Mobility-Aware Computation Offloading for Hierarchical Mobile Edge Computing

Mobile edge computing (MEC) is a promising technology that aims to reduce the total latency of user equipment (UE) by deploying computation resources at the edge of mobile networks. UE mobility is a challenging factor that causes the traditional MEC architecture to suffer from several issues, such as decreased efficiency and frequent service interruptions. One popular method to manage UE mobility is virtual machine (VM) migration, which requires high bandwidth and causes undesirable latency, rendering it impractical for real-time tasks with stringent latency requirements. This paper proposes a hierarchical architecture for MEC networks that facilitates mobility management and mitigates the need for VM migration. In order to utilize this architecture efficiently, a Markov chain-based predictive strategy is introduced to predict UE mobility. Afterward, an optimization problem is formulated to make the optimal long-term offloading decisions for UEs such that their expected cost is minimized subject to latency commitments and resource consumption constraints. Simulation results demonstrate that the proposed scheme reduces the cost of high-mobility UEs by up to 25% compared to traditional schemes. Furthermore, the measures of movement direction predictability and offloading decision popularity are introduced that provide insights into the behavior of the proposed and counterpart schemes.

I. Introduction

In recent years, the demand to run computationally intensive applications, such as facial recognition, object detection, speech processing, etc., on user equipment (UE) has been increasing steadily. However, due to their limited battery capacity and computation power, UEs cannot afford such applications [1]. Cloud computing provided a promising solution to this challenge, where computationally intensive tasks could be offloaded to centralized cloud servers through a mobile core network and the Internet [2]. For instance, in online cloud gaming, UEs could offload the entire game engine to a cloud server and only receive rendered frames via the Internet to be displayed on their equipment [3]. Overall, cloud computing brought many advantages for UEs, such as increased battery lifetime, unlimited storage space, and the ability to run computationally intensive tasks without owning up-to-date equipment [2]. Despite its considerable advantages, cloud computing suffers from some drawbacks, such as long latency to access a distant cloud server via the Internet. These drawbacks impede newly emerged applications with strict data rate, latency, and response time requirements from exploiting cloud computing [4]. For instance, autonomous vehicles need to process an enormous amount of data in a matter of milliseconds in order to detect routes and obstacles in a timely manner. Offloading this data to a distant cloud server would significantly increase the obstacle detection delay and, therefore, endanger the life of passengers [5]. Consequently, a new paradigm called edge computing was proposed to mitigate the challenges of cloud computing and facilitate strict quality of service (QoS) provisioning by deploying computation resources at the network edge and providing computation services to UEs in their proximity [6].