Swarm Optimized Deep Learning Scheduling in Cloud for Resource-intensive Iot Systems

Abstract

Abstract The paradigm Internet of Things (IoT) connects several million devices that can gather information which is stored and processed in the Cloud. This data is analyzed for inferring knowledge and performing predictions and analysis. Generally, in the cloud platform, users are charged based on the resources for storage and computing used. For real-world scheduling applications, these machines are not always available during certain periods of time owing to their stochastic or deterministic causes. The problem of cloud computing scheduling can be as challenging as the Non-deterministic Polynomial (NP) optimization problem, which can be an NP-hard problem. The continuous development in cloud computing and its complexity have made the problem even more challenging. Also, the problem of scheduling optimization is becoming an indispensable topic in academia. In this paper, a new hybrid metaheuristic technique based on Firefly Algorithm (FA), Particle Swarm Optimization (PSO) and Tabu Search (TS) is proposed to enhance task scheduling in resource intensive IoT systems. For the proposed algorithm, there can be a new and complete scheme to handle the task scheduling problems that were designed. To this, the TS algorithm can be incorporated aiming to lookout for the local optimum of every individual. For the purpose of improving solution quality, in every hybrid algorithm step, there has been an effective heuristic that has been proposed. This heuristic can bring down the other overtime costs by means of efficiently using the slack of the operation. Deep Reinforcement Learning (DRL) can be employed to solve problems in resource allocation and time scheduling thus making it easy to handle several tasks and resource heterogeneity. The experimental results demonstrated that the proposed methods (PSO-TS & FA-TS) achieved better performance compared to the other methods.