Research On Controller Placement And Routing Optimization Strategies In SDN

With the continuous development of applications,the traditional network architecture has been unable to carry explosively increasing network traffic.To solve this problem,a new type of network architecture has emerged,namely Software Defined Network(SDN).The separation of control and forwarding planes,logically centralized control,and open interface between different layers of SDN bring a lot of flexibility to network management.SDN brings many possibilities and also brings many challenges.How to improve service quality by optimizing SDN network has become a current research hotspot.In the controller deployment stage,we quantified the intimacy between nodes and improving the construction of the similarity matrix in the spectral clustering,and proposed a controller deployment algorithm based on node affinity to achieve delay optimization between switches.In addition,considering that single-objective optimization cannot solve the problem of SDN overload.We added the load balancing constraint factor,improved the Infomap algorithm and proposed a deployment strategy that optimizes latency and load.After the controller is deployed,we propose a routing optimization algorithm based on Deep Reinforcement Learning(DRL).This algorithm realizes SDN routing optimization by sensing the underlying network status and performing self-learning.The contribution of the paper is mainly summarized in the following three aspects:(1)Multi-controller deployment algorithm based on node intimacyIn the multi-controller deployment stage,the location of the controller will directly affect the service quality of SDN.Our work aims to minimize the propagation delay in a software-defined wide area network(SD-WAN),and by quantifying the intimate relationship between network topology nodes,we improve the similarity in traditional spectrum clustering algorithms matrix,to realize the sub-domain division and the determination of the position of the controller.The experimental results based on the real network topology Internet2 OS3 E and China Net show that the partition result calculated by the algorithm has a significant effect on the SD-WAN propagation delay,which is better than the existing k-means based controller deployment algorithm and spectral clustering based controller deployment algorithm.(2)Multi-controller deployment algorithm based on community discoveryIn the controller deployment phase,the intimacy-based multi-controller deployment algorithm is designed for single-objective optimization is extremely likely to cause uneven load among sub-domains.The existing multi-objective optimized controller deployment scheme does not take into account the delay between controllers,and the number of sub-domains needs to be determined manually.Given the above deficiencies,we proposed a multi-controller deployment algorithm based on community detection.The algorithm takes delay and load balancing as the optimization goals,improves the Infomap algorithm in community discovery to discover the potential community structure in the SD-WAN topology,and realizes the SD-WAN subdomain division and determining the deployment location of the controller.Experimental results show that the proposed scheme can automatically determine the number of partitions.Besides,this scheme can obtain lower delay and a more balanced load.(3)Route optimization algorithm based on Deep Reinforcement LearningAfter the controller is deployed,because the controller deployment in the static environment proposed in(1)and(2)is difficult to adapt to changes in dynamic flow.Considering that the traditional routing protocol in the SDN controller only considers the shortest path to be prioritized by path,while ignoring the problem of perception of the underlying network situation,this solution starts with the design of the SDN routing protocol and combines DRL technology.A DRL-based routing optimization algorithm(DQSP),this method can quickly sense the underlying network status(including the network status under attack),and intelligently optimize the routing path through interaction with the environment,improving the normal network environment and specific attack environment Download Qo S in SDN to optimize overall network performance.Extensive simulation experiments have been conducted with respect to several network performance metrics,demonstrating that our DQSP has good convergence and high effectiveness.Moreover,DQSP outperforms the traditional OSPF routing protocol,at least 10% relative performance gains in most cases.