非均衡负载边缘计算资源调度策略优化方法  

Research on Optimization Method of Resource Scheduling Strategy for Unbalanced Load Edge Computing

作　　者：宋苗 章鹏 李发陵 SONG Miao;ZHANG Peng;LI Fa-ling(Chongqing Institute of Engineering,Chongqing 400056,China;Chongqing Satellite Network System Research Institute Co.,Ltd.,Chongqing 401135)

机构地区：[1]重庆工程学院,重庆400056 [2]重庆星网网络系统研究院有限公司,重庆401135

出　　处：《计算机仿真》2024年第11期233-238,共6页Computer Simulation

基　　金：重庆市自然科学基金项目资助(CSTB2022NSCQ-MSX1298);重庆市教委科学技术研究项目(KJZD-K202101901);重庆市高等教育教学改革研究项目((202126,213406);重庆市教育科学规划课题(2020-GX-145);重庆工程学院科研基金资助项目(2020xzky01)

摘　　要：目前常见的云边协同计算架构中容易因为负载分布不均出现任务时延难以满足需求、云平台资源调度动作频繁执行及网络与计算资源利用率低等问题。针对非均衡特征计算负载的特点和需求场景,提出一种应用趋势移动平均算法及整合LSTM网络的深度强化学习云边协同计算卸载方法MA-LSTM-DQN。通过仿真环境下使用运营商边缘计算负载模拟实验测试结果表明,所提出的方法在非均衡负载应用场景下较基准比较方法平均计算请求处理时延缩短17%,平台资源调度切换动作频次下降40%左右,综合性能指标有明显提升。Currently,common cloud edge collaborative computing architectures are prone to problems such as difficulty in meeting task latency requirements due to uneven load distribution,frequent execution of cloud platform resource scheduling actions,and low utilization of network and computing resources.Aiming at the characteristics and demand scenarios of unbalanced feature computing load,a deep reinforcement learning cloud edge collaborative computing offloading method MA-LSTM-DQN is proposed,which applies trend moving average algorithm and integrates LSTM network.The simulation experiment of the simulation environment and the test results of the actual operator's edge computing load application scenario show that the proposed unloading decision method in the unbalance computing load application scenario decrease the processing delay by 17%and scheduling frequency by about 40%,and improves the overall performance index to a certain extent.

关 键 词：移动边缘计算 资源调度 强化学习 马尔科夫决策过程 计算卸载 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]