Data Centres - Today’s ChangingLandscape
Tony Day Global Director Data Centre Projects & Professional Services for
APC by Schneider Electric
Solutions of the pastare not sufficient for the future
The world’s energy system is at a crossroads. Curre nt global trends in energy supply and consumption are
patently unsustainablepatently unsustainablepatently unsustainablepatently unsustainable…
But that can – and must – be altered; theretheretherethere’’’’s still time to s still time to s still time to s still time to
change the road wechange the road wechange the road wechange the road we’’’’re onre onre onre on
… what is needed is nothing short of an energy energy energy energy
revolutionrevolutionrevolutionrevolution
Source: World Energy Outlook
“
“
The ICT sector
• Technology is key – earlier what used to be done via Human interaction has become technology driven – this is becoming a key business differentiation.
• “A crucial part of the solution to climate change” –European Commission Communication to the Parliament on “Mobilizing Information & Communication Technologies to facilitate the transition to an energy-efficient low-carbon economy”
• The way to “save the first billion tons “ of carbon –World Wildlife Fund.
• “The intelligent application of ICT can reduce our annual global emissions by 15% by 2020 (approx 7.8Gt of carbon dioxide equivalent to carbon savings five times larger than the total emissions from the IT sector and €600B of cost savings)-The Global eSustainability Initiative SMART 2020 report.
Double-edged squeeze
The PLANET
Reduce carbon footprint !Reduce carbon footprint !
The BUSINESS
More computing per watt !More computing per watt !
Data center planning and operation is under increasing pressures
Energy and service Energy and service cost control pressurecost control pressure
Increasing availability Increasing availability expectationsexpectations
Regulatory Regulatory requirementsrequirements
Server Server consolidationconsolidation
Rapid changes in Rapid changes in IT technologyIT technology
High density High density blade server blade server power/heatpower/heat
Dynamic power Dynamic power variationvariation
UncertainUncertainlonglong --term plans for term plans for capacity or densitycapacity or density
In response, we need to change the way the In response, we need to change the way the world designs, installs, operates, manages, and world designs, installs, operates, manages, and
maintains data centersmaintains data centers
Time to valueTime to value
Increasing regulation & codification● UK Carbon Reduction Commitment (CRC)● EU Code of Conduct for Datacentre’s (EU CoC)● EU Energy Performance Building Directive● Revision of Part L of The Building Regulations
2000 “Conservation of fuel and power● Climate Change Levy● F-Gas regulation● EU Eco-design Directive for Energy Using
Products, Energy Star, Eurovent certification● BREEAM, LEED certification,● Carbon Trust ECA Scheme● Green Grid metrics PUE/DCIE● BCS● Financial instruments● Standards ANSI-TIA-942-2005 / ASHRAE TC9.9
etc, ● Legislators should always engage and
consult with the industry to ensure the mosteffective means of dealing with the issues ofenergy consumption are implemented.
• The Green Grid global consortium is dedicated to standards, measurement, processes, and technologies to improve data center performance
• The U.S. Environmental Protection Agency (EPA) is defining data center efficiency standards (Energy Star ratings)
• The European Commission’s Institute for Energy is defining a “Code of Conduct” for data center efficiency
• Enterprise companies are starting to make public carbon commitments
Industry-wide movement is underway to shape policy and behavior
is a founding member and on the board of directors of The Green Grid
Data center governance
Efficiency canEfficiency can ’’ t be ignored anymoret be ignored anymore
Lessons Learned?
“The farther backward you can look, the farther forward you
are likely to see."Winston Churchill
ENIAC - 1945
1970/1985 - Big Iron
IBM 370/168 –1972/1980
•Standard Hardware
•Standard Applications
•Service Levels
•High Cost
•Little Flexibility
1985/1995 - Client/Server Revolution
•Platform Choice
•Many Suppliers
•Chaotic Management
•Low Service Quality
1995/2000 - Internet Revolution (a.k.a. The Dotcom Bubble)
Unnamed Colo , London 2000
•Increased Access
•Uncontrolled Growth
•Inefficient Deployments
•Unintegrated Applications
•Minimal Management
•Low Service
2000/2009 - Age of Maturity
Ferrari 2005
•Fewer Platform Choices
•Consolidation
•Improved Measurement
•SLA’s/ Metrics
•Virtualisation
The New Decade – Into the Cloud•Datacenters no longer a collection of disparate storage devices, communications equipment and servers
•Hardware & software resources working to deliver unparalleled levels of efficiency, availability & performance
•Holistic approach to design & deployment
•The datacenter is the computer
Imperatives for the next decade
• Faster time to value• Higher availability• Lower total cost of ownership
–Capex–Installation–Energy Costs–Maintenance
• Better efficiency• Integrated management systems
The Newest Challenge: EFFICIENCY
Provide power and cooling in the amount needed,
when needed, and wherewherewherewhere needed – but no more
than what is required for redundancy and safety
margins
But we canBut we canBut we canBut we can’’’’t manage what we cant manage what we cant manage what we cant manage what we can’’’’t t t t
measuremeasuremeasuremeasure
Efficiency target:
POWER
system
POWER
system
COOLING
system
COOLING
system
Datacenter Efficiency Data Center Physical Infrastructure
IT
Rack-row-room-building
EndEnd --toto --end, wrapend, wrap --around portfolio enables around portfolio enables holistic designholistic design for highest availability and efficiencyfor highest availability and efficiency
Key Energy consumption data points
• Typically 50% of power going into a datacenter goes to the power and cooling systems – NOT to the IT loads
• Every kW saved in a datacenter saves about £630 ($1,000) per year
• A 1% improvement in datacenter infrastructure efficiency (DCiE) corresponds to approximately 2% reduction in electrical bills
• Every kW saved in a datacenter reduces carbon dioxide emissions by 5 tonnes per year
• The typical 1MW (IT load) datacenter is continuously wasting about 400kW due to poor design (DCiE = 50%, instead of best-practice 70%)
66
White paper
Inefficiencies Create Consumption
• Computing inefficiencies > More servers
• Server inefficiencies > More power and cooling• Power and cooling inefficiencies > More power
consumption
Inefficiencies drive both power consumption Inefficiencies drive both power consumption and material consumptionand material consumption
2008 IDC study
Rising densities have led to an increase in power and cooling issues
Power and cooling:
The new #1 datacenter issue
Causes of Energy Inefficiency in Datacenters
Primary Causes• Over sizing of power and cooling equipment• Pushing cooling systems to cool densities higher than they were designed for• Redundancy (for high availability)• Inefficient power and cooling equipment• Ineffective airflow patterns
Secondary Causes• Mainly operational practices
– ineffective room layout– Inefficient operating settings of cooling equipment– Clogged air or water filters– Disabled or malfunctioning cooling economizer modes– Raised floor clogged with wires
CPU Power Consumption
● Accepting 20% utilization as average the IBM figures say the power dedicated to computation is about 2% of the total data centre power…….
● From Dell’s figures using the same utilization ratio we get 1.6%....
Changes to impact efficiency and power consumption
• At top of pyramid changes will have a minuscule impact on power consumption for the data centre as a whole yet they can have a dramatic impact in the data centre efficiency ie useful computations done.
• Increase CPU utilisation 20% to 60%-80% or more (headroom to ensure more responsive to workload peaks), consolidation/virtualisation
• server refresh can potentially double the output per CPU if the servers are two years old or more than quadruple it if four years old.
• At bottom plain energy saving
EXPECTED LOAD
% C
AP
AC
ITY
YEARS
ROOM CAPACITY
INSTALLED CAPACITY
ACTUAL LOAD
Waste due to over sizingWaste due to over sizing
Reference: APC White paper #37
120%120%
100%100%
80%80%
60%60%
40%40%
20%20%
0%0%
00 11 22 33 44 55 66 77 88 99 1010
Over sizing Datacenter capacity
Convert 2N power and cooling into N+N● Double the IT load from the same power input
• Downside is that some load will need to be dropped in case of power or cooling failure• However – UPS + generator backup on each power feed and cooling can be hardened
too• All critical load is also duplicated in a second DR facility too, so the worst-case impact
is 25%• Business needs to sign off on extra risk and need for preventative maintenance outage
Proposed Dedicated Blade Hall Layout● Assumes 2.1MW @ 2N power (4.2MW @ N), 200KW per compute cell● 10-20 cells, depending on 2N or N+N choice
If “A” or “B” power or cooling completely fails, half of the “N+N”load will be required to shut down. The “2N” load will remain fully operational.
Note that the A/B split is across the centre of a compute cell. This will permit half the cell and half the cooling to remain operational, which gives added HA benefits for grid and clustered usage across each side of the cell.
Tiering-four levels relating to availability and security ANSI/TIA-942-2005
“Is it safe?”
●Would the business not want any additional risk regardless of any potential benefits?
●Recent trends in an increasing power constrained and regulated market are a movefrom Tier 3 or Tier 3+ to Tier 2 and in some cases two Tier1 or three Tier 1 facilities in a delta spaced within the latency limits.
● The modular datacentre (depending on the configuration of its building blocks) offers the opportunity to have multiple tier levels within the same facility and/or use of the lower Tiers to benefit from a better PUEwithout any real loss in availability and security.
The Datacenter Power Chain
PUE
The Datacenter Power Chain
Take the power to the DC or the DC to the power ?
denser with more frequent change, focus on efficiency and cost
●Blade technology adoption as a mainstream architecture
●Consolidation & Virtualisation●Dynamic cooling loads
●Cloud Computing (External / Internal)
●Warehouse-Scale Computer (WSC)● Thin-client architecture
Together with the above, the recent economic turmoil and the ever shrinking business horizon has driven the need for more flexible, scaleable, modular DC facilities
Virtualisation
“Virtualisation is the new operating system for the Datacenter”
Michael Dell IDC ME CIO Conference, Dubai Jan 2010
Minimize the inefficiency of over sizing during virtualization and re-growth
and be prepared for higher densities to come
LoadLoad
CapacityPower/cooling
LoadVirtualized Load
LoadOriginalLoad
Original
LoadVirtualizedLoad
Virtualized
LoadLoadLoad
VirtualizedLoad
Virtualized
LoadVirtualizedLoad
Virtualized
Scale DOWN Scale UPRack density
Scalable infrastructure minimizes waste
Scalable power and cooling results in better PUE by Scalable power and cooling results in better PUE by tracking IT load as it shrinks and growstracking IT load as it shrinks and grows
Raising the efficiency curve
Fixed losses and consolidation
Need to reduce fixed losses
Infrastructure
ITUsefulcomputing
Infrastructure
ITUsefulcomputing
Physical
infrastructure
(DCPI)Watts
IN
Watts
IN
Watts
IN
Watts
IN
Watts
IN
Watts
IN
ITwattsIT
watts
ITwattsIT
wattsIT
wattsIT
watts
=
LOWER is better, 1 is perfectLOWER is better, 1 is perfect
Datacenter Efficiency Metric – Power Usage Effectiveness (PUE)
158
PUE/DCiE-Define basis of calculation●Data center efficiency metrics PUE
and DCiE have yet to develop intoformal detailed standards that define what devices or subsystems should be included in the metric and which should not. Therefore, APC by Schneider Electric has standardized on its own criteria for what devices or subsystems are included andexcluded. The table shows all the common subsystems of a data center and separates them into one of three categories: IT Load, Physical Infrastructure, and Not Included. The physical infrastructure category includes subsystems such as UPS, PDU, generator, etc.
Efficiency Analysis of Proposed Data Centre Concept
●An efficiency model was developed for the proposed data center concept which resulted in an estimated annualized PUE of 1.22 at 100% load or a DCiE of 82% at 100% load . This model provides the entire efficiency curve which allows efficiency estimates at any load level. For comparison, a typical data center in this location at 100% load would have a PUE of 1.76 or a DCiEof 57%. Both of these curves are illustrated in the figure above.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
0.0% 20.0% 40.0% 60.0% 80.0% 100.0%
% IT loadD
CiE
Microsoft concept
Typical data center0.00
1.00
2.00
3.00
4.00
5.00
6.00
0.0% 20.0% 40.0% 60.0% 80.0% 100.0%
% IT load
PU
E
Microsoft concept
Typical data center
Drivers of infrastructure efficiency gains
Baseline: Average of existing installed base
Goal: Reduce industry average PUE from 2.13 to 1.39How to get there: Relative contribution of data center improvements
Cooling ECONOMIZERS
Convert from ROOM COOLING to dynamic ROW/RACK cooling
RIGHT-SIZING via modular power and cooling
HigherUPS EFFICIENCY
415/240 V TRANSFORMERLESS power distribution (NAM)
DYNAMIC CONTROL OF COOLING PLANT(VFD fans, pumps, chillers)
32% contribution.24 PUE reduction
16% contribution.12 PUE reduction16% contribution
.12 PUE reduction
16% contribution.12 PUE reduction
10% contribution.07 PUE reduction
10% contribution.07 PUE reduction
PUE
2.13
1.39
PUE efficiency KPI is the industry standard metric for measuring and benchmarking datacenter energy efficiency
Maximum PUE = Server Virtualisation + Scalable Power and Cooling infrastructure
Scalable Power and Cooling infrastructure and IT virtualization can DOUBLE electrical savings.
Row and Rack cooling architecture addresses increasinghigher IT densities requirements.
Efficiency dashboard
InfraStruXure Central ®
Enterprise portal
Supervision
Supervision
Integration is key to optimise efficiency
PROCESS& MACHINES
BUILDINGCONTROL
WHITE SPACE
POWER
SECURITY Supervision
Energy data access at low cost
Energy managementservices
Datacenter Electrical Efficiency Assessments
• Breakdown power, cooling & lighting losses
• Breakdown cooling losses into CRAC, humidification, outdoor heat losses
• Breakdown power losses into UPS, power distribution
• Calculate DCiE• Detailed recommendations to
improve efficiency• Projected efficiency gains for each
recommended improvement
Software Design Tools
Data Center Operations Management
Security & Environmental•NetBotz•Pelco
Data Center Physical Infrastructure
Energy Management•PowerLogic IONE•PowerLogic Meters•Cisco Energywise•IBM Active Energy Manager
Enterprise Management Systems•Microsoft System Center Operations Manager (SCOM)
•Microsoft System Center Essentials•IBM Tivoli
Building Management Systems•Schneider TAC Product Line, including Andover Conti nuum, Vista
•All major brands
InfraStruXure Central
•Centralized & real-time monitoring•Fault notification & graphical trending•Thresholds & alarm settings•Auto-discovery •Mass configuration•Multi-vendor device support
InfraStruXureOperations
Capacity & Change Manager
Ensure capacity meets demand at row, rack, and server level
• Incremental build out• Standardized components• Targeted density,
efficiency, availability
Modular/standardized/scalable architecture Modular/standardized/scalable architecture reduces both upfront AND operating costsreduces both upfront AND operating costs
““““Step and repeatStep and repeatStep and repeatStep and repeat”””” strategystrategystrategystrategy
“POD” Architecture
Capital Expenditure
"It's a very sobering feeling to be up in space and realize that one's safety factor was determined by the lowest bidder on a government contract."
Alan Shepard (Astronaut 1923-1998)
● A “mini data center”with its own cooling
● Contributes no heat to rest of data center
●Works with existing room-based cooling
● Hot/cool air circulation localized within the pod by short air paths and/or containment
● Achieves optimal efficiency
● Targeted availability
Targeted Zone Architecture
134
White paper
Large Standardised Scalable Facilities
Using the industrial shed
Start with the low hanging fruit first
Beyond the crinkly tin
Prefabricated modular rooms
● Normal DC room heights – 3.8m
● Normal DC aisle widths – 1.35m
● Takes deep racks – 1.2m
● Each room with two 18m rows
● 500kW critical load per room
● N+1 cooling, dual feed power
● Weathered – use externally or internally
● Transportable
Prefabricated substations
Std ISO container with UPS & Chillers
• Utility support for power and cooling
• IT Rack Box
Containers stacked inside – MS Chicago
Modular, scalable using containers
We have been there before
The Modular scaleable DC - tecnikon shown at ECTA Vienna in 2000
• Tecnikon MDC 58RMU air-cooled sealed IP65 fully ducted racks12.5kW per rack
• Pre-assembled fully wired spines (rows – 20 racks per row)
• Plug-in plant modules, stair and lift towers
• Factory built/prefabricated, transported by road, -can be relocated for reuse elsewhere, fast build.
Cooling with external air
Modular approach
Masdar City – Exemplar for ecological design
Summary
Thank YouThank YouThank YouThank You