Modeling temperature distribution and power consumption in IT server enclosures with row-based cooling architectures

Traditional data center cooling methods cannot yet control cooling airflows and temperatures on demand, creating an intrinsic inefficiency. A recent solution places row-based cooling unit adjacent to servers and places the entire assembly within an enclosure, which improves airflow distribution and provides rapid real-time control. This is, in particular attractive for micro-data centers where traditional room-based cooling is less energy efficient. Spatiotemporal predictions of temperatures are required to control and optimize data center performance as the system configuration, and other parameters are varied. Current methods, such as proper orthogonal decomposition,machine learning, and heuristic models are inapplicable in practice because they require a prohibitively large number of a priori simulations or experiments to generate training datasets. We provide an alternative a computationally inexpensive training-free model for enclosedmicro-data centers that are integrated with in-row cooling units that requires no a priori training. The model determines the air flowrate within each zone based on a mechanical resistance circuit analysis. These flowrates are then introduced into a zonal energy balance to predict the temperature of each zone. The methodology is validated with experimental measurements and coupled with a power consumption calculation. Its applicability is demonstrated by evaluating the influence of various system factors, such as IT server configurations, cooling unit air, and water flowrates and the numbers of cooling units, on the temperature distributions, and total cooling power consumption. The method can improve micro-data centers control and help to optimize the design of any data center with a row-based cooling system.