Adaptive Evolutionary Neural Network Model for Cloud Resource Reservation and Management

Ensuring elasticity and scalability in cloud data centers necessitates accurate prior resource reservation, which relies on precise future workload estimation. This task is challenging due to the high variability in resource demands, including sudden spikes, drops, noise, and redundancy. To address this, an Adaptive Evolutionary Neural Network (AENN) model is proposed for predicting average resource usage over successive intervals. Utilizing historical data, the model learns workload patterns through an adaptive Differential Evolution algorithm. The performance evaluation using benchmark Google workload traces of CPU and memory usage shows that our model significantly outperforms existing methods, with significant accuracy improvements over Backpropagation and Support Vector Machine-based prediction methods, respectively.