Session Scheduling Strategy For Streaming Media Edge Cloud Based On Deep Reinforcement Learning

There are some disadvantages of poor reliability and high cost of expansion for the traditional streaming media systems based on CDN and P2P techniques.The maturity of cloud computing technology has enabled streaming media services to evolve into cloud forms,and streaming media can solve the above problems effectively.The streaming media cloud is placed at the edge of the network to relieve the core network traffic load and increase the response speed of user requests.Generally,system resources of streaming edge cloud are allocated on demand.Compared with the traditional streaming media service system,the flexible streaming media edge cloud has higher requirements on resource scheduling.Current solutions to the problem of streaming media edge cloud resource scheduling are always based on traditional heuristic algorithms or planning methods,which suffer from insufficient adaptability and high time complexity.Moreover,they are difficult to adjust the strategy according to the system operation scenario.However,reinforcement learning techniques can automatically adapt to complex environments with interactions under the trial-and-error mechanisms,and learn the optimal strategy by maximizing the cumulative reward.Therefore,we propose to solve the session scheduling problem of the streaming media edge cloud system using reinforcement learning methods,and the main works we have acomplished are listed below:1)We propose a novel streaming media edge cloud session scheduling policy,inspired by deep reinforcement learning techniques and taking the constraints of migration cost and load balancing into consideration.Specifically,we define the required elements of reinforcement learning techniques,including state space,action set and reward function,according to the characteristics of the streaming cloud edge cloud system scheduling.Besides,we utilize convolutional neural network to fit the strategy function and action-value function,in order to solve the problems of high-dimensional input and the storage and generalization of action-value function.We train the neural network using a deterministic strategy gradient reinforcement learning algorithm.2)We implement the deep reinforcement learning algorithm and conduct corresponding simulation experiments.Firstly,build experimental platform and implement reinforcement learning algorithm based on a deterministic strategy gradient,then train neural network according to the process of algorithm.Finally,perform simulation experiment of user requested access by using the well-trained policy network.In summary,this dissertation implements a session scheduling algorithm based on deep reinforcement learning techniques and conducts simulation experiments to verify the effectiveness of the proposed algorithm.The experimental results show that our strategy can not only achieve good request access result,but also reduce the migration cost.Moreover,it can also reduce the running time and provide certain adaptability under uncertain streaming media edge cloud system environment.