A Novel Task-Allocation Framework Based on Decision-Tree Classification Algorithm in MEC

Wenwen Liu, Zhaoyang Yu, Meng Yan, Gang Wang, Xiaoguang Liu

Abstract: Mobile-edge computing (MEC) has emerged as a promising paradigm to extend the cloud computing tasks to the edge mobile devices for improving the quality of service. This paradigm addresses the problems in cloud computing architecture by enabling lower latency, higher bandwidth, and better privacy and security. Previous studies of task allocation in MEC systems generally only consider the single influence factor, such as the distance between the mobile device and cloudlets, to select the suitable cloudlets for tasks. However, there are various types of tasks with complex individual requirements about which we should consider, such as transmission bandwidth, computing capacity and storage capacity of cloudlets, and so on. In this paper, we propose an on-demand and service oriented task-allocation framework based on machine learning technology, called Edgant. It classifies tasks into three types using a decision-tree model according to the tasks’ characteristics including requirement on server resources and user’s requirements. For each type, we provide a selection strategy to allocate task to the most suitable cloudlet based on characteristics of the task and the current state of cloudlets. Simulated evaluations demonstrate that Edgant achieves lower latency and better accuracy compared to other task allocation methods, as well as high reliability under high mobility of the mobile devices.