Hybrid DRL-Enhanced ACO-WWO for Efficient Resource Allocation and Load-Balancing in Cloud Computing

Abstract

Abstract The growing complexity of cloud computing necessitates astute workload allocation and adaptive resource management to enhance performance while minimizing expenses and energy consumption. Conventional optimization methods, including Improved Ant Colony Optimization (IACO) and Water Wave Optimization (IWWO), face challenges in real-time adaptability, exhibit slow convergence, and are inadequate for managing rapidly varying workloads. Although IACO enhances local search efficiency and IWWO specializes in global exploration, neither adequately resolves the complexities of dynamic cloud environments. To address this gap, we propose a Hybrid DRL-IACO-IWWO model, a novel hybrid model that combines DRL with advanced iterations of IACO and IWWO. The model presents an adaptive dual-phase optimization strategy, wherein IACO conducts initial task scheduling, and IWWO enhances global optimization, informed by real-time DRL feedback. Furthermore, DRL dynamically adjusts its heuristic parameters to improve operational cost and energy efficiency, ensuring real-time adaptability. To expedite convergence, our model utilizes a wavelet transformation-based perturbation in WWO, thereby preventing premature convergence and promoting a more balanced equilibrium between exploration and exploitation. An energy-efficient scheduling mechanism is integrated to reduce energy consumption and improve cloud sustainability. The proposed model was evaluated using the workflow dataset, considering constraints, such as task deadlines, resource availability, and cost efficiency. The results indicate that our methodology outperforms leading hybrid techniques, such as ACO-GA, ACO-SMO, and WWO-GA. The proposed model achieved a scheduling duration of 1.25 s, compared to 1.75 s for ACO-GA and 1.68 s for WWO-GA, while reducing operational expenses to $23.80, lowering energy consumption to 15.6 kWh, and achieving a resource utilization score of 0.92. These findings underscore the transformative capacity of our Enhanced ACO-WWO with DRL, offering a highly efficient, cost-effective, and adaptive solution for next-generation cloud resource management.

Keywords

• Improved ACO-WWO
• Deep reinforcement learning
• Cloud resource optimization
• Adaptive scheduling
• Energy-efficient computing
• Hybrid metaheuristics