Article

Mechanical Engineering July 2011 Vol.56 No.20: 21852190

doi: 10.1007/s11434-011-4500-5

SPECIAL TOPICS:

Investigation into the energy consumption of a data center with a

thermosyphon heat exchanger
ZHOU Feng, TIAN Xin & MA GuoYuan*
College of Environmental and Energy Engineering, Beijing University of Technology, Beijing 100124, China

Received October 18, 2010; accepted February 17, 2011

A data center test model was used to analyze the energy dissipation characteristics and energy consumption of a data center. The
results indicate that adequate heat dissipation from a data center cannot be achieved only from heat dissipation through the build-
ing envelope during Beijing winter conditions. This is because heat dissipation through the building envelope covers about 19.5%
of the total data center heat load. The average energy consumption for an air conditioner is 4 to 5 kW over a 24-h period. The
temperature difference between the indoor and outdoor air for the data center with a thermosyphon heat exchanger is less than
20°C. The energy consumption of the thermosyphon heat exchanger is only 41% of that of an air conditioner. The annual energy
consumption can be reduced by 35.4% with a thermosyphon system. In addition, the effect of the outdoor temperature on the en-
ergy consumption of an air conditioner is greater than the indoor room temperature. The energy consumption of an air conditioner
system increases by 5% to 6% for every 1°C rise in the outdoor temperature.

data center, energy consumption, thermosyphon heat exchanger, ambient energy

Citation: Zhou F, Tian X, Ma G Y. Investigation into the energy consumption of a data center with a thermosyphon heat exchanger. Chinese Sci Bull, 2011, 56:
21852190, doi: 10.1007/s11434-011-4500-5

The energy consumption of data centers has increased rap- National Laboratory, the energy consumption of a typical
idly in recent years, and the situation is getting worse. The data center is made up of 44% from IT devices (including
US Environmental Protection Agency (EPA) estimates that servers, switches and routers), 38% from cooling devices
the total energy cost for all data centers in the US in 2006 (including air conditioners and fans), 15% from power de-
was 4.5 billion dollars, and the cost will reach 7.4 billion vices (including Uninterrupted Power Supplies (UPS) and
dollars by 2011 [1]. In addition, the power consumption of Server Power Management (SPM) equipment), and 3%
servers is increasing rapidly. In 1998 the thermal load for a from lights. The energy consumption of the cooling equip-
dense rack-mount server was about 5 kW per rack and in ment is second only to that of the IT equipment.
2006 it increased to 32 kW per rack. According to predic- In response to this growing energy problem, all data cen-
tions of the American Society of Heating, Refrigeration and ters have adopted new techniques to improve their energy
Air-Conditioning Engineers (ASHRAE), the thermal load efficiency [3], with most including accurate temperature
will reach 42 kW per rack by 2014 [2]. The cooling and the control and air flow management of the air conditioning
auxiliary equipment power consumptions are increasing system [4–6]. In addition, 10% of data centers have air
rapidly with a sharp increase in the number and power of cooled or water cooled equipment which can use natural
servers, which in turn rapidly increases the energy con- circulation to effectively cool the data center for free and
sumption. thus reduce the energy consumption of the cooling equip-
According to an investigation by the Lawrence Berkeley ment [7]. In ASHRAE Standard 90, naturally cooled
equipment is required in some zones in the US, especially in
the western and northeastern zones with cooler weather.
*Corresponding author (email: [email protected])

© The Author(s) 2011. This article is published with open access at Springerlink.com csb.scichina.com www.springer.com/scp
2186 Zhou F, et al. Chinese Sci Bull July (2011) Vol.56 No.20

Current investigations into ambient energy for cooling thickness of 100 mm with 0.8 mm double-faced steel panels
data centers have mainly considered air ventilation cooling, and a 18 kg/m3 bulk weight core.
including direct fresh air cooling systems [8–11], and sepa- A 1.1 kW electric heater with a fan was used to simulate
rated fresh air heat exchangers [12], which include alumi- the telecommunication cabinets. The hot air from the elec-
num plate heat exchangers and thermosyphon heat ex- tric module flowed through an air duct from the fan accord-
changers. However, the performance of direct fresh air ing to the air flow direction of the actual telecommunication
cooling systems, which are suitable for cool zones with high cabinets. A thermosyphon heat exchanger, which used am-
quality air, varies by location [13]. The actual temperature bient energy to cool the data center, was installed with a
efficiency of plate heat exchangers, which is limited by the rated heat dissipation capacity of 2 kW. The unit was 750
structure and the heat transfer mode, is not high. Ther- mm × 320 mm × 290 mm. The thermosyphon heat ex-
mosyphon heat exchangers can provide more efficient changer had two separate parts with a partition in the middle.
cooling over larger areas. Thermosyphon heat exchangers The thermosyphon core was made of standard copper tubing
have been proposed for cooling data centers [14], but spe- containing the working fluid. The thermosyphon heat ex-
cific investigations are still needed including analysis of the changer was inclined from the horizontal. Fans were in-
heat dissipation through the building envelope and the en- stalled in both the indoor and outdoor air ducts with a total
ergy consumption of the data centers using thermosyphon rated fan power of 60 W. The air conditioner in the model
heat exchangers. data center was a Media Type KFR-32GW/Y unit with a
Consequently a thermosyphon heat exchanger was built rated refrigeration capacity of 3.2 kW and a rated power of
to use ambient energy to cool a typical data center. The heat 1.19 kW, and was chosen to match the actual data center.
dissipation characteristics and the energy consumption of The heat fluxes, temperatures and active power were
the thermosyphon heat exchanger and the equivalent air measured during the experiments. The heat dissipation
conditioner system were then studied during a Beijing win- through the building envelope was calculated based on the
ter. The heat flux through the building envelope was meas- heat fluxes. The heat fluxes were measured using the Bei-
ured to confirm calculations of the data center heat dissipa- jing Shi Ji Jian Tong Type JTRG-II Building Thermal
tion through the building envelope. In addition, the effects Temperature & Heat Flux Auto Test System. The heat flux
of different indoor set temperatures and different outdoor data were recorded each minute with heat flux meters on
temperatures on the total air conditioner energy consump- each of the six surfaces as shown in Figure 1. The meas-
tion were also analyzed. urement accuracy was 5‰ and the resolution of the heat
flux meter was 0.1 W/m2. The wall positions of the heat
flux meters were chosen with a Testo Type 880 Thermal
1 Experimental systems and instrumentation Infrared Imager. The energy consumption of the air condi-
tioner was measured based on the active power consumed
Figure 1(a) shows an actual data center in Beijing with tel- by the air conditioner P in kW. The active power was
ecommunication cabinets, air conditioners, batteries, and measured each minute using a LEM Type ANALYST 3P
switch powers supplies. The data center test model in Figure Power Quality Analyser. The resolutions were 0.1 V for
1(b) was set up with the equipment in the same locations as voltage and 0.01 A for current. The operating errors were
the actual data center. The temperatures and heat fluxes 0.5% for voltage and 1% for current. The temperatures were
were measured from November 15, 2009 to January 18, measured at six points covering the indoor and outdoor
2010 during the Beijing winter. A polystyrene sandwich room temperatures and the thermosyphon heat exchanger
panel (Type 950) used as the wall material had a total inlet and outlet air temperatures. The temperatures were
measured using a YOKOGAWA Type DX 230-1-2 DAQ
STATION every 2 s with Type T 2 × 0.3 mm calibrated
thermocouples.

2 Results and discussion

2.1 Base room temperature

If there was no cooling equipment in the data center, the

heat would all be dissipated through the building envelope.
The indoor temperature measured in this condition is called
the base room temperature (TB). To get a quantitative un-
derstanding, the base room temperature was measured as a
Figure 1 Distribution of data center equipment and heat flux meters.
1, Air conditioner; 2, battery; 3, telecommunication cabinets; 4, switch power function of the outdoor temperature from 16:47, Jan. 13,
supply; 5, door; 6, thermosyphon heat exchanger; ●, heat flux meter. 2010 to 16:47, Jan. 14, 2010. The temperature variations
 Zhou F, et al. Chinese Sci Bull July (2011) Vol.56 No.20 2187

obtained in this 24-h period are shown in Figure 2.

During this period, the energy was only dissipated
through the building envelope with no auxiliary cooling. As
shown in Figure 2, the base room temperature changed as
the outdoor temperature varied, increasing as the outdoor
temperature increased and decreasing as the outdoor tem-
perature decreased. Throughout this time, however, the
temperature difference between the indoor and outdoor
temperatures remained almost constant.
Even though the outdoor temperature was below 0°C
during the testing period for these typical winter conditions,
the base room temperature was still more than 30°C. The
Figure 3 Indoor and outdoor temperatures and energy dissipation
average base room temperature was 32.7°C with the highest through the building envelope.
being 37.8°C. The temperature difference between the in-
side and outside temperatures was over 30°C. Therefore, the
heat transfer through the building envelope could not dissi-
pate all the heat input to the room on cold winter days.

2.2 Air conditioner power consumption

In the second set of tests, the air conditioner was turned on
with the temperature set at 25°C. The thermal dissipation
characteristics and the energy consumption were then
measured over a continuous 48 hour period from 20:51, Figuer 4 An air conditioner power consumption.
Nov. 15, 2009 to 20:51, Nov. 17, 2009. The variations in
the indoor and outdoor temperatures and the energy dissipa-
mally always on. The cooling system has to operate
tion through the building envelope during these tests are
throughout the year in some areas, which consumes much
shown in Figure 3.
energy. Using ambient energy (outdoor cool air) to cool the
The indoor temperature was kept constant at 25°C, while
data center will shorten the air conditioner running time and
the outdoor temperature changed as shown in Figure 3. The
reduce the energy consumption. The DeST software for
heat dissipation through the building envelope was in the
building energy consumption analysis developed by Tsing-
range 180–250 W with an average thermal dissipation of
hua University was used to model the data center. The
214 W, which is about 19.5% of the total 1.1 kW heat load.
model neglected the center’s personnel thermal load, the
The variation in power consumption of the air conditioner,
lighting load, and the cooling leakage while the door was
as shown in Figure 4, varied greatly as the air conditioner
open. The indoor temperature was set at 18–25°C with rela-
cycled on and off to maintain a constant indoor temperature.
tive humidity of 40%–70%. The start-up temperature dif-
The total power consumption of the air conditioner over the
ference of the thermosyphon heat exchanger was set to 3°C.
48-h period was 7.074 kW. Thus, the air conditioner energy
The simulation results showed that the thermosyphon heat
consumption averaged 3.5 kW over 24 h.
exchanger reduced the annual energy consumption by
around 47% in Beijing.
2.3 Ambient energy cooling
A thermosyphon heat exchanger was then installed in the
The telecommunication equipment in a data center is nor- test model data center. The energy dissipation characteris-
tics of the thermosyphon heat exchanger were then studied
for 47 h from 22:47, Jan. 16, 2010 to 21:47, Jan. 18, 2010.
The indoor and outdoor temperatures and the difference
between the indoor and outdoor temperatures with the
thermosyphon heat exchanger are shown in Figure 5.
The indoor temperature changes with respect to the out-
door temperature as shown in Figure 5. The outdoor tem-
perature affects the thermal dissipation through the building
envelope with lower outdoor temperatures resulting in more
heat dissipation through the building envelope. The energy
dissipation from the thermosyphon heat exchanger also
changes with the outdoor temperature [15]. Thus, lower
Figure 2 Base room temperature in the data center. outdoor temperatures result in more energy dissipation
2188 Zhou F, et al. Chinese Sci Bull July (2011) Vol.56 No.20

sumption would then significantly decrease to less than 41%

of the air conditioner, perhaps to as low as 1/3 of the air
conditioner power.
The thermosyphon heat exchanger would keep working
as long as the outdoor temperature is less than the indoor
temperature and the difference between the indoor and the
outdoor temperature is more than the start-up temperature
difference of the thermosyphon heat exchanger. According
to the weather data from the DeST Software, there are
5244 h each year when the outdoor temperature in Beijing is
less than 25°C and the difference between the indoor and
Figure 5 Indoor and outdoor temperatures with the thermosyphon heat outdoor temperatures is more than 3°C, which is thus occurs
exchanger.
about 60% of the year. If the thermosyphon heat exchanger
can completely replace the air conditioner for these hours
from the thermosyphon heat exchanger. Figure 5 shows that and the energy consumed by the thermosyphon heat ex-
the outdoor temperature in this period ranged from 5.8°C changer is 41% of the air conditioner, then the annual ener-
to 3.8°C, while the indoor temperature ranged from 12.8°C gy consumption of the air conditioning system will be re-
to 21.9°C, which is less than 22°C and much less than the duced by 35.4%.
26.7°C upper limit recommended by the ASHRAE Hand-
book [16]. The temperature difference between the indoor 2.4 Air conditioner set temperature
and outdoor temperatures was 15.1°C to 19.1°C. Therefore,
this thermosyphon heat exchanger and the building enve- If the air conditioner set temperature is increased, the ener-
lope can dissipate enough heat during a Beijing winter day gy consumed by the air conditioner will decrease for the
for the test model data center with a total 1.1 kW heat load, same working conditions. The 2007 ASHRAE Handbook
so there is no need to turn on the air conditioner. recommends that the inlet air temperature should be in-
The energy consumed by the thermosyphon heat ex- creased to 26.7°C from 25°C. The simulation results show
changer, using ambient energy to cool the data center, is that the annual air conditioner energy consumption will be
only due to the fans, which have a total power consumption reduced by 2%–3% in Beijing when the air conditioner
of 60 W. The thermosyphon heat exchanger worked con- temperature increases by 1°C. Thus, the energy consump-
tinuously throughout these tests, so the total energy con- tion and the thermal dissipation characteristics were studied
sumption of the thermosyphon heat exchanger over 48- for different air conditioner temperatures (26°C and 27°C).
hours was about 2.88 kW h, which was much less than the The test results and the total energy consumed by the air
heat dissipation from the thermosyphon heat exchanger [15]. conditioner are listed in Table 1.
The energy consumed by the air conditioner over a 48-h The indoor and outdoor temperatures for the different set
period was about 7 kW h as calculated in section 2.2. Thus temperatures (26°C and 27°C) are shown in Figure 6. The
the energy consumption of the thermosyphon heat exchang- indoor temperature is constant as shown in Figure 6, while
er was only about 41% of that consumed by the air condi- the outdoor temperature varies. The outdoor temperature
tioner, giving a 60% saving. A lower power fan that could fluctuates around 0°C in the first test period, while it is
still provide the required cooling would further reduce the more than 0°C in the second test period.
energy consumption of the thermosyphon heat exchanger. The variations in the power consumed by the air condi-
Also, the air conditioner operates differently from the ther- tioner for the different indoor set temperature are shown in
mosyphon heat exchanger. The air conditioner continuously Figure 7. The power consumption changes rapidly, with the
controls the indoor temperature using an automatic control total power consumption for the different indoor set tem-
system. The thermosyphon heat exchanger used here was peratures being listed in Table 1.
manually turned on and ran continuously in these tests. A The total air conditioner energy consumption for the 46-h
thermosyphon heat exchanger with an automatic control test period ranges from 8.0 to 9.0 kW h as shown in Table 1,
system would be turned on and off to further reduce the with an average energy consumption of 4 to 5 kW h for a
energy consumption. The continuous running of the ther- 24-h period. Thus, when the set temperature is increased by
mosyphon heat exchanger caused the indoor temperature to 1°C from 26°C to 27°C, the total energy consumed by the
drop to 12.8°C, which is much lower than the upper limit of air conditioner increases by 1.83%, even though the indoor
the required indoor temperature. Thus, if the shutoff tem- set temperature is higher. The total energy consumption
perature of the thermosyphon heat exchanger was set to increases for the higher indoor temperature because the
22°C (much higher than 12.8°C) and an automatic control outdoor temperature varies during the test. The higher out-
system was used, the working time of the thermosyphon door temperature will reduce the heat dissipation through
heat exchanger would be greatly reduced. The power con- the building envelope. The air conditioner thermal
 Zhou F, et al. Chinese Sci Bull July (2011) Vol.56 No.20 2189

Table 1 Total air conditioner energy consumption for different indoor set
temperatures

Indoor set Total energy con-

Start time Stop time
temperature sumption (kW h)
2009.11.21 2009.11.23
26°C 8.5
22:51 20:51
2009.11.25 2009.11.27
27°C 8.7
22:51 20:51

Figure 7 An air conditioner power consumption for different indoor

temperature.

first test and 7.634 kW h in the second test. Thus, the ener-
gy consumption increases by 11.7%. The average outdoor
temperature is about 0.1°C in the first test period and
about 1.6°C in the second test period as shown in Figure 8,
giving a temperature difference of less than 2°C. Thus, the
air conditioner energy consumption increases by 5%–6%
Figure 6 Indoor and outdoor temperatures with the air conditioner for
different indoor temperature. when the outdoor temperature increases by 1°C.

load will then increase even though the total thermal load 3 Conclusions
remains constant. Thus, even though the air conditioner
thermal load may decrease with the increasing indoor tem-
Considering the various energy management strategies pro-
perature, the outdoor temperature has a significant effect on
posed for data centers, the use of ambient energy to cool the
energy consumption.
data center with a thermosyphon heat exchanger may be the
most effective because it significantly reduces the air condi-
2.5 Outdoor temperature and air conditioner energy tioner’s running time while keeping the indoor and outdoor
consumption air totally separate, and also maintaining the humidity and
cleanliness of the data center. The heat dissipation charac-
The tests show that the outdoor temperature has a greater
teristics and the energy consumption of a thermosyphon
effect than the air conditioner set temperature on the air
heat exchanger and an air conditioner for a typical data
conditioner energy consumption. Figure 8 shows the varia-
center were investigated. The heat dissipation through the
tions of the outdoor temperature for two different time pe-
riods from Nov. 15, 2009 to Nov. 17, 2009 and from Nov.
17, 2009 to Nov. 19, 2009. The air conditioner temperature
is set at 25°C in both test periods.
The outdoor temperature is almost the same after the
30th hour for both test periods as shown in Figure 8. How-
ever, the outdoor temperature in the first period is lower
than that in the second period during the first 30 h. Thus, the
heat dissipation through the building envelope in the first
test period is higher than that in the second test period, so
the air conditioner energy consumption in the first test pe-
riod is lower than that in the second test period. The total air
conditioner energy consumption is 6.836 kW h in the Figure 8 Outdoor temperatures during two different tests.
2190 Zhou F, et al. Chinese Sci Bull July (2011) Vol.56 No.20

building envelope was also measured. The data showed 2009, 131: 021009-1-9
4 Chandrakant D P, Cullen E B, Christian B. Computational fluid dy-
that:
namic modeling of high compute density data centers to assure sys-
(1) The heat dissipation required by the data center can- tem inlet air specifications. In: Proceedings of IPACK’01, The Pacif-
not only be provided by heat transfer through the building ic Rim/ASME International Electronic Packaging Technical Confer-
envelope in Beijing during the winter. ence and Exhibition, Kauai, Hawaii, USA, 2001, IPACK2001-15622
5 Udakeri R, Mulay V, Agonafer D. Comparison of overhead supply
(2) The heat transfer through the building envelope for
and underfloor supply with rear heat exchanger in high density data
the test conditions was about 19.5% of the total thermal center clusters. In: Proceedings of 24th Annual IEEE Semiconductor
load. The average energy consumed by the air conditioner Thermal Measurement and Management Symposium, San Jose, CA,
was 4 to 5 kW h over a 24-h period depending on the out- 2008, IEEE, 166–173
6 Cho J, Lim T, Kim B S. Measurements and predictions of the air dis-
door temperature.
tribution systems in high compute density (Internet) data centers. En-
(3) The average temperature difference between the in- ergy Buildings, 2009, 41: 1107–1115
door and outdoor temperatures using the thermosyphon heat 7 Mark F. Data center energy a concern, but metrics lacking. 2008,
exchanger was less than 20°C and the energy consumed by http://searchdatacenter.techtarget.com/news/article/0,289142,sid80_g
ci1329440,00.html
the thermosyphon heat exchanger was only 41% of that of
8 Jinkyun C, Taisub L, Byungseon S K. Cooling systems for IT envi-
the air conditioner. Thus, the annual energy consumption ronment heat removal in (internet) data centers. J Asian Archit
could be reduced by 35.4%. Building Eng, 2008, 7: 387–394
(4) The outdoor temperature had a greater effect than the 9 Kumar A, Joshi Y. Use of airside economizer for data center thermal
management. In: Proceedings of 2nd International Conference on
indoor set temperature on the air conditioner energy con-
Thermal Issues in Emerging Technologies, Cairo, Egypt, 2008, IEEE,
sumption. The energy consumed by the air conditioner in- 57–66
creased by 5%–6% on average for an outdoor temperature 10 Li J R, Yang L. Numerical simulation of fresh air exchanger effecting
increase of 1°C. indoor environment. J Harbin Univ Sci Tech, 2009. 14: 127–130
11 Chen Y, Wu J, Wang F. Study on ventilation cooling technology for
telecommunication base stations in Guangzhou. Building Sci, 2008,
This work was supported by the National Natural Science Foundation of 24: 27–31
China (51076003), the Specialized Research Fund for the Doctoral Pro- 12 Bao L L, Wang J G, Zhang M J. Performance test and analysis on the
gram of Higher Education (200800050010), and the Innovation Program air heat exchanger used in communication equipment plant. Refrig
for the Doctoral Students of Beijing University of Technology (bcx-2009- Air Condi Electric Power Mach, 2008, 29: 70–73
049). 13 Shehabi A. Energy implications of economizer use in california data
centers. Lawrence Berkeley National Laboratory, 2008, http://eschola
rship.org/uc/item/4px2n6jn
1 Jonathan G K. Estimating total power consumption by servers in the 14 EPA. U. S. Environmental Protection Agency ENERGY STAR Pro-
U. S. and the world. Oakland, CA: Analytics Press. 2007, 2, http:// gram. Report to Congress on Server and Data Center Energy Effi-
enterprise.amd.com/us-en/AMD-Business/Technology-Home/Power- ciency Public Law 109–431, 2007
Management.aspx 15 Zhou F, Ma G Y. Experiment of characteristics of thermosiphon heat
2 Vincent M. Preventing a data center power crisis. 2009, http:// recovery equipment under summer condition. HV&AC, 2007, 372:
searchdatacenter.techtarget.com/tip/0,289483,sid80_gci1350324_me- 58–62
m1,00.html 16 ASHRAE. 2007 ASHRAE Handbook-HVAC Applications. Chapter
3 Iyengar M, Schmidt R. Analytical modeling for thermodynamic Atlanta: American Society of Heating, Refrigerating and Air-Con-
characterization of data center cooling systems. J Elect Packaging, ditioning Engineers, Inc, 2007

Open Access This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction
in any medium, provided the original author(s) and source are credited.