Rack Inlet Air Temperature Research Articles

Comparison of data driven modeling approaches for temperature prediction in data centers

Energy-efficient thermal management of data centers based on dynamic optimization and provisioning of cooling resources requires rapid (nearly real-time) predictions of temperatures within data centers. This work for the first time compares multiple Data-Driven Models (DDMs) to achieve such rapid temperature predictions. DDM typically employs statistical or machine learning-based tools, in combination with physics-based modeling and/or experimental data to predict system behavior. In general, DDM approaches are well-suited to systems that have multiple operational states based on interactions between the many electrical, mechanical and control parameters typical of data centers.This study compares the performance of three different DDM methods, namely Artificial Neural Networks (ANN), Support Vector Regression (SVR), Gaussian Process Regression (GPR) in predicting both steady-state and transient rack inlet air temperature distributions in data centers. Additionally, Proper Orthogonal Decomposition (POD) was considered for transient modeling. The data used for training and analysis were obtained by performing 300 offline numerical simulations with a room-level, experimentally validated computational fluid dynamics/heat transfer (CFD/HT) model.The performance of the four data-driven models was evaluated based on the absolute mean error for interpolation and extrapolation, and the adaptability of the models to changes in physical domain (data center room) configuration. Additionally, the impact of the size of the training data set on prediction accuracy is also compared for the four models.For the steady-state case study, the predictions for ANN, SVR and GPR models are in good agreement with CFD/HT simulations, with the GPR model having the smallest overall average prediction error of 0.6 °C in rack inlet air temperature, corresponding to a relative error of 2.7% with respect to rack inlet temperature measured in °C. It was found that for all the frameworks the prediction error increases when the size of training data set was less than 300 samples. The GPR model had the best accuracy for smaller training data sets compared with the other models, with an average prediction error for rack inlet temperatures <1 °C when trained on only 50 simulations. For the transient case study, the interpolative prediction error for all the models is very low (<0.3 °C); however, the extrapolative prediction errors are much greater, and appear to be directly proportional to the (here, temporal) “distance” from the interrogation point to the input parameter space.

From Chip to Cooling Tower Data Center Modeling: Influence of Server Inlet Temperature and Temperature Rise Across Cabinet

To achieve reductions in the power consumption of the data center cooling infrastructure, the current strategy in data center design is to increase the inlet temperature to the rack, while the current strategy for energy-efficient system thermal design is to allow increased temperature rise across the rack. Either strategy, or a combination of both, intuitively provides enhancements in the coefficient of performance of the data center in terms of computing energy usage relative to cooling energy consumption. However, this strategy is currently more of an empirically based approach from practical experience, rather than a result of a good understanding of how the impact of varying temperatures and flow rates at rack level influences each component in the chain from the chip level to the cooling tower. The aim of this paper is to provide a model to represent the physics of this strategy by developing a modeling tool that represents the heat flow from the rack level to the cooling tower for an air cooled data center with chillers. This model presents the performance of a complete data center cooling system infrastructure. After detailing the model, two parametric studies are presented that illustrate the influence of increasing rack inlet air temperature, and temperature rise across the rack, on different components in the data center cooling architecture. By considering the total data center, and each component’s influence on the greater infrastructure, it is possible to identify the components that contribute most to the resulting inefficiencies in the heat flow from chip to cooling tower and thereby identify the components in need of possible redesign. For the data center model considered here it is shown that the strategy of increasing temperature rise across the rack may be a better strategy than increasing inlet temperature to the rack.

Experimental-Numerical Comparison for a High-Density Data Center: Hot Spot Heat Fluxes in Excess of 500 W/ft $^{2}$

This paper uses previously published experimental data to present a comparison between test results and numerical simulations. The example considered is a large 7400 ft2 data canter that houses over 130 heat-producing racks (1.2 MW) and 12 air conditioning units. Localized hot spot heat fluxes were measured to be as high as 512 W/ft2 (5.5 kW/m2 ) for a 400 ft2 (37 m2 ) region. A numerical model based on computational fluid dynamics (CFD) was constructed using inputs from the measurements. The rack inlet air temperature was considered to be the basis for experimental versus numerical comparison. The overall mean rack inlet temperature predicted numerically at a height of 1.75 m is within 4degC of the test data with a rack-by-rack standard deviation of 3.3degC.

Optimization of Data Center Room Layout to Minimize Rack Inlet Air Temperature

In a typical raised floor data center with alternating hot and cold aisles, air enters the front of each rack over the entire height of the rack. Since the heat loads of data processing equipment continue to increase at a rapid rate, it is a challenge to maintain the temperature of all the racks within the stated requirement. A facility manager has discretion in deciding the data center room layout, but a wrong decision will eventually lead to equipment failure. There are many complex decisions to be made early in the design as the data center evolves. Challenges occur such as optimizing the raised floor plenum, floor tile placement, minimizing the data center local hot spots, etc. These adjustments in configuration affect rack inlet air temperatures, which is one of the important keys to effective thermal management. In this paper, a raised floor data center with 12kW racks is considered. There are four rows of racks with alternating hot and cold aisle arrangement. Each row has six racks installed. Two air-conditioning units supply chilled air to the data center through the pressurized plenum. Effect of plenum depth, floor tile placement, and ceiling height on the rack inlet air temperature is discussed. Plots will be presented over the defined range. A multivariable approach to optimize data center room layout to minimize the rack inlet air temperature is proposed. Significant improvement over the initial model is shown by using a multivariable design optimization approach.

Cluster of High-Powered Racks Within a Raised-Floor Computer Data Center: Effect of Perforated Tile Flow Distribution on Rack Inlet Air Temperatures

This paper focuses on the effect on rack inlet air temperatures as a result of maldistribution of airflows exiting the perforated tiles located adjacent to the fronts of the racks. The flow distribution exiting the perforated tiles was generated from a computational fluid dynamics (CFD) tool called Tileflow (trademark of Innovative Research, Inc.). Both raised floor heights and perforated tile-free areas were varied in order to explore the effect on rack inlet temperatures. The flow distribution exiting the perforated tiles was used as boundary conditions to the above-floor CFD model. A CFD model was generated for the room with electronic equipment installed on a raised floor. Forty racks of data processing (DP) equipment were arranged in rows in a data center cooled by chilled air exhausting from perforated floor tiles. The chilled air was provided by four A/C units placed inside a room 12.1 m wide×13.4 m long. Because the arrangement of the racks in the data center was symmetric, only half of the data center was modeled. The numerical modeling for the area above the raised floor was performed using a commercially available finite control volume computer code called Flotherm (trademark of Flomerics, Inc.). The flow was modeled using the k-e turbulence model. Results are displayed to provide some guidance on the design and layout of a data center.