SDCN中基于深度强化学习的移动边缘计算任务卸载算法研究
Research on task offloading algorithm of mobile edge computing based on deep reinforcement learning in SDCN
蒋守花 1王以伍1
作者信息
- 1. 成都医学院现代教育技术中心,四川 成都 610500
- 折叠
摘要
随着网络技术的不断发展,基于Fat-Tree的网络拓扑结构分布式网络控制模式逐渐显露出其局限性,软件定义数据中心网络(software-defined data center network,SDCN)技术作为Fat-Tree网络拓扑的改进技术,受到越来越多研究者的关注.首先搭建了一个 SDCN 中的边缘计算架构和基于移动边缘计算(mobile edge computing,MEC)平台三层服务架构的任务卸载模型,结合移动边缘计算平台的实际应用场景,利用同策略经验回放和熵正则改进传统的深度Q网络(deep Q-leaning network,DQN)算法,优化了MEC平台的任务卸载策略,并设计了实验对基于同策略经验回放和熵正则的改进深度Q网络算法(improved DQN algorithm based on same strategy empirical playback and entropy regularization,RSS2E-DQN)和其他3种算法在负载均衡、能耗、时延、网络使用量几个方面进行对比分析,验证了改进算法在上述4 个方面具有更优越的性能.
Abstract
With the continuous development of network technology,the network topology distributed network control mode based on Fat-Tree gradually reveals its limitations.Software-defined data center network(SDCN)technology,as an improved technology of Fat-Tree network topology,has attracted more and more researchers'attention.Firstly,an edge computing architecture in SDCN and a task offloading model based on the three-layer service architecture of the mobile edge computing(MEC)platform were built,combined with the actual application scenarios of the MEC platform.Through the same strategy experience playback and entropy regularization,the traditional deep Q-leaning network(DQN)algorithm was improved,and the task offloading strategy of MEC platform was optimized.An im-proved DQN algorithm based on same strategy empirical playback and entropy regularization(RSS2E-DQN)was compared with three other algorithms in load balancing,energy consumption,delay and network usage.It is verified that the improved algorithm has better performance in the above four aspects.
关键词
软件定义数据中心网络/深度强化学习/边缘计算任务卸载/同策略经验回放/熵正则Key words
software-defined data center network/deep reinforcement learning/edge computing task offloading/re-play the same strategy experience/entropy regularity引用本文复制引用
基金项目
四川省高等学校人文社会科学重点研究基地·四川省教育信息化应用与发展研究中心项目(JYXX23-002)
成都医学院校基金科研项目(CYSYB23-02)
出版年
2024
NETL
NSTL
万方数据