OPTIMIZING CLOUD PERFORMANCE: A SELF-ADAPTIVE EVOLUTIONARY METAHEURISTIC-BASED LOAD BALANCING AND RESOURCE ALLOCATION FRAMEWORK WITH DEEP REINFORCEMENT LEARNING

Abstract

Cloud service providers (CSPs) must manage cloud resources efficiently to optimize resource utilization and increase user application efficiency at the lowest possible cost. The increased resource reservations result in higher resource usage costs, but they improve user application needs' performance. This paper developed a hybrid evolutionary algorithm and a unique dynamic load-balancing architecture to get better load balancing (LB) and effective resource allocation (RA). Initially, compute Virtual Machine (VM) loads and cluster them using a self-adaptive evolutionary algorithm-based clustering technique. Hybrid optimization methods like Owl Optimization Algorithm (OOA) and Wild Goose Optimization (WGO) are used for load computation. Then, it introduces a multi-stage technique for optimal task allocation with Deep Reinforcement Learning (DRL) and metaheuristics. The DRL assigns tasks to underloaded VMs, and then evolutionary techniques are used to refine the assignments. The suggested approach has shown improved efficiency and scalability by providing a suitable solution to the problems associated with LB and job scheduling in Cloud Computing (CC) contexts. The developed model gained better results in makespan, energy consumption, and task prioritization.