摘要
在定购高性能计算机时要避免被所宣传的峰值性能误导而忽略实际可获得的应用性能。高性能计算(HPC)气象应用采用的方法是以实测定规模,设计了以实际预报模式为主、基准测试为辅的测试方案,由厂家在规定的时间内,对计算机的应用性能、系统管理及单项性能进行综合的测试。实测表明:测试可准确确定系统的配置规模;对于并行气象应用,向量机的性能是标量机的2-6倍;测出了某些系统的潜在问题。
We should avoid to be misled by advertised peak performance and ignoring actual achieved application performance when buying a high performance computer.The method we adopted to figure out the scale of machine is the means of measuring actual HPC meteorological application.The scheme for test not only consists of actual operational prediction models but also benchmark suits.The regulations specify that vendors should complete a comprehensive test including application performance measuring,system manageable test and micro benchmark within the time-limit.The results of the test have demonstrated that the scale of system configuration can be accurately delimited by test,the performance of vector machine is 2-6 times of scalar's for the parallel meteorological code and some potential defects of the system are detected.
出处
《计算机工程与应用》
CSCD
北大核心
2004年第34期57-59,67,共4页
Computer Engineering and Applications
基金
国家"九五"大中型项目:短期气候预测业务系统(编号:计农经[2000]1748)
国家863高技术研究发展计划:中国气象应用网格(编号:2002AA104210)
关键词
HPC
性能测试
气象应用
HPC,performance measurement,meteorological application
