多计算机的自动插桩与监测系统

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。

An Automated Instrumentation and MonitoringSystem for Multicomputers

Writing large-scale parallel and distributed scien tific applications that make optimum use of the multiproces-sor is a challenging problem.Typically,computational resource are underutilized due to performance failures in the appli-cation being executed.Performance-tuning tools,which means to achieve the best performance through program tuning,are essential for exposing these performance failures and suggesting ways to improve program performance.A software toolkit that facilitates performance evaluation of parallel applications on multiprocessors,the Automated Instrumentation and Monitoring System(AIMS ),is described in this paper.This tool incorporates the functionality of performance-tuning,the run-time monitoring and the performance analyzing.This paper first addresses fundamental issues in building useful performance-tuning tools,and then describes the architecture of AIMS and the experience with the AIMS toolkit for tuning parallel program.Finally,It illustrates the process of performance tuning using AIMS with two examples.