On the Placement of Edge Servers in Mobile Edge Computing

With the wide deployment of mobile devices, a variety of different computation-intensive applications for mobile platforms, such as online games and virtual augmented reality, have emerged. However, due to the limited computation resources, mobile devices struggle to complete the required computation in time. One potential solution to the problem is mobile edge computing. Over the past years, there have been a series of studies on edge server placement. Most of the existing studies focus on the minimization of the delay between mobile devices and edge servers because many mobile applications are time-sensitive. However, workload balance is also an important metric because we do not prefer a scenario where some edge servers are idle while others are overloaded. In our research, we took both the delay and workload balance into account when we attempted to propose an effective edge server placement strategy. In addition, machine learning techniques were utilized to arrive at the appropriate placement method. Specifically, an innovative edge server placement strategy, which combines the advantages of cluster-based method and heuristic schemes, is proposed in the paper. Our experimental results indicate that, compared with the existing schemes, the proposed strategy can achieve lower communication delay and better workload balance simultaneously.

I. Introduction

Over the past years, with the rapid development of mobile devices and the Internet of Things (IoT), the computing platform for business applications and personal entertainment has gone through a revolutionary change [1]. Mobile devices, such as smartphones and tablets, have become pervasive and critical in our daily lives. In the meanwhile, with the continuous development of machine learning and virtual reality, many smart applications and services are provided to enable a more convenient and intelligent life. However, these smart applications and services have become more computation-intensive. It is expected that the current mobile devices will not be able to satisfy the high demand for computing resources because they are constrained in terms of processing power, storage capacity, and battery life. To tackle the problem, mobile devices could offload computation-intensive tasks to cloud servers. However, cloud servers are typically far from mobile devices. Consequently, this solution often suffers from the long latency problem [2] [3]. For time-sensitive applications that need a prompt response, such as real-time navigation and online gaming, it is infeasible to offload computation-intensive tasks to cloud servers. Namely, centralized cloud computing is not very effective for geo-distributed mobile devices.