Content uploaded by Zhonghua Gou
Author content
All content in this area was uploaded by Zhonghua Gou on Sep 01, 2024
Content may be subject to copyright.
Journal Pre-proof
Towards Energy-Efficient Data Centers: A Comprehensive Review of
Passive and Active Cooling Strategies
Senhong Cai , Zhonghua Gou
PII: S2666-1233(24)00091-6
DOI: https://doi.org/10.1016/j.enbenv.2024.08.009
Reference: ENBENV 354
To appear in: Energy and Built Environment
Received date: 9 May 2024
Revised date: 25 August 2024
Accepted date: 30 August 2024
Please cite this article as: Senhong Cai , Zhonghua Gou , Towards Energy-Efficient Data Centers:
A Comprehensive Review of Passive and Active Cooling Strategies, Energy and Built Environment
(2024), doi: https://doi.org/10.1016/j.enbenv.2024.08.009
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.
Copyright ©2024 Southwest Jiatong University. Publishing services by Elsevier B.V. on behalf of
KeAi Communication Co. Ltd.
This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
1
Highlights
• Passive and active strategies for cooling data centers were summarized;
• Different cooling design strategies in best practice examples was explored;
• Four directions for future work on cooling energy reduction in DCs were indicated;
• Critical emphasis is placed on the development of passive strategies;
• Different active cooling strategies were compared through KPIs.
2
Towards Energy-Efficient Data Centers: A Comprehensive Review of Passive and Active
Cooling Strategies
Senhong Caia, Zhonghua Goub *
a,b School of Urban Design, Wuhan University, Wuhan, China; * correspondence: zh.gou@whu.edu.cn
Abstract
With the rapid growth of cloud computing, the number of data centers (DCs) continuously increases, leading to a
high-energy consumption dilemma. Cooling, apart from IT equipment, represents the largest energy consumption
in DCs. Passive design (PD) and active design (AD) are two important approaches in architectural design to reduce
energy consumption. However, for DC cooling, few studies have summarized AD, and there are almost no studies
on PD. Based on existing international research (2005-2024), this paper summarizes the current state of cooling
strategies for DCs. PD encompasses floors, ceilings, and layout and zoning of racks. Additionally, other passive
strategies not yet studied in DCs are critically examined. AD includes air, liquid, free, and two-phase cooling. This
paper systematically compares the performance of different AD technologies on various KPIs, including energy,
economic, and environmental indicators. This paper also explores the application of different cooling design
strategies through best-practice examples and presents advanced algorithms for energy management in operational
DCs. This study reveals that free cooling is widely employed, with Artificial Neural Networks emerging as the
most popular algorithm for managing cooling energy. Finally, this paper suggests four future directions for
reducing cooling energy in DCs, with a focus on the development of passive strategies. This paper provides an
overview and guide to DC energy-consumption issues, emphasizes the importance of implementing passive and
active design strategies to reduce DC cooling energy consumption, and provides directions and references for
future energy-efficient DC designs.
Keywords
Data center; Passive design; Active design; Cooling load; Energy efficiency
3
Abbreviation
AD
Active design
AI
Artificial intelligence
ANN
Artificial Neural Network
ASHRAE
American Society of Heating, Refrigerating and Air-Conditioning Engineers
CACS
Hot aisle containment system
CCHP
Combined cooling, heating and power
CFD
Computational fluid dynamics
CHP
Combined heat and power
CRAC
Computer room air conditioner
CRAH
Computer room air handler
CUE
Carbon usage effectiveness
ESR
Energy-saving rate
GA
Genetic Algorithm
GPR
Gaussian Process Regression
HACS
Cold aisle containment system
HP
Heat pipe
HVAC
Heating, ventilation, and air conditioning
IT
Internet technology
KPI
Key performance indicator
LSTM
Long Short-Term Memory
MAE
Mean absolute error
MSE
Mean square error
PD
Passive design
POD
Proper Orthogonal Decomposition
PUE
Power usage effectiveness
RMSE
Root mean square error
RSC
Radiative sky cooling
SVR
Support Vector Regression
4
WHR
Waste heat recovery
WSR
Water-saving rate
WUE
Water usage effectiveness
1. Introduction
With the widespread application of artificial intelligence (AI), cloud computing, big data, and other
technologies, the global demand for computing power has shown explosive growth [1]. A computing power crisis
is imminent, and the proper construction of data centers (DCs) is becoming an essential part of the solution to this
crisis. Unlike traditional buildings that mainly serve people, DCs, as an emerging building type in recent years,
provide an operating environment for electronic information devices [2, 3]. It is well known that most companies
rely heavily on DCs for key operational aspects, requiring high-density information technology (IT) equipment to
operate continuously (24 hours a day and 365 days a year), making DCs ultra-high energy-consuming buildings
[4, 5]. By 2030, DCs are expected to use around 3-13% of global electricity in 2030 compared to 1% in 2010 [6].
As shown in Fig. 1, the energy of a DC is mainly consumed by IT equipment, cooling systems, electrical systems,
and other equipment [7]. The cooling system is the second largest energy consumer, with about 40% of the energy
consumption, to maintain the temperature of the DCs in a certain range [8, 9].
Fig. 1. Energy consumption in DCs.
Both passive design (PD) and active design (AD) strategies are commonly used in buildings to reduce energy
consumption [10]. PD focuses more on reducing building energy demand by optimizing the building envelope,
shape, and layout during the design process; AD involves the use of more energy-efficient heating, ventilation, and
air conditioning (HVAC), hot water production, lighting, and other building service applications [11]. However, in
practice, passive and active cooling strategies are used simultaneously in DCs to achieve low energy consumption,
5
and few cooling strategies operate independently. To facilitate the implementation of the review, we discuss these
two strategies separately in this paper. AD relies on energy-consuming building equipment for cooling, which is
itself an energy-consuming process, whereas PD aims to improve the performance of DC without energy
consumption. Whether or not the building consumes energy is what distinguishes them, and this has reached a
consensus in the building sector. This categorization provides a clear framework for understanding and comparing
different cooling strategies and helps researchers and engineers to more systematically evaluate and select the most
appropriate cooling solution. Passive cooling strategies are often more environmentally friendly because they
reduce the need for mechanical power. By differentiating and combining passive and active cooling strategies, the
overall energy use can be better optimized, unnecessary energy consumption can be reduced, and strategies with
lower environmental impact can be more effectively evaluated and selected.
To understand the current research situation and focus of PD and AD strategies in DC cooling, and to
summarize the specific classification and content of both strategies, this paper systematically reviews the existing
review articles (see Appendix A) on building PD and AD. Although research on PD and AD in buildings has been
widely conducted, only a few are specific to a particular building type, such as residential buildings and office
buildings. Regarding DCs, most existing review articles summarize a specific cooling method, such as phase-
change cooling [12] and free cooling [13, 14], which are both AD. However, for DC cooling energy, few studies
summarize the AD methods [15], and there are almost no comprehensive reviews of PD.
Key performance indicators (KPIs) are commonly used to evaluate the performance of DCs to ensure energy-
efficient operation while reducing the cost and environmental impact. Although previous studies have summarized
these KPIs, they often focus solely on energy or lack comprehensive cooling strategies [9, 16-20]. Some studies
have proposed new indicators, such as the Exergy Loss Index [21], Cumulative Energy Efficiency and Cumulated
Performance Efficiency [22], and coefficient of PUE [23]. The pivotal global energy indicator for DCs is power
usage effectiveness (PUE) [24, 25], which measures the total energy consumption relative to IT equipment energy
use; a lower PUE indicates greater efficiency. The cooling systems, significant energy consumers, are crucial for
PUE reduction. The energy-saving rate (ESR) is another important parameter for evaluating the energy
performance of a DC, and it is defined as the proportion of the electricity saved by the optimized cooling system
to the total electricity consumption of the original cooling system[26]. Environmental aspects include water and
carbon [27]. Water usage effectiveness (WUE) and carbon usage effectiveness (CUE) are defined as the ratio of
annual water usage and total greenhouse gas emissions of a DC to the energy consumption of IT equipment [9,
28]. In addition, water savings can be measured by the amount of water saved and the water-saving rate (WSR);
6
carbon reduction is a commonly used value. Economic efficiency, which is vital for operators, is evaluated through
the cost and payback periods [29]. The payback period refers to the time taken for an investment project to recover
its initial investment cost and is a key indicator for evaluating the potential of a system’s application [30].
Overall, there is a lack of a comprehensive review of cooling technologies for DCs and of passive cooling
technologies. In addition, existing research has not summarized or compared these technologies in terms of KPIs,
which not only include energy but also need to consider environmental and economic aspects. To fill these gaps,
this study investigates PD and AD strategies for cooling the particular building type, and the analysis framework
is shown in Fig. 2. Combining the classification methods of existing articles and the definitions of PD and AD, we
classified design strategies that do not require energy consumption as PD and those that require energy
consumption as AD. PD includes floors and ceilings as well as the layout and zoning of racks. We also critically
highlight other passive strategies that have not been studied in DCs. AD includes several design types of DC
cooling systems: air cooling, liquid cooling, free cooling, and two-phase cooling. We also summarize the energy,
environmental, and economic KPIs of these strategies.
This paper provides a comprehensive overview of current cooling strategies for DCs, including a detailed
review of the PD and AD classifications and applications for DC cooling, a comparison of strategies through KPIs,
an exploration of strategies through best-practice case studies, an examination of advanced approaches to energy
management during the operational phase, and a forward-looking look at the trends in the field. This study is the
first to provide a comprehensive understanding of PD and AD approaches to DC cooling to cope with the exploding
arithmetic demand and meet the urgent need for future energy efficiency in DC cooling. This study makes
important contributions to the field of DC energy efficiency in several ways. First, it fills a gap in the existing
research on DC passive cooling strategies and provides an important reference for understanding the advantages
and disadvantages of various strategies. Second, by categorizing and comparing different types of passive and
active cooling strategies, a comprehensive and systematic overview of current DC cooling strategies is provided
to help users understand the latest progress in the field. In addition, the application of different cooling design
strategies is demonstrated by providing actual DC project examples, providing a valuable reference for real
projects. Finally, this paper provides a direction for future research and development in DC energy efficiency
improvement.
7
Fig. 2. Framework of review.
2. Methodology
The research method of this study is to conduct a comprehensive review of the existing literature. It can be
specifically divided into three parts. The first step is data collection. Data was collected through the literature
search. Considering the time from the birth to the development of DCs, the literature searched consisted of peer-
reviewed publications published in the last 20 years (2005 to 2024) on which the innovative and scientific
characteristics of the study could be met. Table 1 lists the relevant subject terms used for the search, thereby making
the literature review more relevant. Item 1 limits the scope of the study to DCs and the subcomponents within
them; item 2 focuses on the energy sector; the words in item 3 aim to locate the different ways and places of
cooling; and item 4 fulfills the purpose of this study's review of KPIs for different strategies. We combined the
search terms of the first item with those of the other items, respectively, and conducted extensive searches in
reputable academic databases, including Web of Science, Scopus, and Google Scholar. In addition, closely related
references were examined using the snowball method, which is a commonly used literature search in similar review
articles method [31]. We expanded the literature base by examining the references in the review literature, which
allowed us to identify newer or lesser-known literature that may have been missed by traditional search methods
8
[32]. In addition to journal articles, projects, reports, and industry standards covered in the reviewed literature
were identified to meet the requirements of this study for actual projects. The literature reviewed was written in
English.
The second step is to process the collected literature. In our initial search, we identified many duplicates and
irrelevant profiles; thus, we conducted further screening. We first performed a quick screening of the titles and
abstracts to determine their relevance to our research topic. Then, we conducted a second round of in-depth analysis
to critically assess the value of the remaining articles by closely reading their chapter headings or full-text content.
Through this screening step, we obtained the literature targeted to fulfill the purpose of this study.
Finally, the screened literature was analyzed and summarized in detail. Based on the framework shown in
Figure 1, we appropriately and rigorously categorized and organized the articles to correspond to different cooling
strategies for DCs. Given that the existing studies did not cover the KPIs of PDs, in the analysis of AD, we also
summarized and analyzed the KPIs corresponding to the different strategies in the studies according to their actual
situation. In addition to the design of cooling design strategies, the best practice examples and energy management
were also analyzed in a corresponding review. Combining the above analyses, we proposed future research
directions.
Considering the limited time and capacity, the review methodology has some limitations. We searched the
literature from 2005 to 2024 and focused primarily on English language literature, which may have missed some
earlier studies or important developments. In addition, the screening process relied on subjective professional
judgment, which may have led to bias. Although this review article is not exhaustive, this research has been
relatively innovative and reliable and can contribute to the field of DC cooling.
Table 1. The terminology used in searching the literature.
No.
Searching query
1
Topic = (“data center” OR “cloud data center” OR “IT facility” OR “network operations center” OR “computer room” OR “server room”)
2
Topic = (“energy” OR “power” OR “electricity” OR “energy efficiency” OR “energy management” OR “energy saving”)
3
Topic = (“passive cooling” OR “active cooling” OR “air cooling” OR “liquid cooling” OR “free cooling” OR “two-phase cooling” OR “air
conditioning” OR “heat reduction” OR “cooling strategy”)
4
Topic = (“key performance indicator” OR “KPI” OR “indicator” OR “metric” OR “power usage effectiveness” OR “energy-saving rate”
OR “water-saving rate” OR “water usage effectiveness” OR “carbon usage effectiveness” OR “cost” OR “payback period”)
3. Passive cooling design strategies
Passive cooling design strategies included floors and ceilings, layout and zoning of racks, and other passive
9
strategies not studied. It is worth mentioning that there are very few comparable results from this part of the
quantitative evaluation in the existing studies; thus, we did not compare the passive strategies’ KPIs.
3.1. Floors and ceilings
The floors in computer rooms are categorized into two main types: hard floors and raised floors; the latter are
more convenient for arranging wiring and cold air for the racks of DCs [16]. The static pressure in a raised floor
is the power source that promotes airflow into the computer room, which is mainly affected by the floor height.
The pressure difference along the flow direction is maintained as the height increases [33], but the inhomogeneity
of the pressure distribution reduces the airflow rate [34]. In addition, the arrangement of obstacles should be
considered, mainly including horizontal/vertical chilled water pipes and hot/cold aisle trays, which are required to
be no more than one-third of the height of the raised floor [35]. Based on thermal performance, Beitelmal [36]
determined that a height of 0.76-0.91 m is optimal. Nada and Said [37] recommended a floor height of 0.6 m.
Zhang et al. [38] recommended a range of raised floor heights: open aisle is 1.0-1.2 m, cold aisle containment is
0.6-0.8 m, and hot aisle containment is 0.4-0.6 m. Underfloor geometry also affects thermal performance. For
example, Lu and Zhang [39] found that cross-sectioned underfloor spaces can reduce inlet and outlet air
temperatures by 1.7-2.9 K. The underfloor space can also be used to reduce inlet and outlet air temperatures.
Raised floors need to be set up with perforated tiles to allow cooling airflow to enter the computer room from
the lower part, and the perforated tiles’ open position, open area, tile geometry and open direction affect the airflow
characteristics in the computer room and thus the cooling energy [33]. The open position should match the air
outlet of the air conditioning: if it is too close to the air conditioning, the cold air cannot come out; if it is too far
away from the air conditioning, the cold air cannot reach the perforations [40]. With the rack layout and raised
height determined, the flow uniformity is related to the open area of the tiles. Numerous researchers have
determined the reasonableness of a 25% open area because this area has better flow uniformity and temperature
distribution uniformity [41-43]. Perforated tiles have different geometries, including shape and size. Tiles with the
asymmetric geometry are also asymmetric in terms of the distribution of velocities [44]; the flow field through
circular and square perforated tiles is similar [45]. Arghode and Joshi [46] found that decreasing the diameter of
the aperture affected the airflow and that reducing the width of the tiles improved the air transport into the racks.
In a study of the open direction of the tiles, Khalili et al. [47] found that directional grids significantly improved
airflow delivery. For specific angles, Ni et al. [48] found that the air distribution is optimal at 60° by comparing
different open angles (30°, 45°, 60°, and 90°). Overall, Sorell [49] described how to select appropriate perforated
10
tiles for DCs of various densities and provided guidelines on how to manage the number and position of tiles.
The ceiling height is another important factor. Sorell et al. [50] found through computational fluid dynamics
(CFD) modeling that ceiling height has a significant impact on the airflow as it relates to the hot air circulation in
a computer room. An increase in the ceiling height affects the inlet and exhaust temperatures of the racks, whereas
a lower ceiling height leads to heat buildup [34]; therefore, a balance must be found. Nagarathinam et al. [51] and
Bhopte et al. [52] suggested a ceiling height of 2.52 m. It is worth noting that increasing the height of the raised
floor also resulted in a decrease in the ceiling height. The development and adaptation of the latest standards is
evolving dynamically based on the latest research findings, such as the ANSI/BICSI 002-2019 Data Center Design
Standard.
3.2. Layout and zoning of racks
The rack layout and zoning of the computer rooms, as a special PD method in DCs, should be given high
priority in the design. In the past, airflow factors were not considered in rack planning, and they were generally
laid out in the same direction, as shown in Fig. 3(a). This resulted in the mixing of hot air from the front rack and
cold air from the rear rack, which increased the temperature of the incoming air from the rear rack and reduced the
cooling efficiency [16]. Nowadays, face-to-face or back-to-back arrangements are commonly used, as shown in
Fig. 3(b). This approach can be used in a passive manner to form hot and cold aisles, where the incoming cold
airflow can pass through the cold aisle into the racks. In the hot aisle, the rear side of the racks is aligned along the
aisle from which the hot exhaust air exits each server and returns to the air-conditioning system from the rear of
the racks, achieving circulation and effectively inhibiting the airflow from mixing at different temperatures [53].
11
Fig. 3. Racks arranged in the uniform orientation and the face-to-face layout [16].
The zoning between racks should meet the requirements of the overall layout of the computer room and the
hot and cold partitioning, and the electricity consumption of the racks should be compatible with the cooling
capacity of the corresponding area; while the local heat island phenomenon should be avoided in the server
arrangement inside the racks [54]. According to the form of air distribution, the server arrangement inside the racks
can be divided into four types: uniform distribution, discrete distribution, segmented distribution, and clustered
distribution [55]. A uniform distribution has equal distances between the servers, maximum cold-air utilization,
and balanced space utilization and heat dissipation, which facilitates network cabling and management [55].
Discrete distribution arranges the servers decentralized according to their functions, and segmented distribution
arranges the servers in segments according to specific requirements. Both approaches are more likely to absorb
cold air and are particularly suitable for high-power servers, which require stronger airflow circulation for cooling
[56]. The clustered distribution places a group of servers in neighboring positions in a rack to form a cluster, where
12
the servers can back up each other, providing high availability and fault tolerance. In this way, Ghosh et al. [57]
found that servers are more energy efficient when they are set up higher in the rack. The designer must select an
appropriate server arrangement according to the specific situation. Jin and Bai [58] found that for DCs with low
load rates, servers evenly distributed in all racks have good air intake temperatures, and suggested that the servers
should be evenly distributed from the middle to the upper part of the rack; Jin et al. [34] also suggested that for
low-density racks, clustering of the upper racks can be used as an alternative to uniform distribution.
3.3. Other passive strategies not studied
The two PD strategies described above have been carried out primarily for the interior design of DCs, however,
PD involves several other aspects, including site planning, building envelope, and radiative sky cooling (RSC).
Their importance in reducing cooling loads in buildings has been demonstrated, but there is almost no ongoing
research on their use in DCs. Here, we summarize the application paths of these contents in DCs.
Site planning includes building location, building layout, and building shape. The location has an impact on
its energy efficiency and cooling requirement [59, 60], and to an increasing degree as such buildings become more
energy efficient [61]. Priority should be given to locations with natural cooling sources to reduce the dependence
on DC cooling systems. Common natural cooling sources include but are not limited to the following: seawater
[62], cold air [63], and caves [64]. High-security requirements prevent DCs from being co-located in a single
building with the facilities they serve or the ancillary facilities that serve them; thus, these buildings are commonly
dispersed to form a park. The building layout changes the energy consumption of a building by altering the
microclimate of the site [65, 66]. Specifically, rational orientation can minimize the reliance on artificial lighting
and cooling systems; the natural airflow formed between buildings allows for convection and heat dissipation, thus
reducing the baseline temperature by utilizing the building’s natural ventilation. Buildings with different shapes
can contribute between 1% and 5% to ESR [67]. In general, designing a square or rectangular building shape is
easier to cool and consumes less energy than a convex building [68-70]. The shape of a DC also requires
consideration of the aspect ratio that maximizes the use of the DC’s cooling capacity. An aspect ratio of
approximately 4:3 seems to work best [71].
The thermal performance of the building envelope largely impacts the cooling load [11]. The building
envelope component may include walls, roofs, and windows in the structure, and shading measures on the facade.
Improving the insulation of walls is also an effective way to reduce cooling energy [72], which can be achieved
by optimizing the wall structure and materials. Generally, Trombe walls can reduce the energy consumption of
13
buildings by up to 30% through a special construction method [73]. In DCs, reducing the external heat gain
generated by roofs can be achieved by using surface materials with high solar reflectance and thermal emittance
or other insulating materials and green roofs. Cool roofs that absorb less heat reduce the cooling energy of a
building by selecting brighter (usually white) roofs to replace darker ones. Green roofs are an effective energy load
reduction strategy to generate evaporative cooling [74], and they also have an impact on air quality and occupant
health [75]. DCs typically avoid windows in the computer room area because of the potential for them to cause
physical damage, as well as light interference, etc. However, they are needed in other non-computer room areas
for natural light. Advanced window technologies typically use low-e coatings, vacuum or inert gas filling, and
aerogels to reduce the U-value and solar heat gain coefficient [76-78]. In addition to optimizing the window
material and construction method, the focus should be on improving the cooling effect by adding adhesive strips
to poorly sealed parts of the windows. Shading is a facade facility in the envelope that protects buildings from
solar radiation, making shading systems very suitable for cooling-oriented DCs. For general buildings, adjustable
shading control by daylight or climate controllers is an attractive way [79]. However, in DCs, winter light and heat
harvesting are not required in the computer rooms, so it is appropriate to use fixed shading to block heat for cooling.
Rooms for other functions can be carefully evaluated and weighed against their pros and cons, and decisions can
be made on a case-by-case basis to balance stability and flexibility.
In recent years, RSC has attracted much attention as a rapidly evolving cooling technology that radiates waste
heat in a non-consumptive and completely passive manner through an atmospheric window of 8-13μm into the
universe at 3K [80]. RSC has great potential for cooling buildings, vehicles, solar cells, and even power plants
[81]. Research on RSC in the last decades has been limited to nighttime use. Nowadays, the use of RSC during the
daytime to reach temperatures below the environment has been experimentally proven to be feasible. RSC has a
promising application in DCs where cooling is required throughout the year [82, 83]. Currently, specialized
research on this passive cooling strategy in DCs is still in its infancy. To the best of our knowledge, only Aili et al.
[84] have simulated three configurations for integrating all-day passive radiant cooling into a DC in the tropical
climate of Singapore, with an average annual ESR of approximately 20.0% and an average annual WSR of
approximately 84.0%. The potential applications of RSC in DCs remain to be studied.
14
4. Active cooling design strategies
4.1. Air cooling
Air cooling technology is the backbone and conventional solution for DC cooling systems, and it is widely
used in large DCs. There are usually four types of airflow organization in DC computer rooms: room-, row-, rack-,
and chip-based [54]. The focus is usually on the first three types because implementing chip-based air cooling
optimization solutions inside racks significantly increases the implementation and maintenance costs of some DCs
[85]. Fig. 4 illustrates the three types of air-cooling approaches. In air cooling, cooling equipment can be called
CRAH/CRAC (computer room air handler/computer room air conditioner).
15
Fig. 4. Floor plans illustrating the basic concepts of room-, row-, and rack-based cooling [86].
16
Table 2 summarizes the advantages and disadvantages of the three approaches, where redundancy refers to a
system design in which key components are replicated so that the DCs can continue to operate even if a component
fails. Room-based cooling can quickly change the cooling distribution pattern by reconfiguring the floor tiles,
which is cost-effective and simplest and better suited for low-density DCs; row-based cooling has advantages in
flexibility, speed of deployment, and density, and is less costly; and rack-based cooling is the most flexible, the
fastest to deploy, and can support the highest power densities but at an additional cost.
Table 2. Comparison of room-, row-, and rack-based air cooling features and their advantages and disadvantages, adapted from [54].
Category
Room-based air cooling
Raw-based air cooling
Rack-based air cooling
Flexibility
Advantages
Rapid changes can be performed in
cooling distribution patterns at
power densities less than 3 kW.
It is convenient to plan; the
cooling capacity can be shared.
It is convenient to plan and is
independent of existing cooling
systems.
Disadvantages
Reduced efficiency when airflow
containment is not applied
Hot and cold aisle layout is
required.
The cooling capacity cannot be shared
with other racks.
System
availability
Advantages
All racks can share redundancy.
Racks in a zone can share
redundancy; cooling near heat
sources eliminates vertical
temperature gradients.
Cooling near a heat source eliminates
hot spots and vertical temperature
gradients; it minimizes human error.
Disadvantages
Airflow containment is required.
Redundancy is required in each
rack area.
Redundancy is required in each rack.