基于自适应功耗管理的高性能计算机作业调度策略的研究  

Research of Job Scheduling Strategy of High-performance Computer Based on Adaptive Power Management

作　　者：王洁[1] 曾宇[2] 

机构地区：[1]首都师范大学管理学院,北京100089 [2]北京市计算中心,北京100094

出　　处：《计算机科学》2012年第10期313-317,共5页Computer Science

基　　金：国家863计划(2011AA0405)资助

摘　　要：作业调度系统是高性能计算机的核心组件,其目标是在满足性能要求的前提下,使得所有任务消耗的总功耗最低。提出了一种自适应功耗管理策略,该策略采用遗传算法作为功耗调度算法,采用作业队列的能效比作为调度因素,与面向资源效率的传统作业调度算法相比,在确保提升资源利用率、减少资源碎片、提升作业吞吐率、减少饥饿作业的前提下,大幅提升了系统的能效比。实验证明该策略能有效提高整机能效,与传统作业调度策略相比能节约9%以上的能耗。The job scheduling system is a core component of high-performance computer,and the goal of job scheduling is to meet the performance requirements under the premise,and get the lowest total power consumption of all tasks.We presented a job scheduling strategy based on adaptive power management.The strategy is based on genetic algorithm,and takes the performance/power ratio of the job queue as the scheduling factor.Comparing with the traditional job scheduling algorithms,it largely increases the system＇s energy efficiency with ensuring the resources utilization rate and the job throughput as well as decreasing the resources pieces and the pending jobs.The experiments show that this strategy can effectively improve productivity,and reduce energy consumption about 9% compared with traditional stra-tegy.

关 键 词：自适应功耗管理 作业调度 遗传算法 高性能计算 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]