Virtual Machine Migration Strategy Based on Multi-Agent Deep Reinforcement Learning

Mobile edge computing is a new model, which pushes cloud power from centralized to network edge. However, with the sinking of power, user mobility brings challenges: since it usually unstable, services should be dynamically migrated between multiple servers maintain service performance, that is, user-perceived latency. Considering Edge Computing highly distributed environment and difficult synchronize information servers, in order ensure real-time performance migration strategy, virtual machine strategy based on Multi-Agent Deep Reinforcement Learning proposed this paper. The method training execution adopted, transfer action guided by global during training, only local observation needed obtain action. Compared control method, alleviates communication bottleneck. other methods, needs information, does not need speeds up perception current environment. Migration strategies can generated faster. Simulation results show better than contrast terms convergence energy consumption.