Enhancing Cloud Resource Allocation with a Hybrid Deep Learning-Based Framework: A Comparative Study

Abstract

Optimal resource provisioning remains a core problem in cloud computing, particularly in dynamic and heterogeneous environments. Traditional provisioning techniques are not very agile in adapting to varied workloads and user demands in real-time, resulting in inefficient underutilization or overprovisioning of resources. In this paper, a hybrid deep learning-enabled method to enhance cloud resource provisioning using predictive modeling and discerning decision-making is presented. The architecture blends Convolutional Neural Networks (CNNs), Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM) networks to learn effectively spatial and temporal connections in historical workload and resource usage data. High-level spatial features are learned from multi-dimensional input matrices using CNNs. In contrast, temporal dynamics and long-term dependencies are learned using GRU and LSTM units, enabling better prediction of future resource demands. Experimental comparisons on standard cloud simulation data sets show that the proposed hybrid model significantly outperforms traditional deep learning and rule-based policy allocation policies in terms of allocation accuracy, task completion time, and system throughout. The findings highlight the potential of advanced deep learning models to enhance resource allocation optimization, reduce running costs, and ensure enhanced service quality in real-time cloud computing.