Performance Evaluation and Machine Learning based Thermal Modeling of Tilted Active Tiles in Data Centers

Abstract

Thermal management system of data center continuously face a lot of challenges, because data center industry has seen a boom growth in power density. In this paper we proposed the Tilted Active Tiles (TATs) to improve the local cold air supply and prevent the air flow blow over the rack. In traditional active tiles, fans are placed horizontally which cause the airflow blows over the rack, rather than into, the racks. To solve this issue, we adjusted the angle of the active tile to direct the airflow into the rack. We further introduced ANN based thermal models to predict the thermal performance of TATs. To train the ANN models, we adopted the data set obtained from a data center of Inner Mongolia Meteorology Information Center. The prediction accuracy of the model was extensively compared and analyzed, and the prediction accuracy and overhead of different neural network structures, i.e., BP and LSTM, were evaluated. Experimental results show that the rack with blanking panels has better thermal performance, and the temperature distribution at bottom, middle and top of the rack were same under smaller PWM. Thermal efficiency model was established by BP and LSTM, in this experiment single output model and multi output model were analyzed. The single output model can predict the temperature at different heights on the rack. In single output model the predicted effect of BP model is better than LSTM. The average prediction error is 0.57. The multi-output model can only predict the temperature at a fixed height of the rack. In multi output model LSTM model is better than BP. LSTM prediction error is less than BP. The average prediction error is 0.07.