The Performance Impact of Integrating Water Storage Into a Chiller-Less Data Center Design

Abstract

Data centers consume an extraordinary amount of electricity, and the rate of consumption is increasing at a rapid pace. Thus, energy efficiency in data center design is of substantial interest since it can have a significant impact on operating costs. The server cooling infrastructure is one area which is ripe for design innovation. Various designs have been considered for air-cooled data centers, and there is growing interest in liquid-cooled server designs. One potential liquid-cooled solution, which reduces the cost of cooling to less than 5% of the information technology (IT) energy use, is a chiller-less or warm water-cooled system, which removes the chiller from the design and lets the cooling water supply vary with changes in the outdoor ambient conditions. While this design has been proven to work effectively in some locations, environmental extremes prevent its more widespread implementation. In this paper, the design and analysis of a cold water storage system are shown to extend the applicability of chiller-less designs to a wider variety of environmental conditions. This can lead to both energy and economic savings for a wide variety of data center installations. A numerical model of a water storage system is developed, validated, and used to analyze the impact of a water storage tank system in a chiller-less data center design featuring outdoor wet cooling. The results show that during times of high wet bulb operating conditions, a water storage tank can be an effective method to significantly reduce chip operating temperatures for warm water-cooled systems by reducing operating temperatures 5–7 °C during the hottest part of the day. The overall system performance was evaluated using both an exergy analysis and a modified power usage effectiveness (PUE) metric defined for the water storage system. This unique situation also necessitates the development of a new exergy definition in order to properly capture the physics of the situation. The impacts of tank size, tank aspect ratio, fill percentage, and charging/discharging time on both the chip temperature and modified PUE are evaluated. It is determined that tank charging time must be carefully matched to environmental conditions in order to optimize impact. Interestingly, the water being stored is initially above ambient, but the overall system performance improves with lower water temperatures. Therefore, heat losses to ambient are found to beneficial to the overall system performance. The results of this analysis demonstrate that in application, data center operators will see a clear performance benefit if water storage systems are used in conjunction with warm water cooling. This application can be extended to data center failure scenarios and could also lead to downsizing of equipment and a clear economic benefit.