EN-Beats: A Novel Ensemble Learning-Based Method for Multiple Resource Predictions in Cloud
Ming Chen, Maria A. Rodriguez, Patricia Arroba, Rajkumar Buyya · 2023
Cloud computing has become an important driving force in the economy and the fundamental facility for digitization transformation. Due to its rapid development and increasing demands, accurate resource usage prediction for the cloud has been a long-term challenge. To address this challenge, this paper introduces RCorrPolicy for resource metrics selection and proposes EN-Beats, an efficient ensemble learning-based approach, for predicting multiple resource usages in the cloud. The paper presents trace-driven experiments conducted on a real-world dataset, demonstrating notable improvements in predicting multiple resource metrics. Ablation experiments conducted on existing methods for RCorrPolicy indicate that the proposed policy enhances the performance of these methods across different evaluated metrics. Furthermore, EN-Beats outperforms existing methods by achieving the lowest NRMSE (lower values indicate better performance) for CPU util rate (up to 8% lower), memory usage (up to 6% lower), and network incoming traffic (up to 3% lower). Additionally, EN-Beats attains the highest R2 score (higher values indicate better performance) for predicting CPU util rate (up to 1.39 higher), memory usage (up to 0.57 higher), and network incoming traffic (up to 0.62 higher).