The Optimization Strategy Of SDN Network Based On Traffic Flow Division

In the existing hardware centric network architecture,network performance is limited in terms of scalability,reconfiguration and flexibility.The proposal of SDN network architecture realizes the decoupling of plane level by separating the forwarding plane and the control plane,which gives targeted solutions to the above problem.In the SDN architecture,the requests from the upper layer can be transmitted directly to the underling network devices,which can make full use of network residual resources and realize network performance optimization through centralized network state management and the formulation of relevant strategies.As an emerging network architecture technology,SDN faces many challenges.First of all,we focus on the application aware issue in SDN networks.The existing OpenFlow protocol only supports the policy based on the network below 4th layer,which means it is impossible to partition traffic flows at a higher level by traffic type.Then,we focus on the global routing strategy optimization problem in the SDN network.In DC network,the existing switchboards only support limited flow tables,and the existing links only support limited bandwidth.The unreasonable use of network resources will lead to single point avalanche problem,resulting in waste of the remaining network resources.Finally,we focus on the modeling and evaluation of controller in SDN network.The existing scheme only assumes the model follows the rule of first come first serve pattern,with no thought for the partition of traffic flows according to its type.Aimed to the problems,which is mentioned above,the following research is carried out in this paper:In this thesis:First of all,aimed at application aware of traffic flow classification problem,I try to establish the training set sequence through capture of network data packets and extracting related features in multiple dimensions.A multi type classifier is constructed by combining multiple SVM classifiers,which means we can jointly judge multiple dimension features based on two type classifiers combination.Based on the One-against-one voting strategy,when an unknown sample is classified,the most votes are the type to be determined.We divide the traffic flows into three ranges:1.Streaming Media Flow(include:video calls,P2P and non P2P multimedia playback.).2.Flexible Flow(HTTP file download,P2P file download.).3.UDP Flow(include:IM(WeChat,QQ),client network game,web browsing,intercom short voice.).In order to provide the basis of traffic flow classification for the two parts of next section work,I have worked as mentioned above.Then,we focus on designing and implementing an efficient routing algorithm for batch traffic flows routing based on partitioned ranges.According to the flow table and bandwidth demand tendency of three ranges of traffic flow,combined with the existing resource occupation of the network,the relevant constraints are established,and the optimization objective is put forward.At the same time,in order to avoid the single node failure problem of local network congestion,we attempt to introduce a penalty term that can make resource utilization balanced in the original target.Because whether there is a known path in the assigned routing path set can be expressed by Boolean values.Therefore,the form of genetic coding can be constructed exactly.In this paper,genetic operators,evolutionary operators and mutation operators are designed to iterate the elite set,and the optimal allocation is selected by scoring the elite individuals to achieve the optimization of K traffic routing path set.Finally,the modeling and performance evaluation of the controller with differentiated service priority are presented.Based on the results of traffic flow classification,the high priority queue is scheduled for the traffic with high delay requirement,and the traffic flow with low delay requirement is arranged to the low priority queue.Then,the flow request in the high priority queue is designed to preempt the service being accepted by the low priority queue with probability P.After the balance equations for the controller end are established,we get the steady-state probabilities of the two queues,which are solved by the controller respectively.Based on this,a queuing service model is built for the traffic flow request,and the balance equation is established again.According to this,we get the critical indexes of that model.By weighing controller in average delay and delay information entropy performance,we find the optimal preemption probability p,and through the analysis on the change trend of the index reached a reasonable assessment of the capability of the controller.The research work shows that the SDN network optimization strategy,which is proposed in this paper,based on the traffic flow type partition is given full consideration to the resource occupation tendency of different types of traffic flow.It can solve the multi objective and multi constrained problems of the link bandwidth and the balanced allocation of the node flow table in the SDN network,and provide a high quality optimization configuration scheme,which has good research value.