Forecasting Server Resource Utilization: An Experimental Study with Long Short-Term Memory Model

Cloud computing has indeed been a transformative milestone in the computing landscape, reshaping traditional practices. However, the expansion of cloud computing faces its hurdles, with forecasting computing resource utilization emerging as the top challenge. This research addresses the crucial need for enhanced monitoring and detection of server crashes in cloud computing environments. Leveraging deep learning techniques, the proposed Adaptive Transfer Learning model with Long Short-Term Memory offers a promising solution to better forecast server crashes. The study investigates the complexity and challenges of multivariate time series forecasting, particularly in datasets containing variables like CPU usage, Free Memory, and Free Drive Space. Through the integration of transfer learning and adaptive tuning, the proposed framework aims to overcome challenges such as missing data and dynamic resource utilization trends. The paper presents a comprehensive overview of the experimental workflow, results, and analysis, which contribute to the improvement of server crash forecasting and resource utilization monitoring in cloud computing environments. Based on several experiments and evaluations, the proposed framework demonstrated significant improvements in forecasting accuracy, achieving a reduction in error rate.