基于策略梯度算法的工业数据中心通信调度性能评估

Communication Scheduling Performance Evaluation of Industrial Data Center Based on Policy Gradient Algorithm

为了提高工业数据中心网络调度能力,设计了一种基于策略梯度算法的工业数据中心通信调度性能评估方法。将Coflow的资源数据转化为待执行图象,通过策略梯度算法实现最优化的通信调度。研究结果表明,逐渐提高Coflow到达率,会增加了Coflow的平均流程处理时间。当Coflow达到率很低时,PGO与GCN具有相似的特性,且均优于DNN。该研究有助于提高工业运行安全和节约能源,具有很好的应用价值。

In order to improve the network scheduling capability of industrial data centers,a communication scheduling performance evaluation method based on policy gradient algorithm was designed.The resource data of coflow is transformed into the image to be executed,and the optimal communication scheduling is realized by the strategy gradient algorithm.The results show that increasing the coflow arrival rate will increase the average process processing time of coflow.When the coflow reach rate is very low,PGO and GCN have similar characteristics,and both are better than DNN.This research is helpful to improve the safety of industrial operation and save energy,and has good application value.