Performance Analysis of Computational Task Offloading Using Deep Reinforcement Learning

Abstract

Edge computing is a distributed approach that was designed for task processing and data management of IoT networks’ tasks. These computing resources are deployed nearby to IoT devices with an aim to reduce response time and to reduce overall bandwidth requirements to handle real-time applications over internet. Its main benefit is that it reduces the communication gaps between the IoT users and cloud server. Task offloading by using edge servers have attracted researchers to explore new scope for performance enhancement IoT network to handle computational complexities. For this there is requirement of latency minimization, task offloading, storage management, power consumption and so on. With implementation of Edge computing these objectives can be achieved. But still there is requirement of more advancement as everyday size of data is increasing which require more processing speed. As local computing is not possible all the time because IoT devices have limited battery power and computation resources. So, to support real time high latency consuming IoT applications, it is needed to reduce the offloading complexities at edge servers. Offloading algorithm are concerned to meet Quality-of-Service (QoS) requirements for different resource-demanding applications. In this work deep reinforcement learning is focused to minimize the computational complexity at IoT user end. The task offloading decision process is designed using Q-Learning that minimizes the system cost and curse of high dimensional data.