DRL-Based Availability-Aware Migration of a MEC Service

Abstract

Multi-access Edge Computing (MEC) allows a mobile user to access a service on a computing device called MEC Host (MEH), enabling lower latency by running the service closer to the users. When the user moves away from the serving MEH, the latency increases, which may cause a disruption of the user experience and of the service continuity. Moreover, the serving MEH may also fail, making the service unavailable. We propose a solution to a service migration problem that maximizes the MEC service availability by jointly deciding (i) migration timing and (ii) target MEH based on latency constraint, resource constraint, and availability status of a MEH. We solve the problem by using Deep Reinforcement Learning (DRL). The experiment shows that our proposed solution can successfully maintain a high service availability (more than 94%) in the presence of different failure probabilities, while another algorithm gives unstable service availability.

Keywords

• Edge computing
• 5G
• service migration
• reinforcement learning