E-Aalo: 面向无先验知识Coflow的高效多级队列调度  

E-AALO:EFFICIENT MULTI-LEVEL QUEUE SCHEDULING FOR COFLOW WITHOUT PRIOR KNOWLEDGE

作　　者：施凌鹏 卢士达 刘云飞 李静[2] Shi Lingpeng;Lu Shida;Liu Yunfei;Li Jing(Information&Communication Company,Shanghai Municipal Electric Power Co.,Ltd.,State Grid,Shanghai 200072,China;College of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,Jiangsu,China)

机构地区：[1]国网上海市电力公司信息通信公司,上海200072 [2]南京航空航天大学计算机科学与技术学院,江苏南京211106

出　　处：《计算机应用与软件》2023年第3期201-209,共9页Computer Applications and Software

摘　　要：过高的Coflow调度时间将会导致数据中心性能下降,如何减小数据中心中语义相关流(Coflow)的平均完成时间是目前数据中心网络流量调度需要解决的重点问题。为了降低先验未知情况下的Coflow调度的时间,提出一种面向无先验知识Coflow的高效多级队列调度E-Aalo(Enhanced-Aalo)。采用Coflow流量放置策略选择合适的节点减少Coflow中需传输的数据量,降低Coflow调度的平均完成时间。通过提前调度多级队列中低优先级队列流量进一步优化Coflow调度的平均完成时间。实验结果表明,与已有的Aalo方法相比,Coflow调度平均完成时间降低了12.28%,更接近先验已知情况下的最优调度。Excessive Coflow scheduling time will lead to data center performance degradation.How to reduce the average completion time of semantically related flows(Coflow)in the data center is a key problem that needs to be solved in data center network traffic scheduling.In order to reduce the Coflow scheduling time under priori unknown situations,an efficient multi-level queue scheduling E-Aalo(Enhanced-Aalo)for Coflow scheduling without prior knowledge is proposed.The Coflow traffic placement strategy was used to select appropriate nodes to reduce the amount of data to be transmitted in Coflow and reduce the average completion time for Coflow scheduling.The average completion time of Coflow scheduling was further optimized by scheduling the low-priority queue traffic in the multi-level queue in advance.Experimental results show that compared with the existing Aalo method,the average completion time of Coflow scheduling is reduced by 12.28%,which is closer to the optimal scheduling under the known prior knowledge.

关 键 词：Coflow调度 多级队列调度 平均完成时间 流量放置 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]