一种基于状态信息聚合的边端协同卸载方法  

An end-edge collaborated computation offloading method based on state information aggregation

作　　者：戴彬[1,2] 任涛 胡哲源 牛建伟 DAI Bin;REN Tao;HU Zheyuan;NIU Jianwei(School of Computer Science and Engineering,Beihang University,Beijing 100083,China;Beihang Hangzhou Innovation Institute Yuhang,Hangzhou 310023,China;Hangzhou Innovation Institute,Beihang University,Hangzhou 310051,China)

机构地区：[1]北京航空航天大学计算机学院,北京100083 [2]北京航空航天大学杭州创新研究院(余杭),浙江杭州310023 [3]北京航空航天大学杭州创新研究院,浙江杭州310051

出　　处：《华中科技大学学报（自然科学版）》2022年第11期114-121,共8页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：浙江省自然科学基金资助项目(LY22F020006)。

摘　　要：针对现有移动边缘计算(MEC)研究中终端资源利用不充分和终端状态信息难表征,导致MEC系统整体利用率低的问题,提出一种基于状态信息聚合的边端协同卸载(SAEEC)方法.首先定义了边缘-拉普拉斯矩阵(ELM)来对终端的状态信息进行高质量压缩表征,然后借鉴联邦学习思想对压缩后的ELM表征信息进行边缘状态聚合,最后将经过全局聚合的决策向量分发给各终端,由终端分布式做出调度决策.实验结果表明:SAEEC方法可以大幅降低状态信息表征与聚合成本,同时提高终端自主决策效率和MEC整体资源利用效能,与传统方法相比可使MEC网络的平均成本下降20%以上.In view of the insufficient utilization of terminal resources and difficult representation of terminal state information in the existing research of mobile edge computing(MEC),leading to low overall utilization of MEC system,a state-aggregation based endedge collaborated computation offloading(SAEEC) method was proposed.First,the edge-Laplacian matrix(ELM) was defined to compress and represent terminal’s state information with high quality,and then the edge state aggregation of the compressed ELM representation was carried out by leveraging the idea of federated learning. Finally,the globally aggregated decision vector was returned to each terminal to make distributed scheduling decisions.Experimental results show that the SAEEC method can greatly reduce the state information aggregation overhead,and improve the decision-making efficiency and the overall resource utilization efficiency of MEC meanwhile.Compared with the conventional methods,the average overhead of the MEC network is reduced by more than 20%.

关 键 词：移动边缘计算 计算卸载 分布式调度 边端协同 状态表征 

分 类 号：TN929.5[电子电信—通信与信息系统]