Intelligent Routing Algorithm Based On Reinforcement Learning In SDN

The traditional Internet structure adopts the TCP/IP system to form a distributed network,which is difficult to meet the rapidly increasing data volume and services of the data center,and cannot cope with the larger-scale traffic transmission and service quality guarantee.The SDN network realizes the separation of the control plane and the data plane.As a new type of network structure,it has the characteristics of transparent structure,programmable,and efficient forwarding.The current SDN network still uses the traditional shortest path algorithm,which often causes congestion in the face of large-scale traffic transmission,resulting in a decrease in bandwidth utilization and packet loss.This paper proposes a routing algorithm using reinforcement learning,which can efficiently find a suitable forwarding path for traffic,effectively improve the throughput and average bandwidth utilization of the SDN network under the Fat Tree topology,and reduce the packet loss rate in the forwarding process.First of all,this article introduces the structure of the SDN network and the working principle of the Open Flow protocol in detail,and uses the virtual network simulation software Mininet and the remote SDN controller Ryu to build the SDN network.On this basis,this article designed multiple Ryu APP modules to implement the control and forwarding functions of the SDN network,and completed the deployment of multiple routing algorithms.In addition,this article designs a loop processing method suitable for SDN to effectively avoid network paralysis caused by flooding of ARP packets and IPv packets.Then,this paper designs a routing algorithm for SDN networks based on the Qlearning algorithm in reinforcement learning.The algorithm improves the traditional reinforcement learning algorithm,designs a dynamically changing exploration rate and introduces a utility trace,and designs a reward function to meet different needs.The simulation results under the Fat Tree topology show that the algorithm can accurately find a suitable forwarding path for the traffic.Compared with the traditional routing algorithm,it can effectively improve the network throughput and average bandwidth utilization,and reduce the average packet loss rate.Finally,this article implements a hardware SDN network platform designed based on pica8 programmable SDN switches,using Raspberry Pi as the main control device,deploying related applications at the upper layer to realize network discovery,maintenance,and routing functions,and sending instructions through the serial port to complete the dynamic topology construction in the hardware switch.The hardware SDN network platform built in this article can realize self-defined dynamic topology changes and ensure the data transmission of terminal equipment.