RL-based Computation Offloading Scheme for Improving QoE in Edge Computing Environments

With the rapid development of Internet of Things (IoT) devices and networks, various services are being provided to people. However, IoT devices are limited by computing resources and cannot satisfy services that require short processing time. To satisfy the requirements of these services, edge collaboration schemes have been researched to utilize the computing resources of other edge servers. Generally, edge collaboration schemes offload tasks received from devices to appropriate edge servers based on Reinforcement Learning (RL). Reinforcement learningbased edge collaboration schemes do not exhibit high performance due to decision-making based on a single objective function. In this paper, we propose a RL-based computation offloading scheme for improving QoE in edge computing environments. The proposed scheme approximates the action distribution based on the multiobjective function to maximize performance. To evaluate the performance of the proposed scheme, comparative experiments were conducted in simulation environment, comparing it with existing computation offloading schemes that utilize reinforcement learning in edge collaboration environments. The experimental results confirm the improved learning performance compared to the existing schemes.

I. Introduction

The evolution of the Internet of Things (IoT) and networks is increasing the demand for various new information and communication services [1]. Among the information and communication services, services such as Augmented Reality (AR), Virtual Reality (VR), and natural language processing require low latency to provide high Quality of Experience (QoE) [2]. However, IoT devices do not have enough computing resources to satisfy user requirements. Edge server-based offloading schemes have been studied to help a large amount of computing resources available to satisfy user requirements. Edge servers have high computing power compared to IoT devices and are located close to devices, so they can satisfy services that require low latency [3]. However, when a large number of devices are connected to one edge server, the service cannot be processed efficiently. To solve these problems, edge collaboration-based task offloading schemes have been proposed.