计算机工程 ›› 2019, Vol. 45 ›› Issue (5): 237-242,248.doi: 10.19678/j.issn.1000-3428.0050875

• 多媒体技术及应用 • 上一篇    下一篇

基于深度强化学习的流媒体边缘云会话调度策略

徐西建,王子磊,奚宏生   

  1. 中国科学技术大学 自动化系,合肥 230027
  • 收稿日期:2018-03-20 出版日期:2019-05-15 发布日期:2019-05-15
  • 作者简介:徐西建(1992—),男,硕士研究生,主研方向为网络多媒体;王子磊,副教授;奚宏生,教授。
  • 基金项目:

    国家自然科学基金(61673362);中央高校基本科研业务费专项资金(WK3500000002)。

Session scheduling strategy for streaming media edge cloud based on deep reinforcement learning

XU Xijian,WANG Zilei,XI Hongsheng   

  1. Department of Automation,University of Science and Technology of China,Hefei 230027,China
  • Received:2018-03-20 Online:2019-05-15 Published:2019-05-15

摘要:

在流媒体边缘云资源调度中,传统启发式方法或规划方法多数存在自适应性不足、时间复杂度高等问题。基于迁移代价、负载均衡等约束,提出一种流媒体边缘云会话调度策略。以流媒体边缘云系统的状态信息作为属性特征,结合深度学习与确定性策略进行梯度强化学习,以解决用户请求接入问题。实验结果表明,该策略具有较好的请求接入效果,且能够降低迁移代价,同时缩短了运行时间。

关键词: 流媒体边缘云, 会话调度, 会话迁移, 深度学习, 强化学习, 确定性策略梯度

Abstract:

In the streaming cloud edge resource scheduling,traditional heuristic methods or planning methods mostly have problems such as insufficient adaptability and high time complexity.Based on the constraints of migration cost and load balancing,a session scheduling strategy for streaming media edge cloud is proposed.The state information of the streaming media edge cloud system is used as the attribute feature,and the deep learning and the deterministic strategy are combined to carry out the gradient reinforcement learning to solve the problem of user request access.Experimental results show that the strategy has better request access effect,and can reduce the migration cost and shorten the running time.

Key words: streaming Media Edge Cloud(MEC), session scheduling, session migration, deep learning, reinforcement learning, deterministic strategy gradient

中图分类号: