摘要
针对基于软件定义网络(SDN)的数据中心中老鼠流带宽小、持续时间短且网络占比高、容易导致流表负载不均衡,而大象流容易引起链路负载不均衡的特点,本文提出一种高效的数据中心网络流表与链路联合均衡算法(JLBFTL)。该算法将新到来的流量全默认为老鼠流,提出路径流表评价指标评价并选取路由路径,实现流表负载均衡;周期性监测网络中的流量,对新监测到的大象流,提出路径链路评价指标评价并选取路径,实现链路负载均衡。当网络中有大量突发流量时,可能导致部分链路负载过重,造成链路负载不均衡,此时,本文提出通过选择合适的大象流,利用备份路径和组表进行有效分流实现链路快速均衡。仿真结果表明,本文提出的JLBFTL算法与SRL+FlowFit、L2RM算法相比,在丢包率、带宽利用率和吞吐量方面均有不同程度的改善,提高了网络性能。
In data center networks based on software defined network(SDN),mice-flows have low bandwidth,short duration and high proportion which tends to cause flow table load imbalance.On the contrary,elephant-flows tend to cause link load imbalance.According to the characteristics mentioned above,an efficient joint load balancing algorithm for flow tables and links(JLBFTL)is proposed in data center.In the algorithm,all new flows are regarded as mice-flows,the evaluation index of flow tables is introduced to evaluate and select the path for routing,realizing flow table load balance.Periodically monitoring the flows in the network,and for newly detected elephant-flows,the evaluation index of links is introduced to evaluate and select the path,realizing link load balance.In addition,a large amount of sudden traffic may lead to the overload of some links,resulting in link load imbalance.At this time,this paper proposes to realize link load balance quickly by selecting appropriate elephant-flow and using the backup path and group table for efficient triage.Simulation results show that,compared with SRL+FlowFit and L2RM,the proposed JLBFTL algorithm can reduce packet loss rate,while increase bandwidth utilization and throughput to different degrees,finally improve network performance.
作者
付琼霄
孙恩昌
王倩雯
李萌
张延华
Fu Qiongxiao;Sun Enchang;Wang Qianwen;Li Meng;Zhang Yanhua(Faculty of Information Technology, Beijing University of Technology, Beijing 100124)
出处
《高技术通讯》
EI
CAS
北大核心
2020年第6期579-587,共9页
Chinese High Technology Letters
基金
国家自然科学基金(61671029,61601330,61571021)
国家留学基金委高等学校青年骨干教师出国研修(201806545031)
先进信息网络北京实验室(PXM2019014204500029)
中国博士后科学基金面上项目(2018M640032)
北京市教委科技计划(KM201610005004,KM201610005007)资助项目。
