Efficient Task Offloading in Multi-access Edge Computing Servers using Asynchronous Meta Reinforcement Learning in 5G

The deployment of 5G networks has ushered in a new era for technologies that demand Ultra Reliable Low Latency (URLL) networks. However, the User Equipment (UE) responsible for handling such applications may not always have the computational capacity to meet the demanding requirements of URLL implementations. In such scenarios, certain tasks can be offloaded to a high-performance computing network known as Multi-access Edge Computing (MEC), located near the UE at a base station. Nonetheless, MEC servers themselves have computational limitations, and they can’t handle all the tasks from all the UEs efficiently. To address this challenge, we propose an efficient task offloading approach based on asynchronous Meta Reinforcement Learning (MRL). Our method involves developing an algorithm that prioritizes and organizes UE tasks based on their importance and interdependencies, and a Deep asynchronous MRL algorithm models the task offloading policy. What sets our model apart is its ability to adapt quickly to a variety of environments, whether they are heterogeneous or homogeneous. Our simulation results show that our solution has a 60% overall performance improvement in terms of training latency, energy consumption, and memory and CPU time usage compared to a benchmark.

I. Introduction

The fifth-generation (5G) standard for broadband cellular networks is designed to provide users with an Ultra Reliable Low Latency (URLL) network with speeds 10 to 100 times that of its predecessor 4G [1]. This has paved the way for deploying applications such as Tactile Internet (TI) that transfers haptic (touch) and kinesthetic (muscular movement) data.