Resource Management Utilization Prediction Technique in Cloud using Transformer Models for Workload Forecasting

Abstract

In recent years, the rapid adoption of cloud computing has transformed how organizations manage and allocate computing resources. As businesses increasingly rely on cloud infrastructure for its scalability and flexibility, the need for accurate predictions of resource utilization has become critical. This study investigates the optimization of cloud resource management through the application of transformer model techniques, addressing the growing need for accurate resource utilization predictions in cloud computing. In comparison to previous works that primarily utilized basic statistical methods and simpler machine learning models, achieving lower accuracy. This study demonstrates that the transformer model significantly enhances prediction accuracy for CPU and memory utilization. The experimental results indicating a 25% reduction in Mean Absolute Error (MAE) and a 30% decrease in Root Mean Square Error (RMSE) are compared to previous approaches that utilized basic statistical methods and simpler machine learning models for resource utilization forecasting. This improvement enables more effective dynamic resource allocation, resulting in a 15% reduction in operational costs by minimizing both over-provisioning and under provisioning. Moreover, the integration of the transformer model within Kubernetes environments leads to enhanced scalability and performance. The application response times improved by 20% during peak demand, showcasing a marked advantage over earlier studies that did not achieve such efficiency gains. Additionally, the model maintains a high Quality of Service (QoS), further ensuring compliance with SLAs. Finally, the findings underscore the transformative potential of transformer models in cloud resource management, highlighting significant advancements over previous methodologies. This research paves the way for future explorations into leveraging advanced machine learning techniques to optimize cloud infrastructure even further