parallel measurement tools beyond computation and communication to include parallel I/O systems. Fourth, it will investigate ways to close the performance measurement and tuning loop to create self tuning applications that can automatically adapt to their computational environment.

这项研究计划的目标是从根本上改变调优并行和分布式应用程序的活动，从手动的、用户密集型的过程变为自动的过程。该项目将研究实现这一目标的四个具体步骤。首先，它将寻求方法，不仅确定瓶颈发生的地方，而且确定瓶颈的原因。其次，它将开发技术，允许程序员通过预测特定调优选项对程序执行时间的影响来评估不同的调优方案。第三，它将扩展并行测量工具，超越计算和通信，包括并行I/O系统。第四，它将研究关闭性能测量和调优循环的方法，以创建能够自动适应其计算环境的自调优应用程序。