| 摘要: |
| 针对移动边缘计算中用户移动性导致服务器间负载分布不均，用户服务质量(Quality of Service,QoS)下降的问题，提出了一种移动性感知下的分布式任务迁移方案。首先，以优化网络中性能最差的用户QoS为目标，建立了一个长期极大极小化公平性问题(Max Min Fairness,MMF)，利用李雅普诺夫（Lyapunov）优化将原问题转化解耦。然后，将其建模为去中心化部分可观测马尔可夫决策过程(Decentralized Partially Observable Markov Decision Process,Dec-POMDP)，提出一种基于多智能体柔性演员-评论家(Soft Actor-Critic,SAC)的分布式任务迁移算法，将奖励函数解耦为节点奖励和用户个体奖励，分别基于节点负载均衡度和用户QoS施加奖励。仿真结果表明，相比于现有任务迁移方案，所提算法能够在保证用户QoS的前提下降低任务迁移率，保证系统负载均衡。 |
| 关键词: 移动边缘计算（MEC） 移动性感知 任务迁移 多智能体强化学习（MARL） |
| DOI：10.20079/j.issn.1001-893x.221121002 |
|
| 基金项目:国家自然科学基金资助项目(62071077) |
|
| A Task Migration Scheme Based on Load Balancing under Mobility Aware |
| XIAN Yongju,HAN Ruiyin,ZUO Weihao,WANG Shuaige |
| (School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China) |
| Abstract: |
| To solve the problem that user mobility in mobile edge computing leads to uneven load distribution among servers and the decline of user quality of service(QoS),a mobility aware distributed task migration scheme is proposed.Firstly,with the goal of optimizing the QoS of the worst users in the network,a long-term max min fairness(MMF) problem is established,and the original problem is decoupled by using Lyapunov optimization.Then it is modeled as a Decentralized Partially Observable Markov Decision Process(Dec-POMDP),and a distributed task migration algorithm based on multi-agent Soft Actor-Critic(SAC) is proposed,which decouples the reward function into node reward and user individual reward.Rewards are imposed based on node load balance and user QoS respectively.The simulation results show that compared with the existing task migration schemes,the proposed algorithm can reduce the task migration rate and ensure the system load balance on the premise of guaranteeing user QoS. |
| Key words: mobile edge computing(MEC) mobility aware task migration multi-agent reinforcement learning(MARL) |
