data center strategy for business transformation paper

7/31/2019 Data Center Strategy for Business Transformation Paper

1/8

Intel ITs Data Center Strategy forBusiness Transformation

IT@Intel White PaperIntel ITIT Best PracticesData Centers and IT InnovationDecember 2011

We believe our new approachto data center costing andinvestment evaluation will

stimulate a bolder approach tocontinuous innovation and will

improve the quality, velocity,and efficiency of Intel ITs

business services, creating

a sustained competitiveadvantage for Intels business.

We have realized hundreds of millions ofU.S. dollars in cost savings since 2006 byproactively refreshing our infrastructure,adopting cloud computing, updating ournetwork, pursuing IT sustainability, andconsolidating data centers. In addition,we have supported business growth andcapability improvements by deploying uniquesolutions that benefit Intels critical businessfunctionsDesign, Office, Manufacturing,Enterprise, and Services (DOMES).

Going forward, we have enhanced ourstrategy to include several new elements:

Changing our investment decisionmodel, which now compares current datacenter capabilities to a best achievablemodel. This model seeks to remove theconventional improvement mindset, whichonly focuses on incremental improvements.Instead, the model will help us transform

our capabilities by identifying furthergroundbreaking innovationslike thosealready used to implement our privatecloud and our highly efficient silicon designcomputing grid.

Implementing holistic key performanceindicators and associated goals for costper service unit, quality of service, andeffective resource utilization.

Using a new unit-costing financial modelthat enables us to benchmark ourselvesand prioritize our investments.

We believe our new approach to data centercosting and investment evaluation willstimulate a bolder approach to continuousinnovation and will improve the quality,velocity, and efficiency of Intel ITs businessservices, creating a sustained competitiveadvantage for Intels business.

Shesha Krishnapura Senior Principal Engineer, Intel IT

Vipul LalSenior Principal Engineer, Intel IT

Raju NallapaPrincipal Engineer, Intel IT

Isaac PriestleyStrategic Financial Analyst, Intel IT

John PereiraDirector, Data Centers and Hosting, Intel IT

Sanjay RungtaSenior Principal Engineer, Intel IT

Ananth SankaranarayananTechnical Program Manager, Intel IT

Executive Overview

To better meet Intels business requirements while providing our internal

customers with optimal data centers and innovative business services, Intel

IT is updating our data center strategy. Building on previous investments and

techniques, our refined strategy will continue to create new business value

while improving data center efficiency.


2/8


3/8

www.intel.com/IT 3

Intel ITs Data Center Strategy for Business Transformation IT@Intel White Paper

DESIGNDesign engineers run 20 to 30 millioncompute-intensive batch design jobsevery week. Each job can potentially takeseveral hours to complete. In addition,interactive Design applications are sensitiveto high latencies caused by hosting theseapplications on remote servers. We have usedseveral approaches in our Design computingdata centers to provide enough computecapacity and performance to supportrequirements, including high-performancecomputing (HPC), grid computing, clusteredlocal workstation computing, and aspecialized algorithm that increases theperformance of the heaviest Designworkloads.3 Together, these investmentsenabled Design engineers to run 25 percentmore jobs without adding more computecapacitywhich equates to faster design andtime to market.

Because Design engineers need to accessDesign data frequently and quickly, wedid not simply choose the least expensivestorage method for this environment. Instead,we have invested in clustered and higherperformance network-attached storage

(NAS), along with parallel storagewhich ishighly scalable in performancefor our HPCneeds. We use storage area networks (SANs)for specific storage needs such as databases.

MANUFACTURINGIT systems must be available 24/7 in IntelsManufacturing environment, so we usededicated data centers for factories. We haveinvested heavily over the last few years todevelop a robust business continuity plan

3 Intel uses grid computing for silicon design and tapeoutfunctions. Intels compute grid represents thousandsof interconnected compute servers, accessed throughclustering and job scheduling software. Additionally,Intels tapeout environment uses a high-performancecomputing (HPC) approach, which optimizes all keycomponents such as servers, storage, network, OS,applications, and monitoring capabilities cohesively foroverall performance, reliability, and throughput benefits.For more information on HPC at Intel, refer to High-Performance Computing for Silicon Design, Intel Corp.,November 2009.

that keeps factories running even in thecase of a catastrophic data center failure.These efforts have paid off, and we havenot experienced factory downtime related

to data center facilities since 2009.

In our Manufacturing environment, wepursue a methodical, proven infrastructuredeployment approach to support high reliabilityand rapid implementation. This copy-exactapproach deploys new solutions in a singlefactory first and, once successfully deployed,we copy that implementation across otherfactory environments. This approach reducesthe time needed to upgrade the infrastructurethat supports new process technologies

thereby accelerating time to market for Intelproducts. The copy-exact methodology allowsfor rapid deployment of new platforms andapplications throughout the Manufacturingenvironment, enabling us to meet a 13-weekinfrastructure deployment goal 95 percent ofthe timecompared to less than 50 percentwithout using copy-exact methodology.

OFFICE, ENTERPRISE, AND SERVICESTo improve IT agility and the businessvelocity of our private enterprise cloud,

we have implemented an on-demand self-service model, which has reduced the timeto provision servers from three months tothree hours. We more than tripled the numberof virtualized applications inside the Intel ITOffice and Enterprise environments in 2010,from 12 percent to 42 percent. That numberis greater than 60 percent today, and weremain on track to virtualize 75 percent ofthe applications in our Office and Enterpriseenvironments.

In contrast to the Design environment, in theOffice, Enterprise, and Services environmentswe rely primarily on SAN storage, with limitedNAS storage for file-based data sharing.

REFINING OUR STRATEGYAs the pace of Intels businessaccelerates, we must continually

refine our data center strategy todeliver world-class capabilities in acost-effective manner. Our refinedstrategy includes a new investmentmodel that helps us determine whichinvestments will have the greatestbusiness value, key performanceindicators that help us measure thesuccess of our investments, and anew unit-costing model that helpsus better understand the true costof providing IT services to each

business function.

These new elements of our data centerstrategy, along with a continued focus onmeeting business needs, will help build on thesuccess we have already achieved through ourdata center initiatives over the last decade.These successes include significant data centerconsolidation and dramatic IT cost efficiencies.The refined strategy will enable us to supportthe future growth of Intels customers, products,and acquisitions, as well as enhance the quality,

velocity, and efficiency of the services IT offersto Intel business groups.

Stimulating Bold Innovationthrough a New InvestmentModelBuilding on a time-tested methodology thathas proven successful in Intels Manufacturingenvironment over multiple process technologygenerations, we adopted a new data centerinvestment decision model that comparescurrent data center capabilities to a best

achievable model that guides us to makeinvestments with the highest impact.

Previously, Intel data center planning teamslooked at existing capabilities and funding toestablish a plan of record (POR). This plan droveincremental improvements in our existingcapabilities; our goal was to minimize totalcost of ownership (TCO) and deliver positivereturn on investment (ROI).
http://www.intel.com/IThttp://www.intel.com/IT


4/8

4 www.intel.com/IT

IT@Intel White Paper Intel ITs Data Center Strategy for Business Transformation

In contrast, the new investment model, calledmodel of record (MOR) internally, ignores theconstraints imposed by what we have today.Instead, it identifies the minimum amount of

resources we should ideally have to supportbusiness objectivesthereby establishing anoptimal state with available technology.

By setting a standard of maximum achievableperformance, the new model enables us to: Determine which investments will have

the highest ROI.

Identify the benefits of using disruptiveinfrastructure technologies and breakthroughapproaches that deliver more optimaldata center solutions across all aspects ofour infrastructure.

Make data center location decisions,including identifying potential data centersto consolidate, upgrade, or close.

Using the new model focuses limitedavailable resources in specific areas formaximum holistic gain.

As shown in Figure 2, because technologyis always changing, peak performance alsochangesthe maximum achievable performancekeeps on getting better through innovation.We know that resource constraints make itimpossible to ever actually achieve the standardset by the new investment modelalthoughour HPC environment comes very close to thatgoal. However, the model enables us to identifygaps between where we are and where wedlike to be. We can then identify the biggest gapsin capability to prioritize our budget allocationtoward the highest value investments first.

APPLYING THE NEW INVESTMENT MODEL

We have begun applying our new investmentmodel to identify actionable gaps betweenthe best achievable performance and ourcurrent plan. The model has identified three

Time

K e y

P e r f o r m a n c e

I n d i c a t o r s

Low

igh

Closingthe Gap

Best Achievable Data Center(Model of Record)

Current Capabilities(Plan of Record)

Figure 2. Our new data center investmentmodel encourages innovation and providessignificant business results.

Enhancing Design JobThroughput throughSoftware Optimization

We developed a system softwarecapability called NUMA-Booster,which performs more optimalscheduling of our workloadscompared to default OS schedulingon the latest multi-core non-uniformmemory access (NUMA) systems.Using the NUMA-Booster algorithmincreased job throughput of ourmost critical Design workloads by17 percent and generated USD20.4 million net present value (NPV)in savings. In addition, the algorithmearned a ranking in IDGs InfoWorldGreen 15 Awards.

areas of investment with potential for highROI for one or more business functions: Further data center consolidation.

Reducing unit-cost for the Design environment.

Extending our use of blade servers in theOffice, Enterprise, and Services environments.

Data Center ConsolidationWe used our new investment model to lookclosely at how many data centers we havecurrently and how many we should have ideally.

The new investment model identifiedopportunities to reduce Intels data centers byas much as 35 percent. Since we identified thisgap, we are developing a plan with positive ROIto close the gap through efforts such as:

Closing, retrofitting, or reclassifying datacenters and improving inefficiencies.

Co-locating local infrastructure withDesign and Manufacturing data centers,or providing services from a server closet.

Managing local infrastructure sitesremotely.

Improving facility power efficiency throughstrategic investments.

Reduced Unit-Cost forthe Design EnvironmentOur investment model has shown that wecan improve Design unit cost by 29 percentthrough continuing investments in four-year server refresh, storage optimization,and deployment of a customized softwarealgorithm to optimize grid performance. Weare basing this projection on the 17-percentdesign job throughput improvement enabledby the NUMA-Booster algorithm (see sidebar),

newer servers that offer more meaningfulindicator of performance per system (MIPS)for the same cost compared to the prior year,storage optimizations that have reducedcosts by USD 1.8 million, and improvementsin utilization.


5/8

www.intel.com/IT 5

Intel ITs Data Center Strategy for Business Transformation IT@Intel White Paper

Reduced TCO fromBlade Server TechnologyOur new investment model has shown usthat moving from rack-mount servers toblades can reduce TCO in our cloud computingenvironment by about 29 percent throughreduced port, network, and cable costs. Forexample, a group of 16 blade servers compared

to 16 rack-mount servers requires only eightEthernet interfaces instead of 128, and onlyfour Fibre Channel interfaces instead of 32.

Based on this data, we are activelydeploying blade servers to support furthervirtualization efforts in Office, Enterprise,and Services environments.

Defining Key PerformanceIndicators and GoalsThe second major change we have made to

our data center strategy is to define threenew key performance indicators (KPIs) thatenable us to measure the effectiveness ofdata center investments: cost per service unit,quality of service, and effective utilization.Because the service output for each businessfunction is different, we do not evaluate allbusiness functions in the same way. Our datacenter investment decisions seek to balanceand meet all business requirements whileoptimizing the KPIs.

COST PER SERVICE UNITAs shown in Table 1, different businessfunctions have a different service unit thatwe can measure. This unit represents thecapacity we enable for our business users.

Our goal for this KPI is to achieve a 10-percentimprovement in data center cost efficiencyevery year. This goal does not necessarilymean we will spend less each year, but ratherthat we will get more for each dollar we spend.For example, we may spend less for the samenumber of service units, or we may spend thesame amount but get more service output.

QUALITY OF SERVICEWe use a tiered approach to service-levelagreements (SLAs), tailored to each businessfunctions sensitivity to performance, uptime,mean time to repair (MTTR), and cost. Our goal

for this KPI is to meet specific performance-to-SLA requirements for defined tiering levels.For example, for our most mission-criticalapplications, we aim at a higher performance-to-SLA than for second-tier applications,which are less critical. The end goal andtrue measure of IT quality of service is zerobusiness impacts from IT.

Table 1. Service Unit for Each Business Function

BusinessFunction

Service Unit

Design Cost per electronic designautomation-meaningfulindicator of performance persystem (EDA-MIPS)

Office,Enterprise, and

Services

Cost per OS instance

Manufacturing Cost per integrated factorycompute environment

Intel IT Data Center DashboardTo better monitor and manage our worldwide network of data centers, we developed anddeployed an integrated business intelligence (BI) dashboard. This BI tool is modeled on a dashboard

used in Intels Manufacturing environment. By polling more than 192 million data records acrossour worldwide data center environment, the Intel IT Data Center Dashboard provides a singleview of all our data center health metricsat both system and component levels. It is capableof delivering on-demand reports to users with a mere five-second page load.

This dashboard will help us monitor our key performance indicators (KPIs) by highlighting thecurrent state and opportunities for optimization, thereby enabling overall improvements thatalign with our data center strategy goals.

For example, the dashboard can report on effective utilization of several data center resources,including electronic design automation - meaningful indicator of performance per system (EDA-MIPS);raw and utilized storage capacity; and facilities space, power, and cooling. This data can reportstatistics by business function or by data center, and can be used to compare KPIs and metrics

across several data centers. The figure to the right shows a sample of the dashboard.

The Intel IT Data Center Dashboardprovides a holistic view of data centerresources to help us track our keyperformance indicators (KPIs) and identifyopportunities for optimization andimprovement.

100%

75%

50%

82.1%

25%

0%

Effective Utilized MIPS - Design


6/8

6 www.intel.com/IT

IT@Intel White Paper Intel ITs Data Center Strategy for Business Transformation

EFFECTIVE RESOURCE UTILIZATIONOur refined data center strategy representsa dramatic shift in how we view resourceutilization. Historically, we measured utilization

of IT assetscompute, storage, network, andfacilitiesby simply determining how busy orloaded an asset was. For example, if a serverwas working at peak capacity 90 percent ofthe time, we considered it 90-percent utilized.If 80 percent of available storage was allocated,we considered that 80-percent utilization.

In contrast, we now focus on the actual outputof an assetthat is, effective utilization. Forexample, if Intels design engineers start onemillion design jobsthereby keeping the

servers very busybut a third of those jobsterminate before completion because therewasnt enough storage available, that is loweffective utilization of compute capacityonly 66 percent. Or, if a customer consumesonly 4 GB of a 10-GB storage allocation, theremaining 6 GB is essentially wasted storageeven though it is allocatedand does notrepresent effective utilization of this asset. Ourgoal for the effective utilization KPI is to achieve80-percent effective utilization of all IT assets.

Implementing a NewUnit-Cost Financial ModelWe evolved our financial model from project-and component-based accounting to a moreholistic unit-costing model. For example, wepreviously used a break/fix approach to datacenter retrofits. We would upgrade a datacenter facility or a portion of the facility inisolation, looking only at the project costs andthe expected ROI of that investment, with noholistic view as to the impact of service unitoutput. In contrast, today we focus on TCO perservice unitusing the entire data center coststack per unit of service delivered. This coststack includes all cost elements associatedwith delivering business services and nowconsiders the worldwide view of all datacenters in the assessment of our investments.

As shown in Figure 3, there are six majorcategories of cost to consider: network, head-count, facilities, servers, OS and management,and storage and backup and recovery (BaR). By

adding these costs and then dividing by thetotal number of appropriate service units for theenvironment, we arrive at a cost per service unit.

Service-based unit costing enables us tobenchmark ourselves and prioritize datacenter investments. Determining service-based unit costs also allows us to measureand compare the performance of individualdata centers to each other, identifying whichare underperforminggiving us the tools tomake decisions on whether to upgrade or

consolidate underperforming data centers.To show how the new unit-based costingmodel works, Figure 4 compares Design costdata and Office, Enterprise, and Services costdata. The headcount category accounts fora greater percentage of total cost in Office,Enterprise, and Services than it does in Design;in contrast, servers are more of a cost factorin Design than they are in Office, Enterprise,and Services. Knowing our exact unit cost ineach environment, as well as the breakdownof that cost, enables us to develop optimizedsolutions for each environment that will havethe greatest effect on cost efficiency and ROI.

DATA CENTERBEST PRACTICESWe have made many strategicinvestments and developed solutionsto make our data centers moreefficient and better serve the needs

of Intels business.

These investments are spread across ourentire infrastructure stackcompute, storage,networking, and facilities. These best practices,and the business value they have generated,are described in Table 2.

Network

FacilitiesServers

OS andManagementHeadcount^

Storageand BaR

+

+

++

+

TotalData Center Cost

=

Total DOMES-SpecicService Units

Cost per Service Unit

^Headcount includes Permanent and ConsultingBaR: Backup-and-Recovery

Figure 3. We arrive at a data center unit

cost by considering all categories of costand dividing by the number of units forthat environment, such as electronicdesign automation-meaningful indicatorof performance per system (EDA-MIPS)in Design and OS instances in Office,Enterprise, and Services.

25%

35%

17%

7%

Design Ofce, Enterprise,and Services

35%

21%

7%7%

16%

14%14% 2%

2010 Cost ofInfrastructure as a Service

for Design Compared to

Ofce, Enterprise, and Services

Headcount (Permanent and Consulting)FacilitiesServersOS and ManagementStorage and Backup-and-Recovery

Network

Figure 4. Knowing the total unit cost, as wellthe individual cost category figures for eachbusiness environment, enables us to betterchoose IT investments that will have thegreatest impact on lowering costs.


7/8


8/8

This paper is for informational purposes only. THIS DOCUMENT IS PROVIDED AS IS WITH NO WARRANTIES WHATSOEVER, INCLUDING ANY WARRANTY OFMERCHANTABILITY, NONINFRINGEMENT, FITNESS FOR ANY PARTICULAR PURPOSE, OR ANY WARRANTY OTHERWISE ARISING OUT OF ANY PROPOSAL,SPECIFICATION OR SAMPLE. Intel disclaims all liability, including liability for infringement of any patent, copyright, or other intellectual property rights, relating to use of information in this specification. No license, express or implied, by estoppel or otherwise, to any intellectual property rights is granted herein.Intel, the Intel logo, and Xeon are trademarks of Intel Corporation in the U.S. and other countries.* Other names and brands may be claimed as the property of others. Copyright 2011 Intel Corporation. All rights reserved.Printed in USA Please Recycle 1211/ABC/KC/PDF 325588-001US

For more information on Intel IT best practices,visit www.intel.com/it .

CONCLUSIONTo provide a foundation for continuousinnovation that will improve the quality,velocity, and efficiency of Intel ITsbusiness services, we have refined ourdata center strategy, building on thepractices established over the last decade.

New elements of our data center strategy include:

Stimulating bolder innovation bychanging our investment model.Comparing our current capabilities to abest achievable model encourages us tostrive for innovation that will transformour infrastructure at a faster rate than ifwe only sought incremental change.

A focus on three primary KPIs. Our goalsinclude achieving an annual 10-percentdecrease in cost per service unit, meetingbusiness-specific performance-to-SLAgoals at defined tiering levels, and striving

to achieve an 80-percent effectiveutilization of our data center assets.

New unit-costing financial model. Thismodel enables us to better assess ourdata center TCO based on the businesscapabilities our infrastructure is supporting.The model measures the cost of a unit ofservice output and enables us to compareinvestments and make informed trade-offdecisions across business functionsthereby maximizing ROI and business value.

Figure 5 summarizes our refined data centerstrategy, which has the overall goal of operatinga world-class data center infrastructure to delivera competitive advantage for Intels business.

Combining the new elements of our data centerstrategy with the best practices of the past willenable us to continue to build world-class datacenters to support the growing and changingneeds Intels business in a cost-effective manner.

CONTRIBUTORSDavid AiresBradley EllisonBill Guyon

ACRONYMSBaR backup and recoveryBI business intelligenceDOMES Design, Office, Manufacturing,

Enterprise, and ServicesEDA-MIPSelectronic design automation-

meaningful indicator ofperformance per system

HPC high-performance computingKPI key performance indicatorMIPS meaningful indicator of

performance per systemMOR model of recordMTTR mean time to repairNAS network-attached storageNIC network interface cardNUMA non-uniform memory accessNPV net present valuePOR plan of recordROI return on investmentSAN storage area network

SLA service-level agreementTCO total cost of ownershipUSD U.S. dollar

SCOPEOptimize infrastructure to supportcritical business functions (DOMES)

KEY PERFORMANCE INDICATORSMaximize business value

through three core metrics

APPROACHSeek transformation instead of

incremental change

Cost perService Unit

Service Quality ResourceUtilization

Design

Ofce andEnterprise

Manufacturing

Services

OS ApplicationsCompute

Storage

Network

Facilities

Optimized For:

DOMES:Design, Ofce, Manufacturing, Enterprise, and Services

Best Achievable Data Center(Model of Record)

Current Capabilities(Plan of Record)

Figure 5. Maximizing the business value of Intels data center infrastructure requires continued business-driven innovation in the areas of compute,storage, network, and facilities; the adoption of new metrics; and a dramatic shift in investment strategy.

data center strategy for business transformation paper

Documents