• cloud computing
• deep-learning
• LSTM
• time series forecasting
• virtual machine forecasting

Understanding the behavior of virtual machines in a cloud computing infrastructure is crucial for their effective management. In particular, an accurate forecasting of the VM load represents a valuable support to the design of scheduling algorithms for an optimized exploitation of hardware resources in cloud IaaS systems.

This thesis explores possible solutions for accurately predicting the resource utilization of a set of VMs operating within a cloud computing infrastructure. The study includes a comparative analysis of several prediction models, ranging from traditional statistical approaches, such as the ARIMA family, to modern Deep Learning techniques, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). Furthermore, the thesis investigates a variety of classification methods to examine whether grouping VMs according to behavioral or workload characteristics can lead to the development of customized prediction strategies, tailored to improve forecasting accuracy and support more efficient scheduling and resource management within the cloud infrastructure.