This paper introduces a methodology for developing a reduced order model, using proper orthogonal decomposition (POD), to predict the IT rack's inlet temperature distribution within a raised floor air-cooled data center. The method used in this paper uses a limited set of computational fluid dynamics data at different useful IT levels and tile airflow fractions. The model was able to reconstruct these datasets to with 0.16 °C rms error and interpolate successfully for alternative configurations that were not included in the original dataset. The reduced order model can produce the temperature distribution in the data center in a fraction of a second on a standard personal computer. Several practical IT load placement options in open-aisle, air-cooled data centers, based on either geometrical traits of the data center, a prior physics-based knowledge of the airflow and temperature patterns or measurements that are easily obtainable during operation, are considered. The outcome of this work is the development of a robust set of guidelines that facilitate the energy efficient placement of the IT load amongst the operating servers in the data center. This work found that a robust approach was to use real-time temperature measurements at the inlet of the racks to remove the unnecessary IT load from the servers with the warmest inlet temperature. This strategy shows superior performance to the other strategies studied. The study considered the holistic optimization of the data center and cooling infrastructure for a range of data center IT utilization levels. The results showed that allowing for significant reductions in the supply air flow rate proved superior to providing a higher supply air temperature to meet the IT equipment's inlet air temperature constraint.
Inverse thermal design technique for predetermined temperature distribution using reduced order model of 3D heat conduction simulation and artificial bee colony algorithm
