sustainable information technology ecosystem · subhiksha trading services “it will be the key...
Post on 28-Jun-2020
0 Views
Preview:
TRANSCRIPT
© 2006 Hewlett-Packard Development Company, L.P.The information contained herein is subject to change without notice
Sustainable Information Technology Ecosystemsupply and demand side management
Chandrakant D. PatelHP Fellowchandrakant.patel@hp.com
HP Laboratories, Hewlett-Packard Company
2 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Drivers of Next Generation of IT ServicesUp and coming Small and Medium Enterprises
IT Advantage in the Emerging Economies• Enabling Business Transformation
Retail Example: Subhiksha Trading Services
“IT will be the key enabler and the key differentiator between operations that are or aren’t well run. Whether it is managing the front end or logistics support or in stock and inventory management”.- R. Subramaniam, MD and Founder of Subhiksha Trading Services
- The Economic Times, March 13, 2007
Subhiksha Store
Vadodara, Gujarat, India
3 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Drivers of Next Generation of IT ServicesMicro businesses in emerging economies
IT services as a means for micro-businesses to improve quality of life if provided at a given price point
4 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Sustainable IT Ecosystem Deconstruct conventional supply chain and replace with Sustainable IT ecosystem
5 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
The Road AheadNext Generation IT Services• IT and telecommunications is a means to overcome physical infrastructure
issues such as transportation, increase productivity and improve the standard of living.
• Necessitates a total cost of ownership (TCO) that enables growth in new areas e.g. 800 million in India who want to use IT as a means to improve quality of life and business competitiveness.− Achieving the TCO necessitates management of available energy in the
IT ecosystem as a key resource.
Devising a Sustainable IT Ecosystem− Evolve an “end to end” metric and process to address sustainability in IT− Design, manufacture, operate and manage end of life of products that
minimize consumption of available energy – a “cradle to cradle”perspective
6 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Devising a Sustainable IT Ecosystemsupply side + demand side management
Ground State
Reclaim
irreversibility
Framework needed to quantify the consumption of available energy (exergy)
7 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Exergy or Available Energy
• “Exergy” is a measure of the quality of energy−Or, alternatively, the amount of work available from a
given amount of energy (‘available energy’)
• From the 2nd Law of Thermodynamics: − Irreversibilities in real processes continuously degrade
the quality of energy, simultaneously destroying exergy−For example, the conversion of coal to electricity
• Or electricity to heat• Or rejection of heat to ambient
8 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Approach
• Can a measure of the total exergy destroyed across a product’s lifetime (“lifetime exergy”) be a measure of the environmental sustainability?
manufacturing
exergy toproducematerials
exergy tofabricate and assemble components
exergy fortransportation
operationdelivery
exergy forinstallation
exergy to power electronics and thermal managment
exergy formaintainanceand repair
exergy formaintainanceand repair
retirement
exergy tode-install anddecommision
exergy todisassembleand/orrecondition for safe disposal or recycle
time
9 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
OperationPowering the electronics and thermal Management
Exergy consumed:• Electronics and electro-mechanical systems• Work required to remove heat
− Flow Work • volume flow, m3/s• pressure drop, Pa
− Thermodynamic Work• Temperature, °C
Heat Sink
Epoxy Glass Printed Circuit Board
Tin, Tout
Operation
10 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Heat Flux, W/cm2
Technology TrendsTechnology Trendschip packaging, integrated photonics, semiconductor technologychip packaging, integrated photonics, semiconductor technology
1998 2006
10
100
1000Frequency
Device Size
# of Devices
Voltage
Capacitance per device
per chip
2010
PGAEnhanced Ball Grid Array
Flip Chip Array, Underfill Flip Chip MCM
Flip Chip, Non Uniform Heat Distribution, Power Mgmt
Pow
er, W
200
3d 3d pkgpkg
400
# of Cores
• High Power Density
• Temperature Control due to optical devices
2002
Source: Chandrakant Patel projections, non industry data
11 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Thermal Management Challengestacked devices: chip and package scale
Heat SinkPassive e.g. body of the handheld
Q, W
High power density microprocessor with stacked devices
12 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Chip, Package and System ScaleWork required to remove the heat
Air (25°-40 °C)
Winterface
Demise of Passive Only Cooling Solution• Work Required at the chip interface level due to high power density and advent of new packaging
Interface
13 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Data Center ScaleWork required to remove heat Drive
T inTout
14 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Modular AC Unit
Cool Fluid Hot Fluid
Data Center CoolingConventional approach in controlling temperature in the data center
Data Center CoolingConventional approach in controlling temperature in the data center
Rack
Cooling Coil
Air Mover
RaisedFloor
Treturn at 20 °C
• Single point temperature measurement at the return of the CRAC (Computer Room Air Conditioning Unit), typically set at 20 oC
Tin ~ 14 °C
15 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
IT EcosystemBillions of Users and 1000s of Data Centers
Wchip
Wsys
Wpump
Wb
Wpump Wcompressor
Wb
Qchip
Qsystem
( ) system Cooling of WorkmicThermodyna Work FlownDissipatioHeat Total COP
+=ensemble
Patel, C.D., Sharma, R.K., Bash, C.E., Beitelmal, M, “Energy Flow in the Information Technology Stack: Introducing the Coefficient of Performance of the Ensemble”, ASME International Mechanical Engineering Congress & Exposition, November 5-10, 2006, Chicago, Illinois
Coefficient of Performance of the Ensemble
Wensemble
Qdata center
16 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Coefficient of Performance of the Ensemble
( )∑ ∑∑∑ ∑ ++++
+
+
=
−−k
ctcompm
pl
crbj
ri
devcp
dcG
WWWWWWW
QCOPsup
Cooling Tower loop
Chiller Refrigerant loop
Chilled Water loop
Data Center CRAC units
Warm Water
Air Mixture In
QCond
Wcomp
Air Mixture In
Return Water
QEvap
Cooling tower wall are adiabatic, Q = 0
Makeup Water
Air Mixture Out
Wp
Wp
( )∑=ondarypdchydronics WQCOP
sec
compWchQ
chCOP =
−
−
−= 1
/)1(
2
3)1(
22nn
P
P
nmotor
nPrefmcompW
pηη
ν&
ct
compdcctctct W
WQWQCOP
+==
Patel, C.D., Sharma, R.K., Bash, C.E., Beitelmal, M, “Energy Flow in the Information Technology Stack: Introducing theCoefficient of Performance of the Ensemble”, ASME International Mechanical Engineering Congress & Exposition, November 5-10, 2006, Chicago, Illinois
17 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
ImpactFlow Work + Thermodynamic Work
Heat Generated
Energy to Remove Heat
2 MW
2 MW
10 to 100 W
Services
20 GWTotal
Global Power Consumption
Annualized Destruction of Available Resources – Coal
2-40 W
48 GW Assume:
5000 data centers
Assume: 4 billion handhelds at 12 W
200 Million Metric Tons of coal
85 Million Metric Tons
175 Million Metric Tons CO2 emission
Operation
18 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Data Center Total Cost of Ownership
( ) ( ) ( )1$,12112
2 1,$ σ+++++++
= depavgtotal
hardwareconsumedgridcriticaltotal ITSMRPULKLKftA
ftCost
Personnel, equipment, SW per rackBurdened power consumption
Real Estate
• Responsiveness• Energy Consumption
Factors that impact the “Burdened Power Cost”
J1 : capacity utilization factor, i.e. ratio of maximum design (rated) power consumption to the actual data center power consumption
K1 = F(J1): burdened power delivery factor, i.e. ratio of amortization and maintenance costs of the power delivery systems to the cost of grid power
K2 = F(J1): burdened cooling cost factor, i.e. ratio of amortization and maintenance costs of the cooling equipment to the cost of grid power
L1: cooling load factor, i.e. ratio of power consumed by cooling equipment to the power consumed by compute, storage and networking hardware (inverse of COPensemble) D
epre
ciat
ion
fact
ors
↓ J1
↓ L1
Lack of vital information from single-input single-output environmental control systems
Patel and Shah, Cost Model for Planning, Development and Operation of a Data Centerhttp://www.hpl.hp.com/techreports/2005/HPL-2005-107R1.html
19 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Compute Power Cooling
Flexible & Configurable Elements
Sensing Infrastructure
Policy based Control Engines, Tools
commodity flexible building blocks that enable dynamic change in configuration
compute, power, cooling resources provisioned based on the need
Smart IT IT-Facility End to End Management
T in at 25 °C, 27 °C ?
What is the operational limit?
20 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Dynamic Smart Cooling Architecture
AC Unit AC Unit
Equipment Racks
T
Room Chilled Water Supply
Ventilation Tiles
Modular Building Control
Floor Level Network (optional)
T
H H
VFDVFD
• CRAC temperature control sensors moved from return to supply side;
• Variable Frequency Drives added to AC units;
• Temperature sensors added to rack inlets;
• Advanced algorithms control operation of AC units.
T in
21 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
HP Labs Smart Data CenterPalo Alto, California
CRAC 292 KW
CRAC 628 KW
CRAC 537 KW
CRAC 439 KW
CRAC 380kW
CRAC 194KW
Res
earc
h
RIT RIT
RIT
RITRIT
Res
earc
h
CRAC 1 CRAC 3
CRA
C 4
CRAC 2
CRA
C 6
CRA
C 5
PDU
PDUPDU
HD Area:
140 KW from 14 fully loaded racks in 400 square feet
HD Area
High Performance Cluster – low load in this state
22 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Conventional Mode
Dynamic Smart Cooling Mode
HP Labs Data Center
Palo Alto, CA
• 35% Energy Savings
• Improved reliability
23 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Local Disturbances – Vent Tile Blockage
50
55
60
65
70
75
80
85
90
11:0
0:17
11:1
7:17
11:3
4:17
11:5
1:17
12:0
8:17
12:2
5:41
12:4
2:41
12:5
9:41
13:1
6:41
13:3
3:41
13:5
0:41
14:0
7:41
14:2
4:41
14:4
1:41
14:5
8:41
15:1
5:41
15:3
2:41
15:4
9:41
16:0
6:44
16:2
4:00
16:4
1:00
16:5
8:00
17:1
5:22
17:3
2:22
17:4
9:22
18:0
6:25
18:2
3:26
18:4
0:27
18:5
7:28
19:1
4:32
19:3
1:33
19:4
8:34
Time
CR
AC
Sup
ply
Air
Tem
p (F
)
0
10
20
30
40
50
60
70
80
90
Rac
k B
4 In
let T
empe
ratu
re CRAC 1CRAC 2CRAC 3CRAC 4CRAC 5CRAC 6B4T1B4T2
September 7, 2006
DSC max allowable inlet temp
CRACs take evasive action
Tiles Blocked Blockage removed
Scenario: IT department performed maintenance on a rack and covered nearby vent tiles for 4 hours with bag. DSC responds and keeps impacted racks below maximum.
B4
24 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Global Disturbance – Chiller Failure
Scenario: Groundskeeper blows leaves into primary cooling tower during routine cleanup. Leaves block filter in cooling tower retarding condenser water flow that ultimately results in chiller failure. DSC delayed the impact of the failure allowing for repair.
0
10
20
30
40
50
60
70
80
10:0
0:05
10:2
2:35
10:4
5:05
11:0
7:35
11:3
0:05
11:5
2:35
12:1
5:05
12:3
7:35
13:0
0:05
13:2
2:35
13:4
5:05
14:0
7:35
14:3
0:05
14:5
2:35
15:1
5:05
15:3
7:35
16:0
0:05
16:2
2:35
16:4
5:05
17:0
7:35
17:3
0:05
17:5
2:35
18:1
5:05
18:3
7:35
19:0
0:05
19:2
2:40
19:4
5:10
20:0
8:08
20:3
0:38
20:5
3:08
21:1
5:38
21:3
8:08
Time
CR
AC
Sup
ply
Air
Tem
p (F
)
15
20
25
30
35
40
45
50
Rac
k In
let T
empe
ratu
re (C
) CRAC 1CRAC 2CRAC 3CRAC 4CRAC 5CRAC 6Rack E7Rack B4Rack A7
Primary Chiller Failure Chiller Back On
Maximum Allowable Inlet Temperature
CRAC 6 (DX)
Early reaction of DSC keeps temperatures under control until chilled water temperature reaches 74 F. Allows more time for resolution of problem.
25 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
HP R&D Lab-Data Center Dynamic Smart Cooling Implementation in Bangalore
•Scope/Sizing
•Layout/Topology
•Comm. Fabric/Integration
•Control and Monitoring
•Management
• ServersNon-Stop serversProliant serversBlade serversCustom Enclosures
• Storage (XP/EVA)• Multiple Network topologies
• Sensor Network•7500 sensors
5 floors @14k sq. ft.900kW cooling per floor
•Chillers3 air-cooled2 water-cooled
•Pumps7 Primary5 Secondary
•CRAC units55 units
•Diesel Generators
Facility Building Blocks IT Building Blocks
26 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Supply Side and Demand Side Management
Demand Side (Consumption of Resources):• Library of flexible and scalable building blocks overlaid with
pervasive sensing and control to provision resources based on the need
Supply Side (Availability of Resources)• Devise a 2nd law based tool that enables a “cradle to cradle”
approach in quantifying and designing the IT ecosystem • Can it be used to:
− Dematerialize the ecosystem i.e. least materials design− Seek out appropriate materials− Seek out appropriate distributed energy sources− Seek out analytical techniques to address operations and “end of
life”• leverage the rich instrumentation used for demand side
management to detect and understand anomalies
manufacturing operationdelivery retirement time
27 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Dematerializing the EcosystemData Center Management: modeling, measurement and inference
Reduce redundancy, manage “end of life”:• Minimize redundancy in the data center facility infrastructure
− Example: empirical data and inference techniques to eliminate excessive redundancy e.g. standby air conditioners
• Hardware “Damage Boundary*” [1] and minimizing hardware redundancy− IT-Facility measurement and inference techniques in place, can we push the
limits of operation• Example: Understanding the impact inlet temperature Tin to life of components by
having thermo-mechanical models in place • Understanding thermo-mechanical behavior to determine root cause of failure and
manage “end of life”• Drive reliability studies have been presented before at USENIX using large
samples, SMASH data – opportunity to extend the work [2][3]
*Used for fragility assessment – can this be extended to other areas to determine operational limts[1] Lee Hedtke, Chandrakant Patel, Damage Boundary Testing of Hard Disk Drives, Proceedings of the ASME
Winter Annual Meeting, 1990[2] Pinheiro, Wolf-Dietrich Weber, Luiz André Barroso. Failure Trends in Large Disk Drive Population. In
Proceedings of the 5th USENIX Conference on File and Storage Technologies (FAST ’08), February 2007.[3] Weihang Jiang, Chongfeng Hu, Yuanyuan Zhou and Arkady Kanevsky, Are Disks the Dominant
Contributor for Storage Failures, 6th USENIX-FAST, February 2008
retirementoperationmanufacturing
28 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Rigid Drive Mechanism (Thermal)
Tin, °C
Tdrive,in
Tdrive core
Tdrive core,°C
Qdrive spindle
retirementoperationmanufacturing
29 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Thermo-Volume ResistanceExpedient Thermo-Fluids Modeling of Systems
“Thermo-Volume” [Ref] Resistance
V', m3/s
Thermal Resistance
System Blower (s)
Extract the temperature at a given location
e.g. chip, hard drive temperature
Tchip
vbypass
v
VolumeResistance
2v21v
21
ductduct
tduct
duct
ldf
df
xP ρρ +=∂∂
x•Patel and Belady, ECTC1997
•Bash and Patel, ISPS1999
•Patel and Bash, Thermal Management Course Notes, 2007
Epoxy Glass Printed Circuit Board
Can we represent sub-systems as “thermo-volumes” with thermo-mechanical attributes?
30 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Rigid Drive Mechanism (Mechanical)
Susceptibility to vibration:
• over the years, stiffness to mass ratio has gone up due to miniaturization
•however, in-situ disturbances in data centers can coincide with natural frequency of the arm
retirement
31 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Dematerializing the EcosystemLifetime Exergy Advisor - Case Study of Sample PC
Ref: HP Labs – UC Berkeley data
manufacturing
Extraction Data
32 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Joules: Currency of the Flat Worldecosystem: billions of handhelds and printers, thousands of data centers and print factories
Cradle to Cradle Design: Least Material, Appropriate materials based on the 2nd Law of Thermodynamics
Least Energy through need based provisioning of resources
Sustainable IT Ecosystem0
10
20
30
40
50
60
70
80
10:0
0:05
10:2
2:35
10:4
5:05
11:0
7:35
11:3
0:05
11:5
2:35
12:1
5:05
12:3
7:35
13:0
0:05
13:2
2:35
13:4
5:05
14:0
7:35
14:3
0:05
14:5
2:35
15:1
5:05
15:3
7:35
16:0
0:05
16:2
2:35
16:4
5:05
17:0
7:35
17:3
0:05
17:5
2:35
18:1
5:05
18:3
7:35
19:0
0:05
19:2
2:40
19:4
5:10
20:0
8:08
20:3
0:38
20:5
3:08
21:1
5:38
21:3
8:08
Time
CR
AC
Sup
ply
Air
Tem
p (F
)
15
20
25
30
35
40
45
50
Rac
k In
let T
empe
ratu
re (C
) CRAC 1CRAC 2CRAC 3CRAC 4CRAC 5CRAC 6Rack E7Rack B4Rack A7
Primary Chiller Failure Chiller Back On
Maximum Allowable Inlet Temperature
CRAC 6 (DX)
Early reaction of DSC keeps temperatures under control until chilled water temperature reaches 74 F. Allows more time for resolution of problem.
Opportunity for collaboration CS-ME-EE
33 7 March 2008 Chandrakant Patel, chandrakant.patel@hp.com
Thank You
• I thank the USENIX-FAST organizers for giving me this opportunity
• I thank my HPL colleagues &• I thank you for your time
top related