Research on Microservice Coordination Technologies based on Deep Reinforcement Learning

In order to solve the problems of scaling, placement and traffic scheduling of a large number of microservices. Based on tensorflow and keras machine learning framework, this paper designs and implements a microservice coordination method based on DDPG (deep deterministic policy gradient) algorithm in DRL (deep reinforcement learning). Firstly, the network model and service traffic model are built, and the optimization objectives are defined, i.e. throughout and delay. Besides, a POMDP (partially observable Markov decision processes) and a traffic scheduling table are designed to build the service coordination framework and finish the ultimate algorithm. Secondly, the utility of our algorithm and other baseline-algorithms is numerically evaluated within different traffic pattern and real-world traffic traces. Thirdly, technology proposed in this paper and 5g core network are combined. Furthermore, the significance of this coordination technology on 5g core network in the trend of microservice is also discussed.