Joint Optimization of Service Migration and Resource Management for Vehicular Edge Computing

The integration of edge computing and vehicle networks with future communications (e.g., 5G/6G networks) enables computation-intensive and delay-sensitive applications, such as autonomous driving. However, the mobility of end devices and vehicles with context switching brings a considerable Quality of Service (QoS) degradation and interruptions in edge services provisioning. Predicting the end-user mobility and context switching enable edge nodes to migrate services proactively to guarantee the QoS. But the cost maybe significant if the prediction is wrong. Service migration seems to be an effective way to minimize service disruptions and maintain service continuity. Service migration can be performed in two ways: proactive and reactive. Most of existing works focus on either reactive or proactive service migration models without considering resource usage efficiency. In this paper, we consider the combination of reactive and proactive service migration and propose a novel framework for joint optimization of service migration and resource management for vehicular edge computing by leveraging reinforcement learning (RL) as an emerging technique. We construct an optimization problem to reduce migration costs in terms of latency and energy usage. Finally, we present Multi-Armed Bandit (MAB) methods for solving the optimization issue. Rigorous theoretical analysis and extensive evaluations, combining both testbed and numerical simulation, demonstrate the superior performance of the proposed intelligent service migration framework in vehicular edge computing environments.

I. Introduction

Intelligent transportation systems equipped with abundant sensors and actuators produce massive volume of data. As a traditional solution for processing such data, cloud-centric approaches with abundant resources have been supporting massive-scale devices deployment scenarios [1]. However, they can hardly fulfill the performance requirements of rapidly emerging intelligent transportation system applications, such as autonomous driving [2], requiring high availability and dependability, as well as high throughput and ultra-low latency [3]. As a result, the mobile edge computing (MEC), a.k.a. fog computing, paradigm has been developed to address the low latency and high throughput difficulties by relocating cloud services closer to local infrastructures to allow real-time applications and improve quality of service [4], [5].