Due to the rapid growth of the power density in data centers, the thermal management of the data center has become more and more important. Computational Fluid Dynamics has been proven to be one of the most effective tools in the design and analysis in data centers due to its ease of use and fast prediction in data centers. Due to the large size of a data center facility, one of the most challenging tasks in the data center CFD simulation is to establish simplified models to provide a fast but accurate prediction for both the temperature and flow field in data centers. In the past few years, people have proposed many tile models specifically to simulate the jet effect of the perforated tile. However, other aspects of the simulation were still in the early stage, e.g. rack models and the effect of buoyancy forces. This paper mainly studied the effect of the three rack models, which were Black Box Model, Modified Black Box Model, and Volumetric Heat Model. The effects of the buoyancy force were also studied. The result showed that buoyancy force had a huge impact on the simulation result, while the three rack models investigated had little difference.

Data centers in recent years have grown so fast so that their energy consumptions become a big issue in the industrial sector. One of the strategies to make better use of energy in data centers is to improve the efficiency in cooling. As the load density in data centers increases dramatically over the years, the number of computer room air handlers (CRAHs) are also increased to accommodate the high cooling demands. However, the number of CRAH units and their layouts really affect the air flow through the perforated tiles. Non-uniform airflow distributions in the perorated tiles in the cold aisles cause inefficient cooling of all the servers mounted in racks in data centers. Application of necessary strategies to minimize airflow non-uniformity is therefore very important because of its direct impact on the power density capacity. In this paper, a simulation study to examine how computer room air handler (CRAH) positions, the number of operating units, and tile types affect the airflow uniformity in selected data center models. Also, the placement of mixed tiles in cold aisles to regulate the airflow through the perforated tiles to accommodate greater heat loads from server racks is evaluated.

Energy Conservation Measures (ECMs) of the primary data center at the University of Maryland are developed. Measurement and simulation are performed to validate the developed ECMs. Three ECMs — 1) Increase in the return temperature at Computer Room Air Conditionings (CRACs) 2) Cold aisle containment 3) Elimination of unnecessary CRACs — are suggested to reduce energy consumption by optimizing the thermo-fluid flow in the data center. Power savings of 12.7 kW – 17.4 kW and 14.1 kW are obtained by increasing the return air temperatures at the CRACs and performing the cold aisle containment, respectively. In addition, a power saving of 11.2 kW is obtained by turning off CRACs 3 and 8 which have an adverse effect on the data center cooling.

Recent developments in applications such as computer data centers, electric vehicle power electronics, avionics, radars and lasers have led to alarming increases in heat dissipation rate, which now far exceeds the capability of air cooling schemes and even the most aggressive single-phase liquid cooling schemes. This trend is responsible for a recent transition to two-phase cooling, which capitalizes upon the coolant’s latent heat rather than sensible heat alone to achieve several order-of-magnitude increases in heat transfer coefficient. Three two-phase cooling configurations have surfaced as top contenders for the most demanding applications: mini/micro-channel, jet and spray. This study will explore the implementation of these configurations into practical cooling packages, assess available predictive tools, and identify future research needs for each. It is shown that the design and performance assessment of high-flux, two-phase cooling systems are highly dependent on empirical or semi-empirical predictive tools and, to a far lesser extent, theoretical mechanistic models. A major challenge in using such tools is the lack of databases for coolants with drastically different thermophysical properties, and which cover broad ranges of such important parameters as flow passage size, mass velocity, quality and pressure. Recommendations are therefore made for future research to correct any critical knowledge gaps, including the need for robust computer algorithms. Also discussed is a new class of ‘hybrid’ cooling schemes that capitalize upon the merits of multiple cooling configurations. It is shown that these hybrid schemes not only surpass the basic cooling configurations in heat dissipation rate, but they also provide better surface temperature uniformity.

Due to the tremendous cooling costs, data center cooling efficiency improvement has been actively pursued for years. In addition to cooling efficiency, the reliability of the cooling system is also essential for guaranteed uptime. In traditional data center cooling system design with N+1 or higher redundancy, all the computer room air conditioning (CRAC) units are either constantly online or cycled according to a predefined schedule. Both cooling system configurations, however, have their respective drawbacks. Data centers are usually over provisioned when all CRAC units are online all the time, and hence the cooling efficiency is low. On the other hand, although cooling efficiency can be improved by cycling CRAC units and turning off the backups, it is difficult to schedule the cycling such that sufficient cooling provisioning is guaranteed and gross over provisioning is avoided. In this paper, we aim to maintain the data center cooling redundancy while achieving high cooling efficiency. Using model-based thermal zone mapping, we first partition data centers to achieve the desired level of cooling influence redundancy. We then design a distributed controller for each of the CRAC units to regulate the thermal status within its zone of influence. The distributed controllers coordinate with each other to achieve the desired data center thermal status using the least cooling power. When CRAC units or their associated controllers fail, racks in the affected thermal zones are still within the control “radius” of other decentralized cooling controllers through predefined thermal zone overlap, and hence their thermal status is properly managed by the active CRAC units and controllers. Using this failure resistant data center cooling control approach, both cooling efficiency and robustness are achieved simultaneously. A higher flexibility in cooling system maintenance is also expected, since the distributed control system can automatically adapt to the new cooling facility configuration incurred by maintenance.

Cooling poses as a major challenge in the IT industry because recent trends have led to more compact and energy intensive microprocessors. Typically microprocessors in current consumer devices and state-of-the-art data centers are cooled using relatively bulky air cooled heat sinks. The large size heat sinks are required due to the poor thermophysical properties of air. In order to compensate for the poor thermal properties of air, it is typical to use chillers to pre-cool the air below the ambient temperature before feeding it to the heat sinks. Operating the chillers requires additional power input thereby making the cooling process more expensive. The growing cooling demand of electronic components will, however, render these cooling techniques insufficient. Direct application of liquid-cooling on chip level using directly attached manifold microchannel heat sinks reduces conductive and convective resistances, resulting in the reduction of the thermal gradient needed to remove heat. Water is an inexpensive, nontoxic and widely available liquid coolant. Therefore, switching from air to water as coolant enables a much higher coolant inlet temperature without in any way compromising the cooling performance. In addition, it eliminates the need for chillers and allows the thermal energy to be reused. All these improvements lead to higher thermal efficiency and open up the possibility to perform electronic cooling with higher exergetic efficiency. The current work explores this concept using measurements and exergetic analyses of a manifold microchannel heat sink and a small scale, first of its kind, hot water cooled data center prototype. Through the measurements on the heat sink, it is demonstrated that the heat load in the state-of-the-art microprocessor chips can be removed using hot water with inlet temperature of 60°C. Using hot water as coolant results in high coolant exergy content at the heat sink outlet. This facilitates recovering the energy typically wasted as heat in data centers, and can therefore result in data centers with minimal carbon footprint. The measurements on both the heat sink and the data center prototype strongly attest to this concept. Reuse strategies such as space heating and adsorption based refrigeration were tested as potential means to use the waste heat from data centers in different climates. Application-specific definitions of the value of waste heat were formulated as economic measures to evaluate potential benefits of various reuse strategies.

Heat transfer phenomena in complex physical systems like multiphase environments, multidimensional geometries can be difficult to capture in terms of correlations, analytical functions or numerical models using conventional techniques. Such systems are designed based on approximations, thumb-rules or semi-empirical correlations between parameters based on averaged values and are operated likewise using another set of rules derived from bulk thermodynamic performance parameters. With the development of nano-scale sensors and advanced data aggregation techniques, there is a need for analytical techniques that can discover the complex interrelationships between the thermodynamic parameters of the process, geometry constraints and the governing outcomes of the process. Such techniques can leverage the possibility of deployment of thousands of sensors to extract the key relationships that drive the transport phenomena for advanced development of process control tools and methodologies. Heat and mass transfer equipment design and operation can benefit from knowledge discovered through analytics applied on thermo-physical data obtained from real time processes. We present illustrative use cases of application of data analytics and knowledge discovery techniques to a richly instrumented data center where computer room air conditioning (CRAC) units provide cooling for IT equipment arranged in rows of racks. Sensors located at each rack provide temperature measurements which are analyzed in real-time and also archived. Rack temperatures, together with operating parameters of CRAC units such as supply air temperature (SAT), and variable speed drive (VFD) settings, are analyzed together to derive design insights and detect anomalies.

Early Data centers can consume 25 to 50 times more electric power than a standard office space of the same footprint. In this paper, a simplified computational fluid dynamics/heat transfer (CFD/HT) model for a unit cell of a data center with a hot aisle-cold aisle (HACA) layout is simulated. Inefficiencies dealing with the mixing of hot air present in the room with the cold inlet air, leading to a loss of cooling potential are identified. The need for a thermal aware job-scheduling algorithm which enhances IT productivity, while maintaining the facility within server inlet temperature constraints is established. The inherent non-linearity of such an optimization problem is explained. A novel algorithm called the Ambient Intelligence based Load Management (AILM) is developed which counters the above issues and enhances the net data center heat dissipation capacity for given energy consumption at the facilities end. It gives a scheme to determine how much and where the computer loads should be allocated, based on the differential loss in cooling potential per unit increase in server workload. Enhancements of heat dissipation capacity of over 50% are proved numerically for the representative values considered. An approach to incorporate heterogeneity in data centers, both for lower heat dissipation and liquid cooled racks has been established. Finally, different objective functions are studied and an ideal combination of the IT objectives and thermal constraints is derived.

This paper presents experimental data showing the response of a computer room air conditioning unit (CRAC) to chilled water (CHW) pump restart. The data are offered to improve and develop modeling of cooling equipment restart events following data center power failure. There are estimates that power failures will increase and limits on availability will affect data center operations at more than 90 percent of all companies over the next five years. Because providing backup power to cooling equipment increases data center first cost, it is important to have accurate models for cooling events and processes following power failure that help predict server inlet temperatures during the transient phase caused by a power failure. Since power density of computing equipment continues to rise, the temperature rise of air within the data center has been predicted to rise more quickly to an unacceptable level, increasing concern. Accurate models of CRAC response to pump restart can aid in data center cooling design, backup power infrastructure provisioning, and even compute equipment selection by predicting the air supply temperature after the generator provides power to the chilled water pump. Previous transient models include zonal models with large time scales and CFD/HT models with boundary conditions developed for steady state. These models can be improved by comparison with experimental data. The experiment consists of measuring the response of the CRAC heat exchanger to the step change in CHW flow rate upon pump restart. Inlet and outlet temperatures of both CHW and air were measured, as well CHW flow rate. A point measurement of air at the CRAC fan outlet was also taken to verify that airflow remained relatively constant. Outlet temperatures from the CRAC follow a first order response curve; it is found that the CRAC under consideration has fan outlet temperature time constant of 10 seconds. A delay of 20 seconds is observed between the fan outlet temperature response and the CHW return temperature response.

With the proliferation of multiple cores on a single processor and the integration of numerous stacked die in chip packages, the resulting non-uniformity in power dissipation necessitates spatially and temporally localized control of temperature in the package. Thermoelectric (TE) devices can potentially provide one mechanism to achieve such control. Unfortunately, at typical junction-to-ambient temperatures, the coefficient-of-performance (COP) of existing bulk TE devices tends to be quite low. As a result, for many high-power systems, the additional power input required to operate a TE cooling module can lead to an increase in the overall system cost-of-ownership, causing TE cooling solutions to be excluded from cost-sensitive thermal management solutions in high-volume computer systems. However, recent trends of compaction and consolidation of computer servers in high-density data centers have resulted in a dramatic increase in the burdened cost-of-ownership of mission-critical IT facilities. For example, the energy consumption of the cooling infrastructure for many high-density data centers can equal or exceed the power consumption of the compute equipment. Thus, for the growing enterprise thermal management segment, the appropriate metric is no longer the COP of the thermal solution but rather the COP of the cooling ensemble, which takes into account the energy efficiency of cooling solutions in the chip, system, rack, data center and facility. To examine the effects of chip-level COP on the ensemble-level COP, this paper explores a case study comparing two ensemble solutions. In one case, local hotspots on a chip in a three-dimensional package are mitigated by increasing fan power at the system level (and subsequently, in the computer room and the rest of the ensemble). In another case, local hotspots at the chip are mitigated through spot-cooling via TE cooling modules. For each of these cases, the COP of the ensemble is evaluated. The model suggests that while feasible, the benefit of using TEs at current performance levels is limited. However, ongoing research that may improve TE performance in the future has the potential to enhance infrastructure energy efficiency.

There is a strong need to improve our current capabilities in thermal management and electronic cooling, since estimates indicate that IC power density level could reach 50 W/cm 2 in near future. This paper presents several possible closed and open loop cooling schemes for thermal management of electronic equipment in data centers. Additionally, analytical modeling of two kinds of direct expansion refrigeration cooling evaporator and a secondary liquid cooling fan coil heat exchanger is incorporated with a computational fluid dynamics (CFD) model to analyze a refrigeration cooled high heat density electronic and computer data center installed on a raised floor. Both models incorporate an accurate tube-by-tube thermal hydraulic modeling of the heat exchanger. The refrigeration coil analysis incorporates a multi region heat exchanger analysis for a more precise modeling of two phase refrigerant flow in the evaporator. The single phase secondary loop fan coil heat exchanger modeling uses an effectiveness method for regional modeling of the spotcooling coil. Using an iterative method, results of the heat exchanger modeling is simultaneously incorporated in the CFD model and an optimal design of spot cooling heat exchanger is developed. This paper was also originally published as part of the Proceedings of the ASME 2005 Pacific Rim Technical Conference and Exhibition on Integration and Packaging of MEMS, NEMS, and Electronic Systems .

The recent miniaturization of electronic devices and compaction of computer systems will soon lead to data centers with power densities of the order of 300 W/ft 2 . At these levels, traditional thermal management techniques are unlikely to suffice. To enable the dynamic smart cooling systems necessary for future data centers, an exergetic approach based on the second law of thermodynamics has recently been proposed. However, no experimental data related to this concept is currently available. This paper discusses the development and subsequent validation of an exergy-based computer model at an instrumented data center in Palo Alto, California. The study finds that when appropriately calibrated, such a computational tool can successfully predict information about local and global thermal performance that cannot be perceived intuitively from traditional design methods. Further development of the concept has promising potential for efficient data center thermal management.