Performance Analysis and Visualization for DSM-Based Parallel Programs
-
摘要: 针对具有共享存贮的分布式可扩展机群系统的结构及实现特点,通过实验的方法,对共享存贮器访问, 进程调度, 和同步开销等性能参数进行了特征化.对并行应用程序中影响性能的几个主要因素进行了分析,并给出了运用性能数据及系统性能参数对程序和系统的性能进行分析的方法. 将此方法运用于系统行为与性能的可视化,实现了一个基于分布式共乡存贮环境的并行应用程序的性能可视化工具,实时显示和后处理分析并行程序的行为和性能.Abstract: According to the architecture of the shared memory based network of stations and the implementation of its memory coherence protocol,using experimental methodology to characterized the shared memory access,process schedule,and synchronize overhead. Several parameters which primarily affect the performance of a parallel application are analyzed.Then gave a method of how to analyze the performance of an application running on the system by using our performance method and characteristic parameters.Applying this method to performance visualization of parallel application and system,implemented a performance visualization tool to display and analyze the behavior and performance of a parallel application running on a distributed shared memory based network of workstations.
-
[1] Dybes S G, Zhang Xiaodong. Graph:A tool for visualization and optimize layout in data parallel problems . International conference on Parallel Processing . USA.:The Pennsylvania University, 1995.231~241. [2]Zhang Xiaodong, Nalluri N S, Qin Xiaohan. MIN-Graph:A tool for monitoring and visualizing minbased multiprocess performance[J]. Journal of Parallel and Distributed Computing,1993,18:231~241. [3]Xu Zhichen, Larus J R, Miller B P. Shared-memory performance profiling . 6th ACM Symposium on Principles and Practical Parallel Programming . Nevada:Les Vegas, 1997.18~21. [4]吴幸福. 可扩展的并行性能模型极其应用 . 北京:北京航空航天大学计算机科学与工程系,1996 .
点击查看大图
计量
- 文章访问数: 2492
- HTML全文浏览量: 77
- PDF下载量: 278
- 被引次数: 0