Learning-based Task-Offloading for Mobile Edge Computing with User Mobility

Mobile edge computing (MEC), envisioned as a cloud computing extension, is becoming an essential architecture for data transmission. The system performance (such as the execution delay) can be improved by offloading tasks from mobile user (MU) to edge servers. However, there is a lack of research on task offloading in the case of random movement of MUs. In this paper, we study the task offloading problem considering the MUs’ mobility. We model the task offloading problem as a Multi-armed Bandit (MAB) model. We then study the task offloading algorithms based on the typical Upper-Confidencial-Bound (UCB) and ε-greedy algorithms, which can make efficient task offloading decisions by adaptively learning the MU’s mobile property. By investigating the impact of the mobile property on network performance, we further propose an EPE-UCB task offloading algorithm. Simulation results demonstrate the efficiency of our algorithm in achieving low cost.