Research On Multi QoS Constrained Routing Based On SDN

In a distributed network architecture,network devices independently complete network control and forwarding work based on neighbor node interaction information during the process of routing and forwarding business traffic.This kind of control and forwarding are tightly coupled,and the network architecture where the devices are fighting each other faces huge challenges in network management configuration and flexible network deployment.Software-defined network(Software Defined Network,SDN)sets up a centralized control center by decoupling the forwarding and control plane of the network,which can obtain network topology information globally and manage the network in a unified manner.In the new generation of 5G(5th Generation Wireless Systems)networks,in the future long-term network evolution plan,operators will adopt the SDN network architecture in the future core network.At the same time,with the current rise and development of technologies such as cloud computing and big data clusters,the data traffic in the Data Center Network(DCN)will face explosive growth.The new generation of major Internet companies such as Google,the data center network adopts SDN network architecture already,SDN is the general trend of network development.Getting rid of traditional routing design ideas,making full use of SDN network unified management and control,global information acquisition and other architectural advantages for routing design has a broad application prospects.In addition,the current network service traffic is becoming more and more diverse and complex(such as 8K live broadcast,short video,car networking service,etc.).These different service types have different requirements at network delay,bandwidth,packet loss and other performance parameters.The existing routing algorithms only use single network parameters as routing weight parameters,such as hop number etc,which is difficult to make full use of the advantages of SDN global network parameters,and can not guarantee the QoS requirements of diversified business flows.Therefore,the design of multi-QoS constrained routing in SDN network has very important research significance.In this thesis,to deal with the problem of multiple QoS constrained routing in the SDN network scenario,our research is presented as follows.Our thesis studies the routing problem of multiple QoS constraints in the core network of SDN-based Internet Service Provider(ISP)firstly.In this scenario,the SDN network combines Network Virtualization technology and service function chain(SFC),when routing flows,flows need to pass through a series of virtual network function nodes standardized by SFC in order to complete specific network functions.Existing algorithms face to different services are not able to make differentiated routing strategies for QoS requirements.Greedy algorithms are usually used to solve the routing paths of virtual function nodes step by step,which makes the algorithm easy to fall into the local optimal solution.This thesis presents a routing algorithm with multiple QoS parameter constraints for multiple types of service flows in the network.The algorithm start with establishing a link cost model for different QoS requirements of different services,and applying different links cost calculation for different flow requirements;at the same time,in order to avoid falling into the local optimal solution,a Viterbi based algorithm which reduces the complexity by reducing the solution space is proposed to solve this routing problem.Before applying the Viterbi algorithm,the algorithm filters the function nodes to be selected according to the remaining resources of the virtual network element function nodes,which can reduce the number of nodes to be selected and reduce the complexity of the algorithm.The thesis implements the above routing algorithm in Matlab simulation environment,compared with traditional routing algorithms such as shortest path and resource awareness,our algorithm can improve the acceptance rate of SFC flow,improve network throughput,and effectively meet the different QoS requirements of SFC flow.The second part of the thesis discusses the problem of multi-QoS constrained routing in the Data Center Network(DCN)under the SDN environment.The routing in this network scenario is to provide services for the business servers deployed in the network.The routing strategy needs to be able to effectively circumvent abnormal events such as network congestion in the network,reduce the occurrence of faults,and avoid problems such as interruption of business traffic or server disconnection.In addition,the service applications in the network are becoming more and more diverse,the design of routing algorithms also needs to fully consider the different QoS requirements of different service types to meet the development of the network.In order to solve the problem that traditional routing algorithms such as ECMP which can not learn and change routing strategies from network congestion or other network emergencies,this thesis proposes a fast-convergent DQN routing algorithm which considers the QoS requirements of multiple types of services,this algorithm applies deep reinforcement learning to the routing problem,in order to solve the different QoS requirements of different service flows in the network,the algorithm classifies different types of flows in the network,and sets different feedback reward functions according to the classification,so that the network converges to meeting the QoS requirements of multiple services.At the same time,the priority-based playback strategy adopted in the algorithm divides the experience playback pool into a priority pool and a non-priority pool,which improves the probability of selecting high learning value cases,thereby improving the learning efficiency of the network and accelerating the convergence of the algorithm.The algorithm can autonomously learn the characteristics of network traffic,adaptively adjust the network routing strategy,after multiple iterations of training,the algorithm converges to a better state.The thesis simulation compares our algorithm with traditional ECMP,DQN.The simulation results show that our algorithm can reduce the probability of network congestion and improve network throughput.There is a big performance improvement.In the last chapter,the thesis summarizes all the work and looks forward to the follow-up research and improvement work.