Research On Diffserv Mechanism Of Fine Grained Video Based On ML In SDN

With the rapidly expansion of video servers such as video chat,the proportion of video traffic in multiledia is also improved quickly.At the same time,due to the lower delay and higher bandwithdth demand than other traffic,which usually caused the congestion of links.And difficult to control,the need to implement fine-grained classification and scheduling of video streams is becoming more and more urgent.Under the traditional network architecture,a best-effort forwarding strategy is adopted,but when the network is large,it is difficult to quickly deploy new functions,which increases the difficulty of network management.Software Define Network(SDN)has greatly enhanced flexibility in network resource management and scheduling due to its programmable,logical centralization,and separation of transfer and control,making it more suitable for deployment of video streams.Classification and scheduling strategy.Therefore,this thesis studies the fine-grained classification and routing of video streams under the SDN architecture and integrates them into a multi-video stream classification and scheduling system,which aims to reduce the resource idleness and provide differentiated transmission for video services.The main work of this thesis is originized as follows:1.Classify video streams according to the differentiated QoS requirements of different video services in terms of delay,bandwidth,and reliability.This thesis first preprocesses the self-collected video traffic data set and public traffic data set,and then comprehensively compares the accuracy and real-time evaluation of common clustering algorithms and supervised machine learning algorithms.Finally,the selected CART algorithm is in the public The video classification accuracy rate on the data set has reached more than 97.3%,and it has high classification real-time performance.2.In view of the differentiated QoS requirements for different viedeo streams,it is necessary to add constraints to the QoS indicators of different video streams,which is abstracted as a multi-metric QoS constraint problem.An optimized version named MMCP is presented to adapts to the multi-metric QoS constraint problem in the context of multi-video stream scheduling,and proposes an approximate algorithm to solve the optimized version.Finally,the time complexity and related properties of the algorithm are analyzed.3.Combining the CART algorithm and the approximate routing algorithm,under the differentiated service model(Differentiated Service,Diffserv)framework,a scheme design of the multi-video classification and scheduling system in the SDN network is proposed.The system was deployed on the SDN experimental platform to test the functions of classification and path selection.Finally,the QoS performance of different video stream transmission paths was analyzed.