Elastic cloud resource allocation using short-term long short-term memory-based workload prediction

This paper proposes a hybrid resource allocation system for cloud computing, combining traditional allocation algorithms with workload prediction methods based on deep learning. The existing approaches to task distribution in cloud systems and their limitations are analyzed. The use of Long Short-Term Memory (Long short-term memory) networks for short-term server load forecasting is investigated, enabling traditional algorithms such as MinMin, Round Robin, and Weighted Allocation to dynamically adapt to changing workloads. It is found that integrating predictions enhances resource utilization, reduces load imbalance, and improves system resilience. Recommendations for integrating predictive methods into existing resource allocation systems are formulated. The use of synthetic data and simulations to evaluate system performance is suggested, showing significant improvements compared to traditional approaches.