Big Data Analysis for Predictive Maintenance at the INFN-CNAF Data Center using Machine Learning Approaches

Predictive maintenance is a hot topic in research. It is widely applicable to the field of supporting and monitoring computing systems with the twofold objectives of increasing the operational efficiency and reducing costs through the prediction of faults. The ultimate goal of this project is to build a complete predictive maintenance solution for data centers, based on the extraction of content from the log files of services running onsite. Currently, the major Italian Worldwide Large Hadron Collider (LHC) Computing Grid data center is the INFN-CNAF, placed in Bologna, that mainly relies on reactive maintenance. In order to improve the data center quality of services (QoS), this work uses the log data as feedstock. Typically, log files are unstructured or semi-structured files, with heterogeneous data, revealing information about the system status that could be useful for its maintenance. In general, because of its characteristics, this kind of data is hard to process with standard Machine Learning (ML) supervised solutions without a deeply time and resource-consuming solution. In addition, this information can be complemented with collected environmental data to further refine event predictions or system diagnostics.