Modeling And Optimization For Video Transmission Service Based On In-network Caching

With the continuous evolution and innovation of network architecture and commu-nication technology,the quantity and performance of user terminals have been upgraded by leaps and bounds.With this opportunity,the speed of popularization of the video business has been further accelerated.On the one hand,it enriches people's material and cultural life remarkably,and promote the development of related industries.On the other hand,they also inevitably lead to the exponential growth of video traffic in the network,and bring unprecedented pressure to the network infrastructure.In the face of massive bandwidth demand,improving the carrying capacity through network expansion and equipment update has been a drop in the ocean.In this case,the emergence of the in-network caching technology brings a turning point for the solution of this problem.Since the popularity distribution of video con-tents is subject to the 80/20 law,by saving hot contents frequently requested by users on the equipments of the edge network,the nearby transmission service can be pro-vided during the occurrence of the cache hits.So that it can not only greatly increase the transmission rate and reduce the transmission delay,but also can eliminate a large number of redundant traffic in the network,thus relieving the transmission pressure of the core links.Moreover,the increasingly mature Software Defined Networking(SDN)and Network Functions Virtualization(NFV)technologies also are capable of provid-ing a good platform for the deployment of in-network caching.Taking advantages of the scalability and programmability of SDN's centralized control plane can formulate specific caching mechanism and strategy,when combining the virtualized functions of content storage and transmission of the network nodes enabled by NFV,it is able to de-sign and implement the corresponding service framework according to the requirements of various video businesses.This dissertation studies the in-network caching assisted video transmission scenarios in SDN based wired local area network(LAN)and mo-bile cellular network respectively,and designs their corresponding system architectures.Meanwhile,the mathematical models are built to formulate the optimization problems,and the optimization control theory and machine learning algorithms are applied to find the optimal cache strategies.The main research contents are summarized as follows:1)This dissertation proposes a variable-length interval caching mechanism based on the finite state machine and the sliding window technique,and designs a cor-responding software-defined video streaming transmission system assisted by in-network caching.This solution aims at the scenario of VoD service in wired LAN based on SDN and NFV.Specifically,the SDN controller is used to collect the link state and cache distribution information of the data plane in real time,and the optimal service node is selected to provide the video streaming service ac-cording to the deployed content transmission strategy.On this basis,combined with the proposed sliding window based variable-length interval caching strate-gy,it can adaptably perceive the trend of video popularity,and adjust the size of cached segments,thus improving the utilization of cache resources.In additionthe mathematical model of the proposed caching mechanism is built,of which the performance is also derived theoretically.After that,the actual prototype system is built with general hardwares to verify its availability and effectiveness,and the comprehensive performance evaluation with multiple indicators as well as under large-scale network scenario is also carried out based on the network simulation platform,i.e.Mininet.Experimental results indicate that the proposed caching strategy can obtain higher caching utility and higher QoS of video transmission.2)This dissertation proposes a distributed cooperative caching strategy between multiple base stations based on multi-agent deep reinforcement learning.This solution is designed for the video transmission scenario in ultra-dense 5G cellular network of which the base stations support content caching,and the correspond-ing system architecture and transmission mechanism are both conceived.Next,the problem of in-network collaborative edge caching among multiple base sta-tions can be modeled as a joint strategy optimization based on Partially Observ-able Markov Decision Process(POMDP),which can be solved according to the proposed distributed cooperative caching strategy.First,by making use of the computing and storage resources deployed on each base station,and obtaining the state information of user access and video requests from local observations,it is possible to predict the changes of local states during future timeslot through the corresponding learning algorithms.On this basis,when combined with the global hidden state information shared by all base stations,it can be used as the input state of multi-agent collaborative edge caching algorithm,so as to find the optimal caching decision,aiming at improving the utilization of cache resources and reducing redundant traffic.Finally,the performance of the proposed caching algorithm in large-scale and high-density deployment scenarios of base stations is evaluated by simulation.The experimental results show that the caching strategy proposed in this dissertation can make full and effective use of the cache resources in the edge network through the collaboration mechanism of in-network caching,and improve the overall performance of the system.3)This dissertation proposes a pre-caching strategy with the perception ability of user mobility based on deep generation model and deep reinforcement learn-ing.This solution aims at the video transmission scenario assisted by in-network caching for mobile users in ultra-dense 5G cellular network,where a correspond-ing pre-caching architecture is designed based on the idea of trajectory prediction,which can fetch the content to be transmitted later to the incoming base station be-fore the user triggers a handoff.First,the neural network based generation model is used to extract the features of behavior patterns of different mobile users.After learning,the model can be leveraged to predict users' movement trajectories in the future.And then,combined with the observed network state information,the content pre-caching decision can be made according to the deep reinforcement learning algorithm,with the goal of improving the utilization of cache resources and the QoS of video transmission.Finally,the accuracy of the trajectory predic-tion algorithm is tested by using the GPS trajectory data sets of mobile users in re-al scenarios,and the proposed pre-caching strategy is also evaluated and verified through simulation experiments.The results indicate that the pre-caching strategy designed in this dissertation can enhance the performance of video transmission of mobile users by leveraging the in-network cache resources more efficiently.The research results of the above three aspects show the performance improvement and bandwidth-saving for the scenarios of video transmission in different network archi-tectures,which are benefited from the in-network caching technique.Meanwhile,the availability and effectiveness of the caching mechanisms and strategies designed in this dissertation are also verified,thus fully demonstrating its bright application prospect in the next generation network.