Proposta de um sistema inteligente de monitoramento baseado em seleção de características

Abstract

This work builds upon the Slice-as-a-Service paradigm proposed in the Novel Enablers for Cloud Slicing (NECOS) project. Assuming the provision of end-to-end slices which are composed by resources coming from multiple infrastructure providers, this work proposes an intelligent monitoring system, aiming to dynamically select a set of features that best fits the real life time management needs of the slices, keeping the accuracy of that management. We want to avoid the movement of unnecessary information from the infrastructure providers, delivering essential data to management functions. Presenting the experiments made in our testbed, which compared both algorithms of correlation and machine learning, it’s possible to notice that the data reduction, through the feature selection, increases satisfactorily the quality of service estimative precision.