Research On Resource Optimization Technologies In Software-defined Networking

With the explosive growth of network traffic and the continuous deployment of various emerging network functions,the traditional Internet system has become more and more complicated,leading to serious degradation of service quality and low network efficiency.Software-Defined Networking(SDN)decouples the control plane and the data plane,and provides the advantages of centralized control,fine-grained data flow control,and open programming by maintaining a global network view to realize flexible and efficient network management and resource scheduling,which is of great significance in simplifying network operation and maintenance costs,improving network resource utility,and improving Quality of Service.However,it is the characteristics of SDN data-control separation,logically centralized control plane,and fine-grained flow control that also bring serious resource shortage problems,which makes SDN still face many challenges in large-scale network deployment and application,including:(1)The logically centralized control plane overloads the controller,resulting in control plane resources shortage problems;(2)The fine-grained data flow control overflows the switch flow table,resulting in flow table storage resources shortage problems;(3)The rigidity of routing scheduling strategy congests the link,resulting in an unreasonable allocation of link bandwidth resources.This dissertation studies the SDN resource optimization technologies in large-scale deployment scenarios.It starts with the control plane and data plane to improve the resource utilization and network service performance of SDN networks,which provides technical support for large-scale SDN deployment.The main innovations are as follows:1).For the irrational control resource allocation problem caused by ignoring data flow characteristics of existing controller association schemes,a control resource optimization mechanism based on controller dynamic association is proposed.The controller association mechanism is divided into two phases: firstly,a minimum set cover algorithm is designed to associate switches on the same data flow path to the same controller as much as possible to minimize control resource consumption.Then,a coalitional game strategy is introduced to achieve a more balanced controller-switch association scheme through cooperative games between controllers.The simulation results demonstrate that our proposed mechanism can reduce the control resource consumption by about 28% and the control traffic overhead by 8%.2).For the insufficient storage resources problem of switches caused by SDN fine-grained flow control,a flow table rule optimization mechanism based on segmented routing is proposed.This mechanism introduces Segment Routing into the SDN architecture and designs an online data flow path encoding algorithm to perform path aggregation on the data flow based on the coincidence of the flow paths,so that the data flow of the same path segment can share the flow table rules to reduce the number of flow table rules to be installed.Simulation results show that our proposed mechanism can effectively reduce the number of flow table rules by about 61% and the number of encapsulated MPLS labels by 56%.3).For the irrational link bandwidth resources allocation problem caused by traditional routing protocols relying on manual configuration,a routing optimization mechanism based on deep reinforcement learning is proposed.This mechanism designs a network utility function with latency and throughput awareness.With the goal of maximizing network utility,it introduces a deep reinforcement learning DDPG algorithm to iteratively learn traffic characteristics and routing strategies for finding the optimal path in SDN networks.The experimental results prove that compared with the traditional routing protocols OSPF and ECMP,our proposed mechanism can greatly improve network throughput by about 30% and reduce the packet delay by about 22%.