Details of Research Outputs

在移动边缘计算中通过将终端设备的计算任务卸载到边缘服务器,可以利用边缘服务器资源解决终端设备计算能力不足的问题,同时满足移动应用程序对低延迟的需求.因此,计算卸载备受关注并成为移动边缘计算的关键技术之一.本文对移动边缘计算的计算卸载研究进展进行深度调研.首先,总结归纳出两类计算卸载方法——基于启发式算法的传统方法和基于在线学习的智能方法;从最小化延迟时间、最小化能耗、权衡时间和能耗三个不同优化目标对基于启发式算法的传统计算卸载进行分析对比;梳理了基于在线学习智能计算卸载采用的底层人工智能技术;然后介绍了边缘服务器资源分配方案和新兴的移动边缘计算应用场景;最后分析计算卸载方案存在的问题并展望移动边缘计算的计算卸载研究的未来方向,为后续研究工作指明方向. 

With the rapid development of the Internet of Things (IoT), cloud computing, and big data, data manifests explosive growth. Traditional cloud computing uploads massive data to cloud servers. Due to the long distance between cloud servers and mobile devices, traditional mobile cloud computing suffers from high energy consumption and network delay, which limits the development of mobile applications. To overcome this limitation, Mobile Edge Computing (MEC), a novel networking and computing paradigm, is proposed and becoming more and more prevalent. In MEC, the computation tasks are offloaded from the resource-limited mobile devices to the powerful network edges, which can leverage the computing resources of the network edges to perform the computation tasks while providing quite low latency as most mobile applications ask. Therefore, computation offloading has become one of the essential technologies of MEC and gained a lot of attention in both the academic community and industrial world. In this paper, we conduct a deep survey of the state-of-the-art works of computation offloading in MEC. First, we divide the existing computation offloading schemes into two categories: the traditional computation offloading based on heuristic algorithms and the intelligent computation offloading based on online learning. We compare these two different methods in detail and analyze the advantages and disadvantages of the existing computation offloading methods. In general, the intelligent computation offloading methods are the mainstream of research directions in the future which outperforms the traditional computation offloading in terms of privacy data and user mobility. We analyze the traditional computation offloading schemes according to their optimization objectives, which include minimizing network delay, minimizing energy consumption, and optimizing the trade-off between network delay and energy consumption. These schemes are based on heuristic algorithm based on an optimization objective, and then design a heuristic algorithm to approach the optimal solutions. We also analyze the intelligent computation offloading schemes based on the underlying Artificial Intelligence (AI) technologies. These schemes based on online learning not only can solve the problem of high delay and energy consumption but also address the problem of data security and privacy, which is not considered in the traditional methods. More importantly, with AI technologies, most intelligent computation offloading does not take the network delay or the energy consumption as a single optimization objective but considers the overall performance of offloading. Then, we introduce the resource allocation schemes of the edge servers, which is an important process after the concrete computation tasks are offloaded to the edge servers. We also present several emerging application scenarios such as Internet of Things(IoT), Internet of Vehicles(IoV), Blockchain, Unmanned Aerial Vehicle(UAV), Virtual Reality(VR), and Augmented Reality (AR). Finally, we conclude technological challenges which include mobility of end devices, edge servers, security, user privacy data, and service heterogeneity and prospect future directions about computation offloading which could point out the direction for the follow-up research. Though this area of research is young, there is much room for improvement. We believe that more studies about computation offloading in mobile edge computing will bring more opportunities soon.