多核系统基于遗传算法的实时任务节能映射  

GA-Based Energy-Efficient Real-Time Task Mapping on Multi-Core Processors

作　　者：吴小东[1,2,3] 韩建军[4] 

机构地区：[1]泉州师范学院数学与计算机科学学院,福建泉州362000 [2]福建省大数据管理新技术与知识工程重点实验室,福建泉州362000 [3]智能计算与信息处理福建省高等学校重点实验室,福建泉州362000 [4]华中科技大学计算机科学与技术学院,湖北武汉430074

出　　处：《微电子学与计算机》2015年第7期1-5,10,共6页Microelectronics & Computer

基　　金：国家自然科学基金项目(61472150;61173045);福建省自然科学基金项目(2015J01663);泉州师范学院科研启动基金

摘　　要：电压/频率岛(VFI)设计将多核处理器划分成VFI,各VFI可设置独立的电压/频率,从而能进行灵活的功耗管理.针对VFI多核处理器上的实时任务应用,提出一种基于遗传算法的节能映射GA-Mapping,将实时任务映射到计算核上调度运行,在满足任务实时性的前提下减小能耗.GA-Mapping算法将任务的映射表示为字符编码串,并通过染色体的选择、交叉和变异算子的设计,在遗传迭代的过程中优化任务-计算核的映射,以减小系统能耗.通过实验将提出的算法与现有节能映射算法进行比较,实验结果表明,在节能与任务的可调度性方面,提出的算法均能获得较好的性能.In the VFI-based multi-core system, processing cores are partitioned into VFIs （voltage/frequency islands）, so that the system is able to realize fine-grained power optimization by using a unique voltage/frequency for each VFL In this paper, an energy-efficient mapping algorithm called GA-Mapping is presented, which is based on genetic algorithm, to map real-time tasks onto the processing cores in the VFI-based multi-core system. The proposed mapping algorithm converts the task-core mapping into digit string. Moreover, during the genetic process, the energy consumption is progressively optimized through the proposed selection, crossover and mutation operations. Experimental results demonstrate that compared with the existing energy-efficient mapping algorithms, the proposed algorithm has better performance with respect to energy efficiency as well as schedulability.

关 键 词：多核系统 实时调度 节能调度 遗传算法 

分 类 号：TP316[自动化与计算机技术—计算机软件与理论]