User Request Allocation Optimization In Intelligent SDN-based CDN

Nowadays,with the development of information technology,availability of intelligent terminals and applications,rapid growth of broadband access,network traffic shows a trend of explosive increasement.The demands of high quality content are increasing as well,which bring huge pressure on network infrastructures.Content Delivery Network(CDN)is an effective solution to offer reliable and efficient content provisioning across the Internet.However,traditional CDN is unable to possess global knowledge of whole network and relies on DNS redirection,therefore lacks the ability to dynamically control routing and server selection in real time.Software Defined Network(SDN)is a novel technology that allows decoupling control plane and data plane to offer flexible and centralized management.Implementing CDN with SDN provides an approach to monitor real-time network status and make intelligent decisions timely.Moreover,it can redirect user requests in finer time-scale.This paper firstly proposes a novel intelligent SDN-based CDN architecture.Through inoducing an intelligent center,it solves the increasing demand of SDN controller and the problem of information sharing and interaction between multiple control domains.This pape proposes a Model Predictive Control(MPC)-based algorithm to assign user requests.In the algorithm,we take two main parameters related to user QoE:response time and bandwidth satisfaction,to optimize routing and server selection jointly.The simulation results shows that the proposed algorithm achieve superior performance both in response time and bandwidth satisfaction.Besides,through changing the value of weight parameter,we can adjust priority of response time and bandwidth satisfaction to meet different network requirements.Further more,this paper proposes another algorithm to assign user requests based on MPC and Neuro-Network.The algorithm generates an overall score for each path based on Neuro-Network at first.Then,we assign user request according to these scores.The simulation results demonstrate that the performance has been further improved because of nonlinear mapping ability and self-learning ability of Nuero-Network.