高性能计算中面向Web的性能工具

Web-oriented performance tool for high performance computing systems

为改善并行应用程序运行效率,对性能数据的采集和分析两个方面进行了研究,分别提出多线程同时采集多种类型的性能数据,以及性能分析应该面向Web的两种思想,并提出了一种对负载均衡程度进行量化的算法,实现了面向Web的性能评价工具。它可以在应用程序运行的同时收集系统负载信息和记录应用程序的行为。用户可以藉此在Internet上任何一台计算机上通过浏览器对性能数据进行多视图、多角度的观察分析,找出性能瓶颈,最终提高应用程序的性能。

Most parallel implementations of performance tools have rather low efficiencies. An efficient performance tool should collect multiple types of performance data, which requires multiple threads, and should be Web-oriented for portability and usability. This paper presents a Web-oriented performance tool which can collect information about both resource utilization and program behavior during runtime and can visualize the information in a browser window with JAVA applets using multiple filters and multiple views. A quantifying algorithm was developed to measure the load balance of a parallel system. With this performance tool, programmers can monitor the runtime state of the application, analyze the relationship between the program processes and system load, and identity performance bottlenecks to improve application performance.