云环境下的程序运行性能自适应模拟方法研究  

A Research of Program Performance Adapting Simulation in Cloud Environment

作　　者：周艺华[1] 解高纯 段淑凤[3] 张常有[2] 

机构地区：[1]北京工业大学计算机学院,北京100124 [2]中国科学院软件研究所并行软件与计算科学实验室,北京100190 [3]石家庄铁道大学信息科学与技术学院,石家庄050043

出　　处：《太原理工大学学报》2015年第6期727-734,共8页Journal of Taiyuan University of Technology

基　　金：中国科学院院地合作项目:面向众核平台的高能效大规模图并行算法研究(61379048)

摘　　要：针对Julia程序运行的时限约束,提出一种基于有色Petri网的程序运行性能自适应模拟方法。面向特定领域构建适应底层计算资源特点的工具库,抽取针对Julia语言的计算原语;程序员通过Julia语言实现原语,动态选取领域工具,适应运行时限约束。结合图像对象识别案例,构建Colored Petri Net模型,借助CPN Tools工具模拟运行,自动优化备选方案。模拟实验表明,基于工具库的性能标定,CPN模拟得到的工具库映射方案接近最优。The parallel program developing is difficult on heterogeneous computing platform . Cloud programming in Julia dynamic language is a reasonable model .For time‐constrained cases of the Julia program ,a performance self‐adaptive simulation method based on the Colored Petri Net is proposed in this paper .Oriented to specific field ,such as the case of image object recognition ,we constructed a tool‐library adapting to the characteristics of the underlying computing resources , and extracted a set of primitives in the Julia language .These primitives are implemented in Julia language ,and they select tools dynamically to adapt the time‐constrained running limitation .Con‐sidering the case of image object recognition ,we constructed Colored Petri Net models ,and simu‐lated this model on CPN tools to observe the auto‐optimizing processes .The simulation experi‐mental results show that the tools mapping scheme produced in CPN simulation is close to the best on the basis of the performance calibration of the tools library .

关 键 词：云编程环境 时限约束 性能自适应 CPN模型 性能标定 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]