Sei sulla pagina 1di 5

The following article was published in ASHRAE Journal, April 2005.

© Copyright 2005 American Society of Heating, Refrigerating and Air-


Conditioning Engineers, Inc. It is presented for educational purposes only. This article may not be copied and/or distributed electronically or

Data Centers
in paper form without permission of ASHRAE.

Meeting Data Center


Temperature
Requirements
By Roger Schmidt, Ph.D., Member ASHRAE; Madhusudan Iyengar, Ph.D., Member ASHRAE; and Richard Chu

T
he increasing compactness of data to the rack compared to the airflow rate be zones where all equipment within
processing equipment size and the through the rack?” and “Is the remainder that zone dissipates very high heat loads.
expanding needs of the informa- of the air supplied to the rack within ac- This arrangement of equipment may be
tion technology industry has led to more ceptable temperature limits?” required to achieve the performance de-
equipment being packed into the same Data centers, such as the one in Figure sired (cabling between computer racks
amount of space. However, the energy 1, typically are arranged into hot and may be restricted in length to achieve
efficiency of computing and storage cold aisles. This arrangement accom- performance requirements). These high
equipment has not risen at the same rate. modates most rack designs that use performance zones can provide sig-
This has resulted in a significant increase front-to-back cooling. And, it somewhat nificant challenges in maintaining an
in power density and heat dissipation. separates cold air exiting the perforated environment within the manufacturers’
Therefore, the cooling of computer and tiles (for raised floor designs) and over- temperature specifications.
telecommunications equipment rooms is head chilled airflow (for non-raised floor Airflow distribution within a data
becoming a major challenge. designs) from the hot air exhausting center has a major impact on the thermal
The problem in many data centers is from the rear of the racks. environment of the DP equipment located
that the chilled air from the air-condition- The racks are positioned so the fronts within these rooms. A key requirement
ing units cannot satisfy the local airflow of the racks face the cold aisle as in Fig- of manufacturers is that the inlet tem-
requirements of the computer and storage ure 1. Similarly, the rear of the racks face perature and humidity to the electronic
racks. This permits only a portion of each each other, and provide a hot air exhaust
About the Authors
electronic equipment rack to be cooled by region. This layout allows the chilled air
Roger Schmidt, Ph.D., is distinguished engi-
the chilled air, while other portions of the to wash the fronts of the data processing neer and chief thermal architect, and Madhu-
rack pull in air from other regions of the (DP) equipment while the hot air from the sudan Iyengar, Ph.D., is an advisory engineer
data center that is generally at a higher racks exit into the hot aisle as it returns to at the Systems & Technology Group at IBM in
Poughkeepsie, N.Y. Richard Chu is IBM fellow
air temperature. the inlet of the air-conditioning units.
and manager of the Advanced Thermal Labs
The primary questions addressed here With the arrangement of DP equipment at the Systems & Technology Group at IBM in
are: “How much chilled air is supplied in rows within a data center, there may Poughkeepsie, N.Y.

44 ASHRAE Journal ashrae.org April 2005


The problem in many data centers is that the
chilled air from the air-conditioning units cannot
satisfy the local airflow requirements of the com-
puter and storage racks.
equipment be maintained within their is pulled from other regions of the data chilled air entering the data center room
specifications. center to satisfy the flow required by the from the air-conditioning units. All Track inlet
For a Class 1 datacom environment as rack. If the temperature specifications temperature measurements reported here
specified in Reference 1, the allowable of the rack encompass this range, this were made at a height of 1.75 m (5.74 ft)
inlet air temperature is between 15°C and would be acceptable, even though pro- measured from the raised floor and at the
32°C (59°F and 89.6°F) while the rela- longed periods of operation at elevated air inlet of the 2 m (6.56 ft) high racks.
tive humidity is between 20% and 80%. temperatures may be detrimental to the The second temperature metric used
Higher elevations require a de-rating of reliability of the system. here to thermally characterize the data
the maximum dry-bulb temperature of Reference 2 compared four ventilation center is the temperature difference
1°C (1.8°F) for every 300 m (984 ft) schemes for cooling electronic equipment ratio:
above an elevation of 900 m (2952 ft) in a data center: underfloor supply/over-
RdT = dTinlet / dTrack (2)
up to a maximum elevation of 3050 m head return, underfloor supply/horizontal
(10,004 ft). return, overhead supply/underfloor return where dTinlet is defined in Equation
These temperature/humidity require- and overhead supply/horizontal return. 1, and dTrack is the air temperature rise
ments are to be maintained over the entire Experimental results were presented, as through the computer rack equipment.
air inlet area of the rack. Three other class well as an analytical model of the under- When this ratio, RdT , is 0, then the air
environments specified in Reference 1 floor supply/overhead return showing entering the rack is at the supply chilled
generally have a wider range of environ- the airflow and temperature distribution air temperature, and there is no recircula-
mental requirements. at the inlet of the electronic equipment tion (at the point of measurement). When
With airflow through an electronic rack rack. The analytical model showed the this ratio is equal to 1, all of the air (at the
nominally front-to-back, the chilled air rack intake air temperature increasing point of measurement) is at the rack’s hot
supply, whether from a raised floor tile or with height from the raised floor chilled exhaust air temperature due to recircula-
via diffusers from the ceiling, is typically air supply to the top of a rack where in- tion. When this ratio is greater than 1, it
only a fraction of the rack airflow rate. creased recirculation of exhaust air from shows the presence of a self-heating loop
This is due to the limitations of tile or dif- the rear of a rack occurs. References 3 (at the point of measurement), where the
fuser flow rate. The remaining supply of and 4 expand on some of the data shown air is recirculating and increasing in tem-
air to a rack is made up of ambient room in Reference 2. perature more than what would have been
air from mixed hot and cold airstreams from one pass through the rack.
and/or recirculation of hot air exhausting Thermal Performance Metrics
from the rear of a rack then returning While the absolute value of the air Data Center Measurements
directly to its intake. temperature that enters the rack is of Measurements were taken in four
This may be acceptable if the rack inlet ultimate importance, for design purposes distinctly different data centers with
air temperature is within the air tempera- and system characterization, two other the results from only one reported here.
ture specifications of the rack. However, rack inlet air temperature-related metrics Similar results were obtained from all
the airflow paths could result in a wide have been used here. The first metric is four data centers. The results of this
range in inlet air temperatures across the the temperature difference between the investigation are summarized in Figure
face of the rack. inlet air temperature (at a specific loca- 2, which refers to the layout depicted
For a raised floor layout, the inlet air tion in front of the rack) and the supply in Figure 3. All four data centers were
temperature could range from 10°C to chilled air, which is given by: thermally profiled, that is, measurements
15°C (50°F to 59°F) at the bottom of the made of temperature, power and airflow.
dTinlet = Track inlet – Tchilled supply air (1)
rack close to the chilled air supply to as The measurement methodology and the
much as 30°C to 40°C (86°F to 104°F) at where Track inlet is the air inlet temperature results are described in detail for one of
the top of the rack where the warmer air to the rack, and Tchilled supply air is the inlet the data centers in Reference 5.

April 2005 ASHRAE Journal 45


For the results reported here, successively larger areas of the 0.49. One interpretation would be that 49% of the flow for
data center were examined for flow and temperature to see if the rack is supplied by the perforated tile in front of the rack
trends might emerge that describe the mixing of the air and the while the remaining rack flow is drawn from elsewhere in
resulting dTinlet at the air intake of the racks. the room. Although in most cases this would probably be the
Region 1 shown in Figure 3 is part of a larger data center case, it cannot be ensured that all the chilled air flow from that
encompassing 687 m2 (7,391 ft2). Four areas of the data center tile passes to the adjacent rack. However, forming this ratio
were examined. The smallest area for the data center included for larger areas of the data center, a more accurate picture is
one rack (Rack 1 in Figure 3) and the perforated tile area directly obtained. This ratio formed with the measurements from all
in front of the rack. The rack chosen was the highest powered four data centers (only one is shown here for brevity) is quite
rack of the two within the local area. clear; the ratios for all areas of all data centers examined fell
The remaining areas examined were successively larger: Lo- between 0.28 and 0.68.
cal Area 1 encompassing two racks facing one another (depicted And, for the data center measurements reported in Figure 2,
by the area enclosed by the dashed lines in Figure 3), multiple the ratio for the rack, local, region and data center areas ranged
racks facing one another and sharing the cold aisle (Region 1 from 0.48 to 0.63. This ratio of less than one but greater than 0
in Figure 3) and finally the entire data center (not shown). Each indicates that some air is supplied to the rack(s) from elsewhere in
grouping of data in Figure 2 displays the results for each suc- the room and not from the perforated tiles in front of the racks.
cessively larger area: rack, local, region and data center. The fifth grouping in Figure 2 depicts the ratio of the mea-
The first grouping shows sured cold aisle plus hot aisle
power for the rack at 7.5 kW Racks Air-Conditioning Unit flow rates and the rack flow
(25,600 Btu/h), local area rate. The cable openings are a
power of 14.2 kW (48,460 source of chilled air from the
Btu/h), regional power of 106 hot aisle and provide cooling
kW (361,770 Btu/h) and the to the hot air exhausting from
data center power of 1089 the rear of the racks. This ratio
kW (3,716,720 Btu/h). (The varied from 0.73 to 1.0.
last bar in each grouping is Even when including the
from the numerical modeling, cable opening flow, the ratio
which is described later.) was less than 1 for all cases
The next grouping shows except for the region showing
the heat flux for each area that in most cases the rack inlet
with the highest occurring air is from other regions of the
for the rack. The correspond- data center. In further support
ing dTinlet’s are shown in the of warm air recirculation the
Perforated Floor Tiles
next grouping (averages are RdT, shown in the last group-
shown when displaying val- Figure 1: A data center arranged in hot and cold aisles. ing, fell between 0.5 and 0.75
ues for more than one rack). indicating that a portion of the
With a dT inlet of 8.1°C (14.6°F) for Rack 1 as in Figure 2 and air for the rack was supplied from elsewhere in the room.
a supply chilled air temperature of 12.9°C (55.2°F), the rack Reference 5 states that based on the reported measurements in
inlet air temperature (at a height of 1.75 m [5.74 ft]) of 21°C that study if the perforated tiles associated with a rack exhausted
(69.8°F) is well below the rack maximum temperature limit of one-quarter to one-half of the rack flow, then the system inlet air
32°C (89.6°F). temperature could be maintained (given the chilled air exhaust
Although not shown, all rack inlet temperatures in the data temperature was approximately 13°C to 15°C [55.4°F to 59°F]).
center were within the maximum rack inlet temperature speci- For this study, and in all cases of the data center (Rack 1, Local
fication. And, every rack inlet air temperature was greater than Area 1, Region 1 and Data Center 1) , the ratio of chilled airflow
the supply chilled air temperature showing that warm air from exhausting from the perforated tiles to the rack flow rate varied
other portions of the room is being drawn into the rack intakes. between 0.48 and 0.63 (the range for all four data centers was
Another way to confirm that warm air is entering the racks is 0.28 to 0.68) , and the air inlet temperature for the racks were
to examine the airflow rates. within specifications. This study shows that the data is similar
The fourth grouping of Figure 2 displays the perforated tile to that reported in Reference 5.
flow rate immediately in front of the rack(s) compared to the
rack flow rate(s). For example, the flow rate for Rack 1 is 0.5 Modeling Experiments
m3/s (1,050 cfm) while the perforated tile flow rate immediately One of the objectives of the modeling was to construct a
in front of the rack is 0.24 m3/s (516 cfm) forming a ratio of simplified model of the region (described previously) used

46 ASHRAE Journal ashrae.org April 2005


2.25

Rack
2 Local
Region
1.75 Data Center
Numerical Model
×10

1.5

×10
Measured Value

1.25
×103
×103
×102

1
×10

0.75
×10

0.5

0.25

0
Power Heat Flux dT Inlet, C Perf. Flow/ Perf. + Cable Flow/ RdT
kW kW/m2 Rack Flow Rack Flow

Figure 2: Results for measurement and modeling of a data center shown below in Figure 3.

for measurements but still capture most of the thermo-fluid The rack was modeled with blowers that force a specific
and geometric parameters and then compare the results. A airflow through these three portions of the rack with the same
representative model of a section of raised floor data center flow characteristics as the rack. The temperature of the chilled
system, depicted in Figure 4, was constructed using a com- air entering the room through the perforated tiles was fixed
mercial software tool. The model was constructed to closely at 15°C (59°F). The rack flow rate, perforated tile flow rates,
resemble the layout in Figure 3, for and cable flow rates matched those
which measurements were taken. Region 1 measured in Region 1 of Figure 3.
The model consists of four pri- The room dimensions were four
mary components; namely, the heat times that of the quarter symmetry
generating datacom equipment model shown in Figure 4, i.e., 6.05
racks, perforated tiles that supply m by 2.8 m (19.8 ft by 9.18 ft), with
chilled air, cable openings that a room height of 2.75 m (9.02 ft).
cause underfloor air to leak into The rack dimensions were 0.61 m
the hot aisle, and computer room (2 ft) wide by 1.22 m (4 ft) deep by
air-conditioning units (CRAC) 2 m (6.56 ft) tall.
that draw in hot return air from the The resulting heat flux of this
racks and exhaust chilled air into configuration is 1.6 kW/m2 (507.6
the raised floor plenum. The model Rack 1 Btu/h/ft2) (Figure 2). Thus, the heat
comprised of a quarter symmetry flux and the flow conditions are the
Local Area 1
of 16 rack sections of a data center same for the numerical model cell
arranged in a cold aisle-hot aisle Figure 3: Rack layout in a data center. as for those measured in the data
fashion. The racks were assumed to center and reported previously. One
dissipate the same heat load as the machines for which the perforated tile with dimensions of 0.61 m by 0.61 m (2 ft by
experimental results are reported. The geometry of some of 2 ft) was placed in front of each rack. The cable openings at
the racks from the field measurements was replicated in the the back of the rack were modeled as 17.8 cm (7 in.) deep
computer model. and the width of the rack.
The racks were modeled to dissipate 6.7 kW (22,870 Btu/h) The thermo-fluid interactions were numerically solved using
each (average of the racks measured in the data center), with the k-e turbulence model, which is the preferred choice for such
the heat load distributed in three portions of the rack (same as data center fluid flow modeling. The cell count for the models
those shown in the measurements of Figure 3): input/output was approximately 50,000. Temperature data was recorded for
portion at the bottom of the rack, central electronic complex the numerical simulations at nine vertical locations in front on
that includes memory and processors in the center of the rack, the rack, namely at heights of 200 to 1600 mm (7.87 to 63 in.)
and ac/dc power supplies located at the top of the rack. in increments of 200 mm (7.87 in.), and the highest point at

April 2005 ASHRAE Journal 47


1750 mm (68.9 in.) (same as the location for measurements). 2. Although the room flow rate is adequate to cool the overall
The criteria for satisfactory convergence took into consideration heat load in the data center and the local flow rate adjacent to
the trends and absolute values, respectively, for the temperature,the racks does not appear adequate, the convection currents
continuity residuals and the monitor point temperatures. that occur at the room level are adequate to bring the local
air temperatures for the high powered racks within the rack
Measurements and Modeling Comparisons air temperature specifications. Flow patterns observed from
Although several comparisons can be made between the simple modeling of a data center show the upper regions of a
measurements reported from Data Center 1 and the numeri- rack being satisfied with flows from other portions of the room.
cal modeling, the desire is to find a metric that can easily be Mixing of the cold and hot airstreams can occur in the vicinity
used in data center characterization and, therefore, the focus of the rack or in other portions of the data center. However, the
was on the dimensionless metric RdT defined in Equation 2. mixing that can occur allows the air temperature to be within
As seen in Figure 2, the temperature data from the numerical the air temperature specifications of the racks.
results were higher than those measured, for the data center 3. A novel methodology was demonstrated by which a large
level case, when the airflow complex data center can be
ratios and the heat flux were Chilled Air-Conditioning Unit separated into global, region,
modeled to be same as the local, and rack level regions
experimental case. This dem- for analysis and modeling.
onstrates the need for signifi- 4. A temperature difference
cantly more work in this area formed by the temperature
if these tools are to be used rise of the rack inlet air and
more generally to model the the supply chilled air tempera-
thermal and airflow behavior ture, and a temperature ratio
of data centers. formed by the temperature
Several weaknesses need to rise of the rack inlet air and the
Air Temp., °C
be further explored based on air temperature rise through
> 48
the following observations: the rack, was used to col-
39.75
• The difficulty in making lapse the data for comparison
Cable Openings 31.5
detailed and exhaustive purposes.
Racks Perforated 23.25
measurements at the data 5. A comparison between
Tiles Symmetry Plane <15
center level due to the size measurement data and model
of the facility; Figure 4: Computer-based simulation of a data center. results showed the need for
• The complex nature of the more work in this area.
rack design, which does not always lend itself to simplifica-
tion in numerical modeling; and Acknowledgments
• The large number of variables and boundary conditions that We want to thank the people who aided in the collection of the
need to be matched to make an accurate comparison. data for this study: Bob Wasilewski and Tom Juliano of DLB As-
Future work will be devoted to addressing these issues to sociates Consulting Engineers, Ocean, N.J., and Saurabh Shrivat-
close the gap between modeling and measurements. sava of the State University of New York at Binghamton.

Summary/Conclusions References
1. Measurements in a data center showed with significantly 1. Thermal Guidelines for Data Processing Equipment. 2004.
ASHRAE Special Publication.
less flow from the perforated tiles in front of racks, compared 2. Nakao, M., H. Hayama, M. Nishioka. 1991. “Which cooling
to the rack flow, the inlet air temperatures into a rack could air supply system is better for a high heat density room: underfloor
be met. Specifically, measurements showed that for the rack, or overhead?” Thirteenth International Telecommunications Energy
local area, region and data center the ratio of perforated tile Conference, (INTELEC ’91), Paper 12-4, pp. 393 – 400.
3. Hayama, H. M. Nakao. 1989. “Air flow systems for telecommunica-
flow to rack flow ranging from 0.48 to 0.63 were acceptable tions equipment rooms.” The Eleventh International Telecommunications
to meet inlet air temperature. (The data from four data cen- Energy Conference (INTELEC ’89), Vol. 1, Paper 8.3, pp. 1–7.
ters ranged from 0.28 to 0.68). This range was similar to that 4. Hayama, H., M. Nakao, and M. Sanabe. 1990. “Airflow distribu-
reported in Reference 5, stating that if the perforated tiles tion in telecommunications equipment rooms.” Twelfth International
Telecommunications Energy Conference (INTELEC ’90), Paper 11-4,
associated with a rack exhausted one-quarter to one-half the pp. 206 – 212.
flow of the rack, then the system inlet air temperature could 5. Schmidt, R. 2004. “Thermal profile of a high density data
be maintained (given the chilled air exhaust temperature was center—methodology to thermally characterize a data center.” 2004
below approximately 15°C [59°F]). ASHRAE Annual Meeting.

48 ASHRAE Journal ashrae.org April 2005

Potrebbero piacerti anche