Transient power unit efficiency prediction of a small data center through thermal and energetic analyses

Data Centers (DCs) are proliferating throughout the world, and their energy consumption is increasing. Therefore, optimizing the energy efficiency of the server environment becomes crucial. For this purpose, the usual key metric is the Power Unit Efficiency (PUE). However, this measurement only occurs when a Data Center (DC) becomes operational. The current study builds a methodological tool to estimate PUE~ metrics before the DC is built, by bringing together concepts from Information Technology (IT) and energy laws. Firstly, a consumption study is carried out on a small university DC with a power of 140 kW. An energy behavior pattern is drawn based on 5 minutes time-step data collected over a recent period of 7 years, from 2016 to 2023, making this dataset relevant for a benchmark of historical DC server energy consumption. Secondly, a dynamic model is built to integrate the variable energy behavior and the weather-dependent system environment. In this way, the short time PUEτ reflects interesting non-constant values, exhibiting maximum and minimum performances of the DC. The results of the dynamic model are compared to the actual measurements and show a deviation of 5% between measurements and predictions. Temporal variations of PUE are strongly correlated with temperature variations during a day. In contrast, server environments exhibit an almost constant consumption, except for High Performance Computing (HPC) during the working days, for which the energy demand is higher than during nights. This approach is also used to make long-term predictions evaluating the impact of global warming on DC efficiency as an example. In this study, a prediction was made for the years 2035 and 2065 and could be easily implemented by manufacturers as a reliable method to estimate PUE~ during the pre-construction and projection phases of DC in a context of rapid climate change.