SHF: Small: Adapting Dataflow Analysis for Efficient and Precise Parallel Program Monitoring

SHF：小型：采用数据流分析实现高效、精确的并行程序监控

• 批准号: 1116898
• 负责人: Todd Mowry
• 金额: $ 47.86万
• 依托单位: Carnegie-Mellon University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2011
• 资助国家: 美国
• 起止时间: 2011-08-01 至 2016-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1116898&HistoricalAwards=false
• 关键词: SHF; Small; Adapting; Dataflow; Analysis

Due to the industry-wide shift to multi-core processing, futureapplications will only run faster if they are written as parallelsoftware (i.e. where the work for a single program is accomplished bysimultaneously performing different parts of that work on differentprocessors). Unfortunately history has taught us that writingparallel software is a very error-prone task for programmers. Toaddress this challenge, this research project is creating a powerfulnew mechanism for identifying and hopefully fixing bugs in parallelsoftware on large-scale production systems. Hopefully the resultingframework will not only support the parallel execution of existingmonitoring tools, but it will also spawn the creation of new classesof powerful tools that can recognize subtle bugs in parallel software.Such tools can help programmers continue to reap the performancebenefits of future microprocessors, thereby continuing to enable allof the benefits that increasingly-faster computation has had on theeconomy, science and technology, and our everyday lives in anInternet-connected world.More specifically, the focus of this project is on sophisticatedonline program monitoring tools that model various aspects of programcorrectness at an instruction-by-instruction granularity. To strike apractical balance between performance, precision, and convenience forthe tool developer, this research has developed a novel framework thatintroduces explicit "windows of uncertainty" combined with a newapproach for avoiding a state space explosion in the analysis(inspired by Tarjan's interval analysis approach to dataflow analysis).

由于整个行业都在向多核处理转变，未来的应用程序只有以并行软件的形式编写时才能运行得更快（即单个程序的工作是通过在不同的处理器上同时执行该工作的不同部分来完成的）。 不幸的是，历史告诉我们，编写并行软件对程序员来说是一项非常容易出错的任务。 为了应对这一挑战，该研究项目正在创建一个强大的新机制，用于识别并修复大规模生产系统上并行软件中的错误。 希望最终的框架不仅能支持现有监控工具的并行执行，而且还能催生新的强大工具，这些工具可以识别并行软件中的细微错误。这样的工具可以帮助程序员继续获得未来微处理器的性能优势，从而继续实现越来越快的计算对经济、科学和技术的所有好处，更具体地说，这个项目的重点是简化的在线程序监控工具，这些工具以逐指令的粒度对程序正确性的各个方面进行建模。 为了在性能，精度和工具开发人员的便利性之间取得实际的平衡，本研究开发了一种新的框架，该框架引入了明确的“不确定性窗口”，并结合了一种新的方法，用于避免分析中的状态空间爆炸（受Tarjan的区间分析方法的启发）。