基于SDN的IP骨干网流量调度研究与实现
发布时间：2018-11-09 17:49
【摘要】：在传统IP骨干网中一般采取分布式的流量调度方式和静态的路由策略。这些方法缺乏全局视角,导致网络链路的利用率较低和网络拥塞等问题。Google在2012将SDN(Software Defined Network)技术成功引入到IP骨干网中,实现了跨数据中心之间广域网的集中式流量调度机制,使网络中链路利用率提高到95%以上。这为运营商对其电信广域网的流量调度带来了新的思路。本文探索将SDN技术引入到IP骨干网的可行性问题,针对IP骨干网的特殊性,设计了基于SDN的流量调度架构和粒度可变的流量调度算法,有效解决传统IP骨干网中链路利用率低与链路负载不均衡等问题。本文的主要研究成果如下:(1)分析了将SDN引入到IP骨干网所面临的五个挑战:控制器与交换机交互压力成为性能的瓶颈,控制器下发路由表项的不一致性,控制器获取网络信息难度大,底层设备表项容量限制,流量调度算法性能要求高。针对这些挑战,设计了基于路径标识的集中式IP骨干网流量调度架构。该架构的主要特点:1)预分配流表,将全局路由信息预先下发到底层设备,减少控制器与交换机交互信息;2)路径标识,使用全局唯一标识表示全局唯一的路径,解决表项下发不一致问题;3)分类非重叠表项,流表项匹配范围不重叠,并且流表按照匹配规则进行分类,结合SDN南向接口协议实现网络汇聚流监测机制;4)汇聚流粒度调整与流量调度交互执行,汇聚流粒度调整为流量调度提供汇聚流预处理,将汇聚流粒度控制在合理范围,避免表项数量过多,并且提高流量调度执行效率,以及汇聚流迁移成功率。基于Floodlight+Mininet平台,验证了该架构能够完整执行,并正确做出决策。(2)为解决链路拥塞问题,设计了粒度可变的流量调度算法。该算法从调度对象和调度方法两个方面设计,提出汇聚流粒度调整算法和流量调度算法。汇聚流粒度调整算法的作用是控制汇聚流粒度在合理范围。为防止汇聚流粒度过细,提出了汇聚流聚合算法和汇聚流调换算法,为防止汇聚流粒度过粗,提出汇聚流拆分算法。流量调度算法的作用是通过汇聚流迁移降低拥塞链路负载。本文提出三个算法:1)MILP流量调度算法,以最小化重路由业务数量为目标构建混合整数线性规划(MILP)模型;2)分层链路动态上限调度算法,以最小化重路由业务数为目标构建MILP模型,以及通过改变链路容量上限实现负载均衡;3)组合拥塞链路算法,优先迁移流经数条拥塞链路的汇聚流,将拥塞链路上的数条汇聚流均衡分配到数条备选路径。经仿真验证,三个算法在不同的性能上有不同的优势。
[Abstract]:Distributed traffic scheduling and static routing strategy are generally adopted in traditional IP backbone networks. These methods lack a global perspective, resulting in low utilization of network links and network congestion. Google successfully introduced SDN (Software Defined Network) technology into IP backbone network in 2012. The centralized traffic scheduling mechanism of WAN across data centers is implemented, and the link utilization rate in the network is increased to more than 95%. This brings a new idea to the traffic scheduling of telecom wide area network (WAN). This paper explores the feasibility of introducing SDN technology into IP backbone network. According to the particularity of IP backbone network, the traffic scheduling architecture based on SDN and the traffic scheduling algorithm with variable granularity are designed. It can effectively solve the problems of low link utilization and unbalanced link load in traditional IP backbone network. The main results of this paper are as follows: (1) five challenges of introducing SDN into IP backbone network are analyzed: the interaction pressure between controller and switch becomes the bottleneck of performance, and the inconsistency of routing table items under controller is analyzed. It is difficult for the controller to obtain network information, the capacity of the underlying equipment table is limited, and the performance of the traffic scheduling algorithm is high. Aiming at these challenges, a centralized IP backbone network traffic scheduling architecture based on path identification is designed. The main features of the architecture are as follows: 1) preassigned flow table, sending global routing information down to the underlying device in advance, reducing the interactive information between controller and switch; 2) path identification, which uses the global unique identity to represent the globally unique path, which solves the problem of inconsistency of table items; 3) Non-overlapping table items are classified, the matching range of stream table items is not overlapped, and the flow table is classified according to matching rules, and the network convergence flow monitoring mechanism is realized with the combination of SDN southward interface protocol. 4) aggregate flow granularity adjustment interacts with traffic scheduling. Convergence flow granularity adjustment provides convergence flow preprocessing for traffic scheduling, controls aggregation flow granularity within a reasonable range, avoids excessive number of table items, and improves the efficiency of traffic scheduling. And the success rate of convergent flow migration. Based on the Floodlight Mininet platform, it is verified that the architecture can be implemented completely and the decision is made correctly. (2) in order to solve the problem of link congestion, a traffic scheduling algorithm with variable granularity is designed. The algorithm is designed from two aspects of scheduling object and scheduling method, and proposes a convergence flow granularity adjustment algorithm and a traffic scheduling algorithm. The function of the convergence flow granularity adjustment algorithm is to control the convergence flow granularity in a reasonable range. In order to prevent the convergence flow granularity from being too fine, the convergence flow aggregation algorithm and the convergent flow exchange algorithm are proposed. In order to prevent the convergence flow from being too coarse, the convergent flow splitting algorithm is proposed. The function of traffic scheduling algorithm is to reduce the congestion link load by converging flow migration. This paper proposes three algorithms: 1) MILP traffic scheduling algorithm to minimize the number of rerouting traffic as the goal to build a mixed integer linear programming (MILP) model; 2) hierarchical link dynamic upper bound scheduling algorithm, aiming at minimizing the number of rerouting services, constructing MILP model, and realizing load balancing by changing the upper limit of link capacity; 3) combining the congestion link algorithm, the convergent flow flowing through several congestion links is transferred first, and several converging flows on the congestion link are equalized to several alternative paths. Simulation results show that the three algorithms have different advantages in different performance.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.0
本文编号：2321128
[Abstract]:Distributed traffic scheduling and static routing strategy are generally adopted in traditional IP backbone networks. These methods lack a global perspective, resulting in low utilization of network links and network congestion. Google successfully introduced SDN (Software Defined Network) technology into IP backbone network in 2012. The centralized traffic scheduling mechanism of WAN across data centers is implemented, and the link utilization rate in the network is increased to more than 95%. This brings a new idea to the traffic scheduling of telecom wide area network (WAN). This paper explores the feasibility of introducing SDN technology into IP backbone network. According to the particularity of IP backbone network, the traffic scheduling architecture based on SDN and the traffic scheduling algorithm with variable granularity are designed. It can effectively solve the problems of low link utilization and unbalanced link load in traditional IP backbone network. The main results of this paper are as follows: (1) five challenges of introducing SDN into IP backbone network are analyzed: the interaction pressure between controller and switch becomes the bottleneck of performance, and the inconsistency of routing table items under controller is analyzed. It is difficult for the controller to obtain network information, the capacity of the underlying equipment table is limited, and the performance of the traffic scheduling algorithm is high. Aiming at these challenges, a centralized IP backbone network traffic scheduling architecture based on path identification is designed. The main features of the architecture are as follows: 1) preassigned flow table, sending global routing information down to the underlying device in advance, reducing the interactive information between controller and switch; 2) path identification, which uses the global unique identity to represent the globally unique path, which solves the problem of inconsistency of table items; 3) Non-overlapping table items are classified, the matching range of stream table items is not overlapped, and the flow table is classified according to matching rules, and the network convergence flow monitoring mechanism is realized with the combination of SDN southward interface protocol. 4) aggregate flow granularity adjustment interacts with traffic scheduling. Convergence flow granularity adjustment provides convergence flow preprocessing for traffic scheduling, controls aggregation flow granularity within a reasonable range, avoids excessive number of table items, and improves the efficiency of traffic scheduling. And the success rate of convergent flow migration. Based on the Floodlight Mininet platform, it is verified that the architecture can be implemented completely and the decision is made correctly. (2) in order to solve the problem of link congestion, a traffic scheduling algorithm with variable granularity is designed. The algorithm is designed from two aspects of scheduling object and scheduling method, and proposes a convergence flow granularity adjustment algorithm and a traffic scheduling algorithm. The function of the convergence flow granularity adjustment algorithm is to control the convergence flow granularity in a reasonable range. In order to prevent the convergence flow granularity from being too fine, the convergence flow aggregation algorithm and the convergent flow exchange algorithm are proposed. In order to prevent the convergence flow from being too coarse, the convergent flow splitting algorithm is proposed. The function of traffic scheduling algorithm is to reduce the congestion link load by converging flow migration. This paper proposes three algorithms: 1) MILP traffic scheduling algorithm to minimize the number of rerouting traffic as the goal to build a mixed integer linear programming (MILP) model; 2) hierarchical link dynamic upper bound scheduling algorithm, aiming at minimizing the number of rerouting services, constructing MILP model, and realizing load balancing by changing the upper limit of link capacity; 3) combining the congestion link algorithm, the convergent flow flowing through several congestion links is transferred first, and several converging flows on the congestion link are equalized to several alternative paths. Simulation results show that the three algorithms have different advantages in different performance.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.0
【参考文献】
相关博士学位论文 前1条
1 杨仝;骨干网路由表压缩、查找及增量更新技术研究[D];清华大学;2013年
，本文编号：2321128
本文链接：https://www.wllwen.com/guanlilunwen/ydhl/2321128.html
