Multithreaded Execution of Object Oriented Programs

面向对象程序的多线程执行

• 批准号: 9902748
• 负责人: Dharma Agrawal
• 金额: $ 18.93万
• 依托单位: University of Cincinnati Main Campus
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1999
• 资助国家: 美国
• 起止时间: 1999-09-15 至 2002-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9902748&HistoricalAwards=false
• 关键词: Multithreaded; Execution; Object; Oriented; Programs

Development of software in OO languages has become verywidespread in every sphere of computing. On the hardware front,multithreaded high-performance workstations and future VLIWmachines or SMP workstations have emerged as an affordablealternative to distributed and parallel computing. Thesemachines, single- or multi-CPU, are often equipped with operatingsystem supported multithreading and are usually connected througha communication network. This proposal intends to develop aplatform for partitioning and scheduling OO programs forconcurrent execution on a multithreaded system. The first step ofproposed work is to perform coarse-grained program partitioningof the flow graph into tasks. We have successfully implementedsuch an experimental framework for programs written in C usingthe Unravel flow graph and Metis graph partitioning package. Thenext step of our approach is to build a framework for runtimeconcurrent execution of the partitioned flow graph based on ourexperience of the Threaded-Executor model. We plan to extendthis implementation to experiment with our recently introducedScalable Task Duplication Based Optimal Scheduling Algorithm. Wealso propose to investigate opportunities for extractinginstruction level parallelism more aggressively for VLIWarchitectures as a result of better static type information andinterprocedural analysis.

用面向对象语言开发软件已经在计算的各个领域变得非常普遍。在硬件方面，多线程高性能工作站和未来的vliwmachine或SMP工作站已经成为分布式和并行计算的一种经济实惠的替代方案。这些机器，单cpu或多cpu，通常配备了操作系统支持的多线程，通常通过通信网络连接。本提案旨在开发一个面向对象程序的分区和调度平台，以便在多线程系统上并发执行。建议工作的第一步是将流图执行粗粒度的程序划分为任务。我们已经成功地在C语言编写的程序中实现了这样一个实验框架，使用了解开流图和Metis图划分包。我们方法的下一步是基于我们对thread - executor模型的经验，为分区流图的运行时并发执行构建一个框架。我们计划扩展这个实现，以实验我们最近引入的基于可扩展任务复制的最优调度算法。我们还建议，作为更好的静态类型信息和过程间分析的结果，研究在vliarchitecture中更积极地提取指令级并行性的机会。