一种基于强化学习的CDN流量调度系统设计方法

A Design Method of CDN Traffic Scheduling System Based on Reinforcement Learning

针对传统的CDN流量调度系统大多采用启发式方法或规划方法,存在维护成本高,实时性不足等缺点,提出一种基于深度强化学习的CDN流量调度系统设计框架。该框架基于马尔科夫链设计了故障告警网络来触发调度,建立了基于stacking模型的质量评估奖励函数,并在此基础上对流量调度进行定义和建模,构建了基于DQN的深度强化学习模型。最后,通过仿真实验验证了该调度框架的有效性。

The traditional CDN traffic scheduling system mostly adopts heuristic method or planning method,which has the disadvantages of high maintenance cost and poor real-time performance,resulting in weak quality of service control.This paper studies a design framework of CDN traffic scheduling system based on deep reinforcement learning.The framework defines and models traffic scheduling,firstly it triggers scheduling by constructing fault alarm network based on Markov chain,and constructs deep reinforcement learning model based on DQN which constructs quality evaluation reward function based on stacking model.Simulation results verify the effectiveness of the scheduling framework.