On Machine Learning for Cloud Resource Management

The dynamicity and heterogeneity of cloud resources require significant expertise and manual labour to properly configure these resources for delivering acceptable quality of service levels to the end users. Cloud resource configuration and orchestration is, in facts, challenging because it requires dynamic coordination across different infrastructure layers. For this reason, we need to automate the whole resource management process and Machine Learning (ML) tools have the potential to facilitate such automation. Moreover, future cloud infrastructure will be more and more dynamic and heterogeneous and for this reason we need ML algorithms able to dynamically adapt and change their parameters according to the new data. In this position paper we classify and analyses the more suitable ML algorithms that fit in such highly-heterogeneous and high-dynamic scenario.