摘要
并行程序运行测试耗时长,成为并行软件投入应用的主要瓶颈问题之一。深入研究并行程序的运行本质,指出可以通过并行程序的计算量表达计算时间,提出数学估算模型。所提模型通过数学公式表达并行计算时间函数,求取该函数导数为零时处理器规模值,该值即为并行计算时间最短时处理器规模参考值。所提模型能够极大地减少程序运行次数,并且由于求导计算去掉了单位数据量计算时间和系统软、硬件开销影响,能够实现平台无关性处理。实验结果表明该方法的有效性。
It is time-consuming to test parallel program,which has becomes one of the bottlenecks in applying parallel software.In this pa-per,it intensively studies the essence of parallel program when it is running,and points out that the parallel computing time can be expressed by the amount of computation.Based on this,a mathematical estimation model is proposed.In this model,the paral-lel computing time function is expressed by mathematical formula,and the parallel node size can be calculated when the deriva-tive is zero.The calculated parallel node size is the reference value when the parallel computing time is shortest.This model can greatly reduce parallel program running time,and can also supports platform-independent handling by calculating the derivative,which ignores the influence of the unit data computing time and the running system’s software and hardware.The experimental results indicate the effectiveness of the discussions.
作者
胡悦
闫振
HU Yue;YAN Zhen(Shanghai XingJian College,Shanghai 200072 China;Shanghai Astronomical Observatory Chinese Academy of Sciences,Shanghai 200030 China)
出处
《自动化技术与应用》
2023年第11期73-76,97,共5页
Techniques of Automation and Applications
基金
国家科技部基金项目(2020SKA0120104)。
关键词
并行程序
计算量
数学估算模型
计算时间最短
处理器规模
parallel program
the amount of computation
mathematical estimation model
shortest computing time
computing node size
