A Mobility-Aware and Fault-Tolerant Service Offloading Method in Mobile Edge Computing

Mobile edge computing (MEC) is a prospective technology to render services through resources to fulfill the requirements of IoT (Internet of Things) devices at the cloud edge. The highly dynamic and heterogeneous characteristics of IoT devices bring both opportunities and challenges, i.e., a higher-than-usual occurrence rate of failures. Such failures occur at all architectural levels of the IoT applications: IoT sensor and actuator nodes can be missed, network links between IoT nodes can be down, and processing and storage IoT components can fail. In this work, for optimizing service offloading efficiency, energy consumption, and system reliability, a semi-online fault-tolerant offloading method (UDQF) was proposed for countering MEC failures by adopting a semi-online-learning-based service offloading strategy. The proposed strategy leverages a Dueling Deep Q network-based algorithm to determine user offloading behavior and utilizes an adaptive checkpointing mechanism (periodically storing the system state and restarting the system at the last checkpointing) to improve the task reliability. To valid and compare the model, the simulated results indicate that the proposed method outperforms other counterparts in multiple metrics.

I. Introduction

Mobile Edge Computing (MEC) is significant in real-time computing services among various branches. Many typical computation-intensive applications, e.g., face identification, interactive game, auto navigation, augmented reality, and remote control aircraft, benefit from the high-speed processing and large-scale distributed computing capabilities of mobile edge computing [1]. However, due to the fact that MEC relies on unreliable wireless communications and decentralized resource infrastructures, the MEC-based applications usually show a higher tendency to encounter varying types of system faults or failures, in terms of resource overflow (an overloaded MEC), software, or hardware breakdown, which further lead bad user-perceived QoE (quality of experience). Thus, implementing fault-tolerant techniques for the MEC infrastructure is an imperial need.