Failures Forecast in Monitoring Datacenter Infrastructure Through Machine Learning Techniques: A Systematic Review.

With the trend of accelerating digital transformation processes, datacenters (DC) are gaining prominence as increasingly critical components for business success. Modern DC are complex systems. To maintain the operation with high efficiency and high availability, it is necessary to carry out detailed monitoring, which easily results in hundreds or thousands of items being monitored. Also, the large number of possible configurations is a challenge for monitoring and correcting failures on time. However, recent advances in the field of Artificial Intelligence (AI), especially in the area of Machine Learning (ML), have been generating unprecedented opportunities to improve the efficiency of the analysis of historical monitoring data, facilitating the recognition of patterns and enabling scenarios for early detection of failures. In this sense, significant research has been published discussing the applications of ML techniques in the context of DC monitoring. Based on this context, in this paper, we aim to present a systematic literature review (SLR) that helps to understand the current state and future trends of the application of ML techniques in DC monitoring, specifically those aimed at early fault detection. This SLR also aims to identify gaps for further investigations. As main results, we identified 51 papers reporting unique studies published in conferences and journals between 2009 and 2020. Most of the works (60%) were applied in supervised algorithms, in which the most used algorithm was the Random Forest (19,60%). The main types of data used were S.M.A.R.T attributes (14 papers) and log data (10 papers).