Intelligent Resource-schedule Strategy For Media Edge Cloud

The traditional streaming media systems based on CDN and P2P are suffering from limitations on scalability,transparency and reliability.Cloud Computing,a relatively new technology,has advantages in resource virtualization,reliability and flexible scal-ability.Hence,as a combination product of them,streaming media cloud becomes one of future developing trends about streaming media technology.Further more,due to the high real-time requirement,streaming media edge cloud(MEC)usually be deployed in many regions to provide better user experiences.Similar to the traditional streaming media service system,bandwidth and cache allocation are still important problems for MEC.Traditionally,session migration or video redeployment is separately proposed to handle the cache allocation problems.But,neither of them can achieve a good balance between cost and performance solely.Deep learning model has advantages in efficiency and generals,which provides a new direction for resources scheduling area.Hence,besides some researches about joint optimization method on MEC resources scheduling problem,we introduce deep rein-forcement learning techniques to achieve improvement.The main content can be sum-marized as follows:1)Considering the fluctuation of video popularity,a two-phase schedule strategy is proposed.In detail,to handle the slight variations of video popularity,a session mi-gration strategy considering load and popularity is proposed.A video redeployment strategy considering cost and load balance is adopted to handle serious change of pop-ularity.With the cooperation of them,high adaptability about popularity and low cost can both be achieved.2)For improving efficiency of the two-phase schedule strategy,an enhanced method based on DRL is proposed and we build a numerical simulation platform to analyze the performance of the algorithm.In detail,to decrease the complexity of scheduling ac-tion,strategy based on separate subnetworks is adopted.Strategy based on access ratio and session remaining ratio is adopted to make a good control of cost.A state generated strategy based on simulator is proposed to enhance the efficiency of method.By intro-ducing the DRL method,the efficiency of the two-phase schedule strategy is increased,which provides a good practicability for the method.In summary,we proposed a two-phase schedule strategy for MEC resources schedul-ing problem.Moreover,we introduce DRL method to make a improvement of the strategy.The result shows that the method achieves better speed,while remaining the advantages of old method.