Ainet0: AI Forecasting Based Carbon Neutral Cloud Resource Management for Net Zero Targets

Abstract

Cloud computing is increasingly integrating artificial intelligence (AI) techniques to optimize resource management and address the complex and fluctuating demands of modern cloud environments. In particular, accurate CPU usage prediction-achieved through appropriate time-series forecasting, is crucial for minimizing both over-provisioning and under-provisioning of cloud resources. High predictive accuracy not only mitigates operational inefficiencies but also aligns with net-zero emission goals by reducing unnecessary energy consumption. To address these challenges, we propose a new framework AINet0, which uses AI forecasting models such as neural networks and time-series models for a dynamic resource management approach. The AINet0 integrates and examines Amazon's DeepAR+ and Convolutional Neural Network-Quantile Regression (CNN-QR) within the Amazon Web Services environment, aiming to enhance energy efficiency and sustainability to enable carbon-neutral cloud services. We evaluated the AINet0 framework using a real testbed, CloudAIBus, and compared its performance against established baselines. The experimental results demonstrate that the AINet0 enhances forecasting accuracy, achieving lower metrics such as Mean Absolute Percentage Error (MAPE), allowing for more precise resource allocation. Furthermore, performance evaluation conducted on the Grid Workload Archive-T-12 dataset shows that the proposed models outperform traditional methods, enabling more sustainable and cost-effective operations that further support net-zero emissions objectives.