Upload
hoangque
View
215
Download
2
Embed Size (px)
Citation preview
Data Center Airflow:A Predictive Model
Roger R. [email protected]
Chief Thermal Architect, IBM Corporation
Suhas V. [email protected]
President, Innovative Research, Inc.
Agenda
• Product Heat Load Trends• Data Center/Equipment Cooling• Data Center Flow Measurements• Data Center Predictive Model/
Comparison with Measurements
Microprocessor Power Trends
1995 2000 2005 2010 2015Year
60
80
100
120
140
160
180
200
Max
imum
chi
p po
wer
(W) 1994 NTRS 1997 NTRS 1999 ITRS
26 - 43 W/cm2
30 - 88 W/cm2
Consortium of Computer and Telecom ManufacturersWill be published by Site Uptime Group
• Amdahl • Compaq• Cisco • Cray • Dell • HP
• IBM • Lucent • Nortel • Sun • Unisys
Heat Load Trends
Data Processing/Telecom Equipment Heat Load Trends
Year of Announcement1992 1994 1996 1998 2000 2002 2004 2006 2008 2010
Hea
t Loa
d pe
r Mac
hine
/Equ
ipm
ent A
rea
- wat
ts/ft
2
6080
200
400
600800
2000
4000
60008000
100
1000
10000
Hea
t Loa
d pe
r Mac
hine
/Equ
ipm
ent A
rea
- wat
ts/m
2
800
2000
4000
60008000
20000
40000
6000080000
1000
10000
100000
TapesWorkstationsServers & DASD(tall racks)
Communication Equipment(Frames)
Projections for Heat Loads
The Site Uptime Group
IBM’s xSeries Servers
1 frame7.7 sq. ft. 7.6 kW max Max 982 watts/ft2
1800 lbs(133 lbs/sq.ft.)
Environmental Facts
IBM’s zSeries Servers
1 or 2 frames14.4 sq. ft. for 1 fr.7.6 kW max for 1 fr.Max 527 watts/ft2
60 Amp dual power plugs2621 lbs(121 lbs/ft2)
Environmental Facts
Product Family Heat Density Trend Chart
Year of Product First Shipment1992 1994 1996 1998 2000 2002 2004 2006 2008 2010
Hea
t Loa
d pe
r Pro
duct
Foo
tprin
t - w
atts
/ft2 -fo
otpr
int
6080
200
400
600800
2000
4000
60008000
100
1000
10000
Hea
t Loa
d pe
r Pro
duct
Foo
tprin
t - w
atts
/m2 -fo
otpr
int
800
2000
4000
60008000
20000
40000
6000080000
1000
10000
100000
CommunicationEquipment(Frames)
Workstations(standalone)
Tape Storage SystemsServers & Disk Storage Systems (1.8-2.2m tall)
xx
The Site Uptime Group
zSeries Cluster
8 tiles
16 tiles
527 watts/ft2
based on systemfootprint
263 watts/ft2
based on minimumsystem service clearances
Typical Functional Allocation within Data Center
Electrically active IT hardware - 30%Service Clearances around products - 30%Site infrastructure support equipment - 20%Main aisles and other inactive areas - 20%
Complete Data Center
0.30 x 527 watts/ft2 = 158 watts/ft2
Partially Configured Systems(Ave. of 50% total load) 0.50 x 158 watts/ft2=79 watts/ft2
Agenda
• Product Heat Load Trends• Data Center/Equipment Cooling• Data Center Flow Measurements• Data Center Predictive Model/
Comparison with Measurements
CRAC CRAC
Servers/Storage
Servers/Storage
Servers/Storage
Servers/Storage
Subfloor
Raised Floor
Computer Data Center Cooling
Perforated Floor Tiles (6, 11, 25, 40, 60% open)
Data Center Cooling
Data Center Characteristics
Raised Floor
Sub floor
zSeries
EPO Panel
dT
cfm
Power Dissipated P
dT = fcn(cfm,P)
Perforated Floor Tile
Objective: Maintain EPO panel thermistors below 35C
Chiller Air Flowrate Required for zSeries
Area of Floor Tiles to SupplyChilled Air to Freeway System
zSeries
Other Computer Systems
Centerline ofAisle Walkway
29.5" to 32.6"
65.4"
(depends on whether side covers are installed)
Rear
Front
Chilled Air Flow Requirements
Underfloor Air Temperature(C)8 10 12 14 16 18 20 22 24 26
Air F
low
Rat
e(cf
m) f
or c
ross
-hat
ched
are
a
200
400
600
800
1000
1200
1400
1600
4.0 kW
6.5 kW
9.0 kW
Agenda
• Product Heat Load Trends• Data Center/Equipment Cooling• Data Center Flow Measurements• Data Center Predictive Model/
Comparison with Measurements
Data Center Floor Plan - Test Area
30'
26'
L
L
Current Flow Test Area
66'
20'
CRAC Units
Test Area Perforated Tile Test CasesA
B
AB
AB
4 x
15 ti
les
2 x
15 ti
les
AB
1 x
10 ti
les
1 x
10 ti
les
1 x
10 ti
les
2 x 7 tiles
Case 1 Case 2 Case 3 Case 4
2 x 7 tiles
2 x 7 tiles
2 x 7 tiles
AB
4 x
15 ti
les
4 x
4 ti
les
Case 5
1 x 7 tiles
Calibrated Flow Tool
Boxed-In Test Area
“25%” Open Floor Tile
Data Center Test - Case 1
AB
4 x
15 ti
les
1 2 3 4 5 6 7 8 9 10 1112 1314 15
Tile Row no.
-300
-100
100
300
Air F
low
rate
(cfm
)
Flow Distribution - Test #10
Data Center Test - Case 4
Wes
t 1W
est 2
Wes
t 3W
est 4
Wes
t 5W
est 6
Wes
t 7W
est 8
Wes
t 9W
est 1
0W
est 1
1W
est 1
2C
hille
r 1 2 3 4 5 6 7 8 910 11 12 13 14 15
Tile Row no.
-200-100
0100200300400500
Air F
low
rate
(cfm
)
AB
2 x 7 tiles2 x 7 tiles
2 x 7 tiles
2 x 7 tiles
1 x 7 tiles
Data Center Test - Case 5
Wes
t 1W
est 2
Wes
t 3W
est 4
Wes
t 5W
est 6
Wes
t 7W
est 8
Wes
t 9W
est 1
0W
est 1
1W
est 1
2C
hille
r 1 2 3 4 5 6 7 8 910 11 12 13 14 15
Tile row no.
-200-100
0100200300400500
Air F
low
rate
(cfm
)
AB
1 x
10 ti
les
1 x
10 ti
les
1 x
10 ti
les
Agenda
• Product Heat Load Trends• Data Center/Equipment Cooling• Data Center Flow Measurements• Data Center Predictive Model/
Comparison with Measurements
Objectives of the Model
• Calculation of airflow distribution through the floor tiles for the whole data center
• Ability to vary chiller flow, tile perforations, and raised-floor height
• Easy-to-use, fast, flexible• Intended for routine use by data-center
designers and operators
Basis of the Model
• A two-dimensional (depth-averaged) calculation of the velocity and pressure fields under the raised floor is performed.
• The chiller flow is specified as inflow.• The outflow through the perforated tiles is
calculated from:∆p = A Q2
where ∆p is the pressure drop andQ is the volumetric flow rate
The Cause of Flow Maldistribution
Results of the Model
Comparison With Measurements
Comparison With Measurements
Comparison With Measurements
Comparison With Measurements
Flow Maldistribution Revisited
Effect of Tile Open Area
Effect of Plenum Height
Industry Needs
• Heat loads of Data Processing Equipment are increasing.
• Sufficient chilled airflow must be supplied to maintain high reliability.
• Complex layouts of systems, chillers, and perforated tiles present extra challenges.
• A validated predictive tool is needed to aid in balancing/optimizing airflow distribution throughout the data center.
Our Contribution
• Detailed airflow measurements have been conducted on a data center with a variety of chiller and tile arrangements.
• A computational model has been developed for the prediction of the flow through the perforated tiles.
• The model predictions give good agreement with the measurements.
• One run of the model requires 1-2 minutes on a 300MHz Pentium computer.