A learning-based framework for optimizing service migration in mobile edge clouds

Abstract

Mobile edge computing is gaining traction due to its ability to deliver ultra-low-latency services for mobile applications. This is achieved through a federation of edge clouds in close proximity of users. However, the intrinsic mobility of users brings a high level of dynamics to the edge environment, calling for sophisticated service migration management across the edge clouds. Previous solutions for edge service placement/migration are architecture-specific, centralized, or are based on restricted cost models. These limitations leave doubts about the practicality of these approaches due to the lack of a standardized reference model for edge clouds. In this paper, we propose a general framework for optimizing edge service migration based on reinforcement learning techniques. Using our framework, edge service migration strategies can be learned with respect to a large variety of optimization goals. Moreover, our learning-based algorithm is agnostic to the underlying architecture and resource constraints. Preliminary results show that our model-free learning-based approach can compete with model-based baselines and adapt to different objectives.