万亿次机群系统高性能应用软件运行现状分析

State-of-art Analysis of High Performance Application Software on Domestic Tera-scale Linux Cluster System

通过调用PAPI(PerformanceApplicationProgrammingInterface)接口函数[6]对2004年3月～4月之间运行在国家应用“973”计划项目LSSC-II万亿次机群系统上部分应用程序进行了跟踪,收集到了大量宝贵的性能数据。依据这些性能数据信息,对我国当前高性能软件的运行情况给出了初步分析。分析结果表明,目前大部分应用程序性能都处于较低水平,并行程序使用处理器的数目范围一般为1～64个,处理器平均效率低于10%,平均性能低于300Mflops。

All the programs running on the first national “973” LSSC-Ⅱ tera-scale Linux cluster system from March to April in 2004 through calling PAPI（performance application programming interface）functions have been tailed after, and lots of important performance data are gathered. This paper presents a primary analysis to high performance application in nowadays in China. The analysis shows that the real performance level of many high performance software is still very low. The amount of CPU used by parallel programs is between 1 and 64. The average efficiency of application is lower than 10%. The average performance of CPU is lower than 300Mflops.