Study on Prediction of Rack Temperature in Server Room Using Machine Learning

Abstract

In recent years, data centers (DC) have become increasingly important. Accordingly, highly efficient and reliable operation and management of DCs have been required. The conventional layout design of racks and ICT equipment in a server room considers the current space and current capacity of power and cooling status, but it does not take into account the temperature or energy performance after environmental changes such as rack removals or additions. On the other hand, for temperature management in the server room, the DC users usually requires DC operator to comply with a service level agreement (SLA) that stipulates that the rack intake temperature is to be kept below a certain level. However, it is not possible to know temperature after environmental changes. To address this problem, we have constructed some models to predict rack intake air temperatures in a server room by using information of ICT equipment, electric power equipment, and air conditioning equipment. In this paper, we propose a method of predicting the temperature after environment changes by using machine learning are developed, and the results of verification and effectiveness in the verification room are reported.