Research On Edge Caching Mechanism For Video Service

With the rapid development of mobile Internet in the 5G era and the popularization of various smart mobile terminal devices,everyone can create and deliver content.Video services such as short videos and live broadcasts are booming,and people’s demand for high-quality video services has also risen sharply.At the same time,a large number of smart devices connected to the core network increase the load on the backhaul link,and traditional cloud computing cannot meet users’ low-latency requirements for video services.In order to make up for the shortage of cloud computing,a new computing paradigm called Mobile Edge Computing(MEC)is proposed.By deploying edge nodes with computing and storage capabilities at the edge of the network,computing and storage services are provided on the edge side closer to the user,thereby reducing the cost of data transmission delay and alleviating network congestion.However,in the environment of edge computing,the difficulties of video caching include how to select appropriate content for caching among the continuously generated massive videos,the limited storage space of edge nodes,the mobility of users,the rapid changes in popular content,the complexity of the edge network and so on.Therefore,it is very significant to study the mechanism of edge caching for video.Considering the problems mentioned above,this paper mainly studies video caching in edge computing scenarios,our study includes selecting content to be cached,caching strategies and how to perform online caching in scenarios such as short videos with fast changes and high real-time requirements.The main contents are as follows:1)For selecting content to be cached,caching popular and popular videos can improve the cache hit rate and meet the total demand of all requesting users to the greatest extent.However,predicting popular and popular content often needs to be completed in a remote data center,which occupies a lot of network traffic and consumes communication.costs,and may also lead to user privacy leakage.To this end,this paper proposes a video request prediction algorithm based on federated learning.First,combined with LSTM and CNN,a video request prediction model(DRPN)is established,which predicts the probability that the video will be frequently requested in the future according to the historical request information sent by the user to the video,so as to determine the future trend of the video and obtain the video worth caching in advance.Then,for the proposed prediction model,the deployed multiple edge nodes are used for joint training,which improves the training speed and reduces the communication cost under the premise of ensuring the prediction accuracy.2)For the caching strategy,most of the traditional caching strategies are based on the architecture that contains a single edge node.When facing the edge caching system of multiple edge nodes,additional information cannot be used.To this end,this paper combines multiple adjacent edge nodes to propose a greedy algorithm-based cooperative caching strategy(GCCS),which aims to reduce the delay as much as possible.GCCS enables the local node and neighbor nodes to cooperatively cache some duplicate content to save more space to cache other content and increase the diversity of content.For Internet service providers,it can also reduce unnecessary redundant content in adjacent areas to a certain extent,saving caching costs.3)In view of the low latency requirements in the 5G era and the rapid changes in short video trends and strong user mobility,it is necessary to design a new and appropriate online caching system to realize timely selection according to time and space changes.Appropriately cache content and make caching decisions.To this end,this paper proposes an online video caching system based on federated neural collaborative filtering under the consideration of the application scenario where the user’s mobile device is more intelligent.Firstly,combined with GMF and MLP,an RPNCF prediction model is proposed to model the implicit features between users and requested videos.At the same time,the training method of federated learning between user devices is further improved.Before each round of model training,users are clustered and grouped,and the training parameters representing the users are used to update the parameters of the subordinate users in the same group,which further accelerates the training and enables Users can gain benefits early in training.Then,an appropriate online caching algorithm is designed to calculate and utilize the real time probability a video would be requested.The algorithm may ensure timely tracking of the content popularity that changes dynamically over time and space.Finally,the feasibility of the proposed online video caching system is proved by experiments,and it is better than the classical online caching strategies.