软件定义网络中面向时延和负载的多控制器放置策略

在多控制器管理的软件定义网络(SDN)中，时延和负载是控制器放置问题(CPP)要考虑的重要因素。该文以降低控制器之间的传播时延、流请求的传播时延和排队时延、均衡控制器间负载为目标，提出一种控制器放置及动态调整的策略，其中包括用于初始控制器放置的负载均衡算法(BCRA)和遗传算法(GA)，用于动态调整控制器负载的在线调整算法(ADOA)。以上算法均考虑网络连通性。仿真结果表明:在初始控制器放置时，在保证流请求的传播时延、排队时延和控制器传播时延较低的情况下，BCRA部署在中小型网络中时，其负载均衡性能与GA相近且优于k-center和k-means算法；GA部署在大型网络中时，与BCRA, k-center和k-means算法相比，使得负载均衡率平均提高了49.7%。在动态情况下，与现有动态调整算法相比，ADOA可以保证较低排队时延和运行时间的同时，仍能使负载均衡参数小于1.54。

Multi-controller Placement Strategy Based on Latency and Load in Software Defined Network

In Software Defined Networks (SDN), latency and load are important factors for Controller Placement Problem (CPP). To reduce the transmission latency between controllers, the propagation latency and queuing latency of flow requests, and balance the controller load, a strategy on how to place and adjust the controller is proposed. It mainly includes Genetic Algorithm (GA) and Balanced Control Region Algorithm (BCRA) which are used to place the initial controller and one Algorithm of Dynamic Online Adjustment (ADOA), that is an online adjusting algorithm in term dynamic controlling. The above algorithms are all based on the network connectivity. The simulation results show that in initial controller placement situation, under the premise of guaranteeing the lower propagation latency, queue latency and controller transmission latency of flow request, when BCRA is deployed in small and medium-sized networks, its load balancing performance is similar to that of GA and superior to k-center and k-means algorithm; When GA is deployed in large networks, compared with BCRA, k-center and k-means, the load balancing rate increases averagely 49.7%. In the dynamic situation, ADOA can guarantee lower queuing delay and running time, and can still make the load balance parameter less than 1.54.