基于大象流两级识别的SDN负载均衡研究
发布时间：2019-06-20 09:18
【摘要】：随着云计算、大数据等业务的兴起,数据中心网络规模及用户数量快速增长,网络流量呈爆发式增加,以TCP/IP架构为核心的传统数据中心网络难以高效管控海量的数据流量,网络阻塞问题加剧,网络资源浪费严重,如何均衡数据中心网络负载以提高数据传输效率成为当前亟待解决的问题。近几年,以控制与转发分离为特征的软件定义网络(Software Defined Networking,SDN)技术凭借其强大的流量管控能力得到了IT界的广泛关注。逻辑集中的SDN控制器基于全局网络视图可实现高效、细粒度的网络流量调度,这使其在网络流量管控方面较以TCP/IP架构为核心的传统网络具有较大的优势。本文从大象流识别的角度,对SDN数据中心网络负载不均衡的问题进行研究。首先,针对现有大象流识别方法识别开销大的问题,提出一种大象流两级识别方法。依据大象流数据量大的特点,该方法在识别第一阶段提出基于TCP发送队列的可疑大象流识别算法(Suspicious Elephant Detection based on Write Queue,SED-WQ),通过监测主机端发送队列缓存中的数据量特征以识别可疑大象流,剔除数据量较小的老鼠流以降低第二阶段控制器的处理开销;依据大象流持续时间长的特点,该方法在识别第二阶段提出基于流持续时间的真实大象流识别算法(Real Elephant Detection based on Duration Time,RED-DT),通过监测网络端可疑大象流的持续时间特征以识别真实大象流,剔除不满足条件的大象流以提高大象流识别准确性。其次,针对SDN数据中心链路负载不均衡的问题,提出一种基于大象流两级识别的SDN网络负载均衡策略(Elephant Load Balancing,ELB)。针对网络中的大象流,采用基于均匀分布的大象流调度算法,利用SDN控制器动态精细地规划大象流的最佳转发路径以保证ELB策略的管控效率;针对网络中的老鼠流,采用基于随机选路的老鼠流调度算法,利用SDN控制器静态粗放地选取老鼠流的最佳转发路径以降低ELB策略的控制器处理开销。最后,利用Mininet软件对所提出的大象流两级识别方法和负载均衡策略ELB进行仿真分析。实验分析表明,在保证大象流识别的高准确性前提下,大象流两级识别方法较基于采样的大象流识别方法可以降低约85%的控制器识别开销;在保证流量识别开销较低的前提下,ELB策略较传统基于等价路由(Equal Cost Multipath Routing,ECMP)的网络负载均衡策略降低约10%的平均传输时延,提升约5%的链路平均利用率。
[Abstract]:With the rise of cloud computing, large data and other services, the network size and the number of users of the data center are rapidly increasing, the network traffic is increasing, and the traditional data center network with the TCP/ IP architecture as the core is difficult to control the mass data traffic efficiently and the network blocking problem is exacerbated, The network resource waste is serious, how to balance the data center network load to improve the data transmission efficiency becomes the current problem to be solved. In recent years, the Software Defined Networking (SDN) technology, which is characterized by the control and forwarding separation, has been widely concerned by the IT community with its powerful flow control ability. The SDN controller in the logic set can realize the high-efficiency and fine-grained network flow scheduling based on the global network view, which has the advantage that the traditional network with the TCP/ IP architecture as the core in the aspect of network flow control. In this paper, the problem of unbalanced network load of SDN data center is studied from the perspective of elephant flow identification. Firstly, aiming at the problem of large identification cost of the existing elephant flow identification method, an elephant flow two-level identification method is proposed. according to the characteristics of large amount of data of the elephant flow, the method provides a suspicious elephant flow identification algorithm (SED-WQ) based on the TCP transmission queue in the first stage, and the data volume characteristic in the queue cache is transmitted through the monitoring host end to identify the suspicious elephant flow, The method comprises the following steps of: removing a mouse stream with a small data volume to reduce the processing cost of the second-stage controller; and according to the characteristic of long duration of the elephant flow, the method provides a real-Elephant Detection based on Duration (RED-DT) based on the flow duration in the second stage. And monitoring the duration characteristic of the suspicious elephant flow at the network end to identify the real elephant flow, and removing the elephant flow which does not meet the condition to improve the identification accuracy of the elephant flow. Secondly, aiming at the problem of unbalanced load of the SDN data center, an SDN network load balancing strategy (ELB) based on two-level identification of the elephant flow is proposed. aiming at the elephant flow in the network, the optimal forwarding path of the elephant flow is dynamically and finely divided by the SDN controller to ensure the control efficiency of the ELB strategy by using an elephant flow scheduling algorithm which is evenly distributed, the rat flow scheduling algorithm based on the random routing is adopted for the rat flow in the network, The optimal forwarding path of the rat stream is selected statically and loosely by the SDN controller to reduce the controller processing overhead of the ELB policy. Finally, using the Minet software, the proposed two-level identification method and load balance strategy ELB are simulated and analyzed. The experimental analysis shows that at the premise of ensuring the high accuracy of the identification of the elephant flow, the method of the two-level identification method of the elephant flow can reduce the controller identification cost of about 85% based on the sampling of the elephant flow identification method, The ELB strategy reduces the average transmission time delay of about 10% compared with the traditional equivalent-based routing (ECMP) network load balancing strategy, and increases the link average utilization rate of about 5%.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.02
[Abstract]:With the rise of cloud computing, large data and other services, the network size and the number of users of the data center are rapidly increasing, the network traffic is increasing, and the traditional data center network with the TCP/ IP architecture as the core is difficult to control the mass data traffic efficiently and the network blocking problem is exacerbated, The network resource waste is serious, how to balance the data center network load to improve the data transmission efficiency becomes the current problem to be solved. In recent years, the Software Defined Networking (SDN) technology, which is characterized by the control and forwarding separation, has been widely concerned by the IT community with its powerful flow control ability. The SDN controller in the logic set can realize the high-efficiency and fine-grained network flow scheduling based on the global network view, which has the advantage that the traditional network with the TCP/ IP architecture as the core in the aspect of network flow control. In this paper, the problem of unbalanced network load of SDN data center is studied from the perspective of elephant flow identification. Firstly, aiming at the problem of large identification cost of the existing elephant flow identification method, an elephant flow two-level identification method is proposed. according to the characteristics of large amount of data of the elephant flow, the method provides a suspicious elephant flow identification algorithm (SED-WQ) based on the TCP transmission queue in the first stage, and the data volume characteristic in the queue cache is transmitted through the monitoring host end to identify the suspicious elephant flow, The method comprises the following steps of: removing a mouse stream with a small data volume to reduce the processing cost of the second-stage controller; and according to the characteristic of long duration of the elephant flow, the method provides a real-Elephant Detection based on Duration (RED-DT) based on the flow duration in the second stage. And monitoring the duration characteristic of the suspicious elephant flow at the network end to identify the real elephant flow, and removing the elephant flow which does not meet the condition to improve the identification accuracy of the elephant flow. Secondly, aiming at the problem of unbalanced load of the SDN data center, an SDN network load balancing strategy (ELB) based on two-level identification of the elephant flow is proposed. aiming at the elephant flow in the network, the optimal forwarding path of the elephant flow is dynamically and finely divided by the SDN controller to ensure the control efficiency of the ELB strategy by using an elephant flow scheduling algorithm which is evenly distributed, the rat flow scheduling algorithm based on the random routing is adopted for the rat flow in the network, The optimal forwarding path of the rat stream is selected statically and loosely by the SDN controller to reduce the controller processing overhead of the ELB policy. Finally, using the Minet software, the proposed two-level identification method and load balance strategy ELB are simulated and analyzed. The experimental analysis shows that at the premise of ensuring the high accuracy of the identification of the elephant flow, the method of the two-level identification method of the elephant flow can reduce the controller identification cost of about 85% based on the sampling of the elephant flow identification method, The ELB strategy reduces the average transmission time delay of about 10% compared with the traditional equivalent-based routing (ECMP) network load balancing strategy, and increases the link average utilization rate of about 5%.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.02
【参考文献】
相关期刊论文 前10条
1 白磊;田立勤;陈超;;基于流抽样和LRU的高速网络大流检测算法[J];计算机应用与软件;2016年04期
2 赵辉;丁鸣;程青松;卢凌;孔晨晟;;SDN与NFV技术在云数据中心的规模应用[J];电信科学;2016年01期
3 周桐庆;蔡志平;夏竟;徐明;;基于软件定义网络的流量工程[J];软件学报;2016年02期
4 刘晓陆;刘渊;王春龙;;一种基于FEFS与CBF的网络大流识别算法[J];计算机工程;2015年09期
5 张朝昆;崔勇;唐，
本文编号：2503106
本文链接：https://www.wllwen.com/guanlilunwen/ydhl/2503106.html
