Autoscaling in Mobile Edge Computing Based on Multi-Agent Reinforcement Learning
Ling Cui, Tianyi Shi, Ruifang Lu, Tiankui Zhang · 2023
Mobile edge computing (MEC) has emerged as a transformative paradigm by deploying computing and storage resources at the edge of access networks. This enables real-time processing, reduces latency, and improves user experiences, particularly with the advent of 5G networks. However, the limited resources of edge servers pose a challenge in ensuring service-level agreement (SLA) while optimizing resource utilization. Autoscaling, a dynamic resource adjustment mechanism, can play a pivotal role in meeting these demands. This paper explores the application of multi-agent reinforcement learning (MARL) to address autoscaling in MEC. The proposed algorithm models the horizontal autoscaling problem in MEC, defines state and action spaces, and introduces a global reward function. Additionally, time-series data prediction enhances scaling responsiveness. A testing platform is employed to validate the algorithm’s effectiveness. This work contributes a novel approach to autoscaling in MEC, addressing the complex interplay of autonomous scaling decisions, SLA compliance, and resource optimization within edge server clusters.