论文部分内容阅读
在设计大规模的并行应用程序时,如何使多处理器的利用率达到最优,这对程序设计人员来讲是一个很大的挑战。一般说来,由于应用程序在运行时性能上的缺陷,计算资源得不到充分利用。因此,迫切需要对应用程序进行“性能调试”,即在正确性的基础上,通过揭示这些缺陷,对程序进行细调而提高程序性能。在这篇文章里,介绍了一个软件工具包—自动插桩和监测系统(theAutomatedInstrumentationandMonitoringSystem),它集程序插桩、运行监测和性能分析为一体,支持在多处理器上对并行应用程序进行性能评估。文章首先论述了一些建立性能调试工具的基本问题;然后,详细描述AIMS系统的体系结构以及在利用AIMS工具包进行性能调试工具的开发中的经验;最后,使用两个例子详细地描述使用AIMS系统进行性能调试的过程。
How to optimize multiprocessor utilization when designing large-scale parallel applications is a big challenge for programmers. In general, computational resources are underutilized due to application performance run-time defects. Therefore, there is an urgent need for “performance tuning” of an application program, that is, on the basis of correctness, the program performance is improved by revealing the defects and fine-tuning the program. In this article, a software toolkit - theAutomatedInstrumentationandMonitoringSystem - is introduced that combines program instrumentation, operation monitoring, and performance analysis to support performance evaluation of parallel applications on multiple processors . The paper begins with a discussion of some of the basic issues involved in setting up performance tuning tools; then, the architecture of the AIMS system and its experience in the development of performance tuning tools using the AIMS toolkit. Finally, two examples are used to describe in detail the use of the AIMS system Performance tuning process.