building a „real‟ data · challenges converging on data centers it density increased stress on...
TRANSCRIPT
Building a „Real‟ Data
Center During the Great
Recession
Steven Carlini Sr Director, Data Center Global Solution Offer Management
APC by Schneider Electric
Schneider Electric at a glanceglobal, innovative and responsible
Listed on the Paris
Stock Market – CAC40
billion euros sales in 2009
people in 100+ countries
in Fortune 500 ranking
Sales by geography – 2009
North
America
27%Asia
Pacific21%
RoW11%
Western
Europe
35%
EasternEurope
6%
of salesin new economies
Specialist in
Energy ManagementGlobal
of sales devoted to R&D
Innovative
Responsible • Environment
• Ethics
• Business
• Access to energy
Schneider Electric at a glancecovering all end markets
Buildings
Data Centers Energy &
Infrastructures
Industry
Residential
31%Healthcare
Offices
Retail
18%
25%
Electrical power
Oil & gas
Marine
10%
16%
Machine
construction
Mining, minerals
& metals
Water
Industrial
% of 2009 sales
IT & Finance
Individual &
collective housing
The energy dilemma is here to stay
vsEnergy demandBy 2050Electricity by 2030
CO2 emissions to
avoid dramatic climate
changes by 2050
The facts The need
Source: IEA 2007
Source: IPCC 2007, figure (vs. 1990 level)
Energy management is a key
to address the dilemma
From production to consumption the
energy train
Coal100 units 35 units 33 units
unit
saved at the data center
( physical infrastructure)
units
saved at the power plant
unit
saved at the server
31
1units
saved at the power plant
6
Companies should focus on end of the chain
to save energy
Challenges converging on data centers
●IT density
Increased stress on power and cooling due to …
● Centralization / consolidation of distributed data centers
● High density technology – Virtualization, on-demand
computing, multi-core processors
●Energy efficiency
A growing a global awareness resulting in …
● Data centers in the spotlight due to high energy
usage that is increasing rapidly
● Opportunity to redirect cost savings toward the business
●Design / build cycle time
Need for agility, speed, predictability in response to
escalating pace of change in …
● Business climate
● Service quality level and business continuity standards
● Compliance requirements
… all putting the squeeze on AVAILABILITY
AVAILABILITY is the ever-present constraint in
data center planning
Why build a new data center
Data centers
Go green
• Leveraged more
energy efficient IT
equipment
• Energy and
management
measurement software
•Reduction of carbon
footprint
Go fast
• Shorten design build
cycle time
• Design with
consideration of
preferences vs.
constraints
• Use tools to evaluate
design impacts
Go dense
• Data center
consolidation
program
• Virtualization of
servers & storage
• Consolidation of
applications
Overheard in planning meetings…
“We‟d like to have multiple data center
proposals ready to present to management, but
that‟s cost prohibitive”
“We redesigned our facility several times
before our plan got approved”
“We organized our resources to support an
$8 million project and then were given a $1
million budget”
The Root Cause of These Problems is PLANNING
Why projects succeed or fail
Source: The Standish Group, 1995
User involvement
Executive mgmt support
Clear statement of requirements
Realistic expectations
Smaller project milestones
Competent staff
Ownership
Clear vision and objectives
Proper planning
0% 5% 10% 15%
Lack of user involvement
Lack of resources
Lack of executive support
Changing requirements and specs
Lack of planning
Don’t need any longer
Lack of IT management
Unrealistic expectations
Incomplete requirements
0% 5% 10%
Why projects succeed Why projects are impaired
Lessons From LEED
PLANNING - Better Buildings, Same Cost
Surprise lesson learned from LEED projects:
LEED-certified buildings cost the same as “normal” buildings
“ The only effective way to budget for sustainable features within buildings
is to identify the goals, and build an appropriate cost model for them.
… It is possible to establish goals and budgets from the very beginning of
the project. Other methods are ineffective and unnecessary.”
● Establish team goals, expectations, and expertise
● Include specific goals in the program
● Align budget with program
● Stay on track through design and construction
How?
Source: The Cost of Green Revisted - Davis Langdon report, 2007
St. Louis data center – Planning timeline
Decision to build
state-of-the-art,
highly-efficient
data center
Data Centers
burdened by
Power, Cooling,
Space challenges
in West Kingston
Reduction of
original budget by
50%
Review of the
original design
due to new
constraints
Data center to be
co-located with
Schneider
Technology
Center in St. Louis
APC acquired by
Schneider Electric
Schneider Electric
is comprised of
well-known brands
including TAC,
Square D,
PowerLogic and
more
=S= was
operating 330
data rooms &
number of data
centers globally
Budget revision
caused a two week
deadline on creation
of new plan
Build completedConsolidation
project in final
stages
Over 500 servers
moved
67% servers
virtualized as of
April 2010
St. Louis data center – moving fast
Technical decisions:
First Floor of SETC
2N UPS Redundancy
4500sq feet
415V, Modular, Row-
based Data Center
Core SME Group
Engaged – pulled for
specialized Data
Center design build
knowledge
The end result – Implemented July, 2009
DP-UPS1
DP
-UP
S2
DP-UPS3
HP-CDC
LP-1D
T1D
CDC-HP-C1
TC1
500A CDC-HP-C2
TC2
500A CDC-HP-C3
TC3
500A
SAN
8300
Front
FrontFront / cold aisle
250 kW Odin - A1 250 kW Odin - A2
RACK 18 RACK 19 RACK 20 RACK 21 RACK 22 RACK 23 RACK 24 RACK 25 RACK 26 RACK 27
RACK 28 RACK 29 RACK 30 RACK 31 RACK 32 RACK 33 RACK 34 RACK 35 RACK 36 RACK 37
R
P
P
A2
R
P
P
B2
Design allows for future
expansion without shutdown
Extra HACs
Extra HACs
• 4,500 ft2 (400m2)
• Average density 5.8 kW/rack
• Zone based approach
● PUE 2.86 1.52 (with a target of 1.35 @ fully loaded)
● Build with scalable, energy-efficient technologies● UPS rightsizing, In-row cooling
● Hot aisle containment
● Energy smart servers
● Future ready● Connections to economizer and solar panels in place
Budget cuts didn’t stop us from going efficient
... Nor compromise services delivered
Capacity
Availability
Security
160 kW 450 kW(expand up to 1.3 MW)
• Multiple SPOF
• Minimal redundancy
Data center availability
~ 99.98%
• Limited access control• Biometric access control
• Rack level access control
2,500 sq ft 4,500 sq ft
Old data center St. Louis
LeanVirtualization
~ 55%Virtualization
~ 67%
Trade-offs lead to gains
ORIGINALDesign
ADJUSTEDDesign
Capacity• 1.3 MW capacity
• 600 m2 (6500 ft2)
2008/2009 economic
crisis
Budget cut
by 50%
Capacity• 700 KW capacity
• 400 m2 (4500 ft2)
Efficiency• Top priority
• PUE 1.34 at full load
Efficiency• Still top priority
• PUE 1.35 at full load
Redundancy• 2N dedicated chiller
• 2N piping, 2N air handlers
• 2N power path (1 onsite substation)
• 2N standby generators
Redundancy• Existing N+1 building chiller
• Existing N piping, N+1 air handlers
• 2N power path (1 onsite substation)
• Existing N building generator
Energy management• Real-time efficiency and energy
usage measurement
Energy management• Real-time efficiency and energy
usage measurement
Changed
Changed
Trade-offs – Planning for the future
Economizer
• Reduce cooling costs
aka “free’ cooling
• Reduce PUE
• Require air filtration
• Humidity an issue
• Estimated Cost -
$230,000
Solar/Alternative
Energy
• Going green
• Internal energy
generation
• Currently costly to
generate power
• Estimated Cost -
100,000 panels needed
to generate 1MW of
power
High Efficiency
Chiller
• Reduce data center
electrical
consumption
• Reduce PUE (when
used with an
economizer)
• Reduce
maintenance costs
• Estimated Cost -
$370,000
2008/2009
economic crisis
We balanced preferences and constraintsPreference:
Result:
• Dedicated chillers
• Dedicated diesel generator
• Water side Economizer
• Dedicated electrical substation
• Executive conference room overlooking data center
• Unfinished space on 1st floor
Constraint:
• Using existing 2MW stand-by diesel generator
• Using “base building” chiller plant
• Capacity to meet 48 month window – 2N - 400kW UPS
• Growth to 750kW 2N UPS and N+2 Cooling without shutdown
• Ultimate growth 1500kW 2N UPS and N+2 Cooling
• Criticality – Equal to Tier III
• Efficiency – PUE of 1.52 as built / 1.35 with future economizer
Evaluate
Preferences
& Constraints
Trade-Off
Analysis
Good planning with use of reference design enabled the success of the project
IT equipment at a glance
• 4,500 ft2 (400m2)
• 500+ servers
• 50% of servers virtualized
Servers
Storage
Network
Dell PowerEdge (rack and blade)
HP Integrity (RISC and Itanium)
9000
IBM Bladecenter L&H
p595 Power 6 / AIX HACMP
p520 TSM
HS21 blades / Novell SUSE Linux
Intel standalone servers
EMC
IBM Quantum tape library
3584 LTO tape backup
DS8300 SAN
Cisco Catalyst 6500 switches
Design considerations at a glance
415/240V distribution throughout the data center
Row-based cooling equipment architecture
Hot Aisle containment system
Metered and controlled chilled water loop
Modular equipment architecture
2 Megawatt existing onsite generator
34KV 12.5KV existing onsite substation
Schneider Electric products usedSupplier Subsystem
Pelco Whole-room video security
Management
NetBotz Rack-level video security
TAC Building management system
APCManagement system for physical infrastructure
Ion E Power monitoring
PowerLogic Metering and instrumentation
Power
Square D Switchgear and transformers
APC UPS
Defem Cable trays
Caterpillar Generator
APC Row-based cooling
Cooling
APC Hot-aisle containment
Veris Metering
Armstrong Int‟l Humidifier
Carrier Chillers
APC Racks
OtherNotifier Fire detection
Front / cold aisle
RACK 18 RACK 19 RACK 20 RACK 21 RACK 22 RACK 23 RACK 24 RACK 25 RACK 26 RACK 27
RACK 28 RACK 29 RACK 30 RACK 31 RACK 32 RACK 33 RACK 34 RACK 35 RACK 36 RACK 37
R
P
P
A2
R
P
P
B2
A data center “zone”
Hot aisle
Cold aisle
Cold aisle
6 Zones organized by density not application
88768136
Actual PUE is not staticSpot interventions done
in the data center
Long-term trend
as IT load grows
Questions?
1.52
Thank You!