一种近似最小有效瓶颈优先的Coflow调度机制  

An Approximate Smallest-Effective-Bottleneck-First Coflow Scheduling Mechanism

作　　者：李文信 周晓波[2] 徐仁海 齐恒[1] 李克秋[2] LI Wenxin;ZHOU Xiaobo;XU Renhai;QI Heng;LI Keqiu(School of Computer Science and Technology,Dalian University of Technology,Dalian,Liaoning 116024,China;College of Intelligence and Computing,Tianjin University,Tianjin 300350,China)

机构地区：[1]大连理工大学计算机科学与技术学院,辽宁大连116024 [2]天津大学智能与计算学部,天津300350

出　　处：《计算机工程》2019年第10期19-25,32,共8页Computer Engineering

基　　金：国家重点研发计划(2016YFB1000205);国家自然科学基金重点项目(61432002)

摘　　要：针对先验知识未知场景下的Coflow调度问题,提出一种近似最小有效瓶颈优先的Coflow调度方法。通过结合Coflow当前大小和宽度决定Coflow的调度顺序,并区分出流大小以及短与长等特征的Coflow,从而加大调度优化的空间。实验结果表明,与先验知识未知场景下的Aalo方法相比,该方法可使Coflow的平均完成时间降低33.2%,相较于先验知识已知场景下的SEBF方法,Coflow平均完成时间与其仅有7.3%的性能差距。Aiming at the Coflow scheduling problem in the prior knowledge unknown scene,an Approximate Smallest-Effective-Bottleneck-First(A-SEBF)Coflow scheduling method is proposed.Coflow’s scheduling order is determined by combining the current size and width of Coflow,and the Coflow is characterized by large and small flows,as well as features such as fat,short and thin,so as to increase the space for scheduling optimization.Experimental results show that compared with the Aalo method in the prior knowledge unknown scene,the method can reduce the average completion time of Coflow by 33.2%.Compared with the SEBF method in the prior knowledge known scene,the average completion time of Coflow lags only 7.3%in performance.

关 键 词：数据中心 并行计算 Coflow调度 流量调度 近似最小有效瓶颈优先 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]