SSRC:时延敏感流的数据源端速率控制算法  

SSRC: source rate control algorithm for delay-sensitive flow in data center network

作　　者：杨洋 曹敏 杨家海[2,3] 车嵘 刘伟[1] YANG Yang;CAO Min;YANG Jiahai;CHE Rong;LIU Wei(School of Information and Communication,National University of Defence Technology,Xi’an 710106,China;Institute for the Network Sciences and Cyberspace,Tsinghua University,Beijing 100084,China;Tsinghua National Laboratory for Information Science and Technology (TNList),Beijing 100084,China)

机构地区：[1]国防科技大学信息通信学院,陕西西安710106 [2]清华大学网络科学与网络空间研究院,北京100084 [3]清华信息科学与技术国家实验室(筹),北京100084

出　　处：《通信学报》2019年第7期14-26,共13页Journal on Communications

基　　金：国家重点研发计划基金资助项目(No.2016YFB0801302);国防科技大学计划科研基金资助项目(No.ZK18-03-59)~~

摘　　要：当前的研究工作针对如何保证时延敏感流的传输时间进行了大量研究,但普遍存在时效性不够强的问题。基于 SDN/OpenFlow 架构,提出了数据源端控制算法 SSRC。该算法依据网络的全局视图,快速定位拥塞可能发生的节点,并及时对目标流的源端速率进行调节,可以缩短算法的响应时间。实验结果表明,与DCTCP 等算法相比,所提算法的流完成时间平均缩短了 75%,且能够保证时延敏感流的传输时间,很好地解决 Incast 问题。Previous work had studied how to ensure the transmission time of delay-sensitive flows, but falled short in its effectiveness for a given period. Motivated by that and based on SDN/OpenFlow framework, a SDN-based source rate control (SSRC) algorithm was proposed. With a global view of network, SSRC can quickly locate the potential conges- tion node, adjust the transmission rate of source and reduce the response time of SSRC. The experiment results show that compared with DCTCP and other algorithms, SSRC can shorten the completion time of flows by 75% in average, which can ensure the transmission time of delay-sensitive and solve the problem of Incast well.

关 键 词：时延敏感 长流 软件定义网络 缓存溢出 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]