A Neighborhood Inspired Multiverse Scheduler for Energy and Makespan Optimized Task Scheduling for Green Cloud Computing Systems

• Published in: IEEE access 2024, Vol.12, p.157272-157298
• Main Authors: Tiwari, Shalini, B. M., Beena
• Format: Article
• Language: English
• Subjects: Cloud computing; Convergence; Costs; Energy consumption; Green cloud computing; Green products; local search; Metaheuristics; multiverse optimizer (MVO); Nearest neighbor methods; neighborhood search; Optimization; Processor scheduling; task scheduling; Throughput
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3484388

In the digital era, cloud computing is vital for scalable and efficient infrastructure, but its growing energy consumption raises serious environmental concerns. Green cloud computing strategies, particularly efficient task-scheduling algorithms, are key to addressing this challenge. Task scheduling in cloud computing is NP-hard due to the complexity of managing numerous tasks, resources, and optimization metrics. To address this, we propose a novel task scheduling algorithm named NIMS (Neighborhood Inspired Multiverse Scheduler), designed to optimize two often conflicting metrics: makespan and energy consumption. NIMS improves the performance of the original MVO (Multiverse Optimizer) by incorporating a novel fitness-based neighborhood search strategy during solution updates. This enhancement improves the quality of solutions, particularly when the standard update mechanism of MVO underperforms. By promoting a more effective exploration of the solution space, our approach significantly enhances the algorithm's convergence rate. Further, we performed a comprehensive performance evaluation of the proposed NIMS algorithm against seven advanced algorithms: EMVO, IMOMVO, OPSO, LJFPPSO, TSIGA, FPGWO, and MVO, using the CloudSim toolkit under various test scenarios, leveraging three widely adopted real-world benchmark datasets. Our extensive simulations and experiments exhibit that the proposed NIMS algorithm significantly outperforms the competing algorithms across five key performance metrics: makespan, energy consumption, throughput, load imbalance, and average resource utilization.