管理下一代的IT基础架构.docx

管理下一代的IT基础架构.docx

《管理下一代的IT基础架构.docx》由会员分享，可在线阅读，更多相关《管理下一代的IT基础架构.docx（12页珍藏版）》请在冰豆网上搜索。

管理下一代的IT基础架构

Managingnext-generationITinfrastructure

Thedaysofbuildingtoorderareover.Thetimeisripeforanindustrialrevolution.

JamesM.Kaplan,MarkusLöffler,andRogerP.Roberts

TheMcKinseyQuarterly,Webexclusive,February2005

Inrecentyears,companieshaveworkedhardtoreducethecostoftheITinfrastructure—thedatacenters,networks,databases,andsoftwaretoolsthatsupportbusinesses.Theseeffortstoconsolidate,standardize,andstreamlineassets,technologies,andprocesseshavedeliveredmajorsavings.Yeteventhemosteffectivecost-cuttingprogrameventuallyhitsawall:

thecomplexityoftheinfrastructureitself.

Therootcauseofthiscomplexityisthebuild-to-ordermind-settraditionalinmostITorganizations.Thetypicalinfrastructuremayseemtobehightechbutactuallyresemblesanold-fashionedautomobile:

handmadebyanexpertcraftspersonandcustomizedtothespecificationsofanindividualcustomer.Todayanapplicationdevelopertypicallyspecifiestheexactserverconfigurationforeachapplicationandtheinfrastructuregroupfulfillsthatrequest.Theresult:

thousandsofapplicationsilos,eachwithitsowncustom-configuredhardware,andajumbleofoftenincompatibleassetsthatgreatlylimitacompany'sflexibilityandtimetomarket.Sinceeachservermaybeconfiguredtomeetanapplication'speakdemand,whichisrarelyattained,vastamountsofexpensivecapacitysitunusedacrosstheinfrastructureatanygiventime.Moreover,applicationsaretightlylinkedtoindividualserversandstoragedevices,sotheexcesscapacitycan'tbeshared.

Now,however,technologicaladvances—combinedwithnewskillsandmanagementpractices—allowcompaniestoshedthisbuild-to-orderapproach.Adecadeintothechallengingtransitiontodistributedcomputing,infrastructuregroupsaremanagingclient-serverandWeb-centeredarchitectureswithgrowingauthority.Companiesareadoptingstandardizedapplicationplatformsanddevelopmentlanguages.Andtoday'shigh-performanceprocessors,storageunits,andnetworksensurethatinfrastructureelementsrarelyneedhand-tuningtomeettherequirementsofapplications.

Inresponsetothesechanges,someleadingcompaniesarebeginningtoadoptanentirelynewmodelofinfrastructuremanagement—moreoff-the-shelfthanbuild-to-order.Insteadofspecifyingthehardwareandtheconfigurationneededforabusinessapplication（"Ineedthisparticularmaker,model,andconfigurationformynetwork-attachedstoragebox..."）,developersspecifyaservicerequirement（"Ineedstoragewithhigh-speedscalability..."）;ratherthanbuildingsystemstoorder,infrastructuregroupscreateportfoliosof"productized,"reusableservices.Streamlined,automatedprocessesandtechnologiescreatea"factory"thatdeliverstheseproductsinoptimalfashion（Exhibit1）.Asproductordersrollin,afactorymanagermonitorstheinfrastructureforcapacity-planningandsourcingpurposes.

Withthismodel,fillinganITrequirementisratherlikeshoppingbycatalog.Adeveloperwhoneedsastorageproduct,forinstance,choosesfromaportfolioofoptions,eachdescribedbyservicelevel（suchasspeed,capacity,oravailability）andpricedaccordingtotheinfrastructureassetsconsumed（say,$7amonthforagigabyteofmanagedstorage）.Thesystem'stransparencyhelpsbusinessusersunderstandhowdemanddrivestheconsumptionandcostofresources.

Companiesthatmakethetransitiongainbigbusinessbenefits.Byreducingcomplexity,eliminatingredundantactivity,andboostingtheutilizationofassets,theycanmaketheirinfrastructure20to30percentmoreproductive—ontopofthebenefitfrompreviousefficiencyefforts—therebyprovidingfargreateroutputandflexibility.Evenlargersavingscanbeachievedbyusinglow-cost,commodityassetswhenpossible.Developersnolongermustspecifyanapplication'stechnicalunderpinningsandcanthereforefocusonworkthatdeliversgreaterbusinessvalue;thenewmodelimprovestimestomarketfornewapplications.

Nevertheless,makingthistransitioncallsformajororganizationalchanges.Applicationdevelopersmustbecomeadeptatforecastingandmanagingdemandsothat,inturn,infrastructuregroupscanmanagecapacitymoretightly.Infrastructuregroupsmustdevelopnewcapabilitiesinproductmanagementandpricingaswellasintroducenewtechnologiessuchasgridcomputingandvirtualization.1AsforCIOs,theymustputinplaceanewmodelofgovernancetomanagethenewinfrastructureorganization.

Theroadforward

DeutscheTelekomknowsfirsthandthechallengesinvolved:

over18months,hopingtobalanceITsupplyanddemand,itimplementedthisnewinfrastructure-managementmodelattwodivisions（seesidebar,"Next-generationinfrastructureatDeutscheTelekom"）.Intheolddays,thecompany'sITinfrastructure,likemost,wasalandscapeofapplicationsilos.Todayaccurateforecastsofuserdemandarecritical,sonewlymintedproductmanagersmusttakeahorizontalview,acrossapplications,toassessthetotalneedsofthebusinessandcreatetherightproducts.Theymustthenworkcloselywithinfrastructureteamstoalignsupply—infrastructureassetssuchashardware,software,andstorage—withdemand.

Inthepast,employeesoftheinfrastructurefunctionwereordertakers.Now,theycanbemoreentrepreneurial,choosingthemixofhardware,software,andtechnologythatoptimizestheinfrastructure.Tokeepcostslow,theycanphaseingridsoflow-endservers,cheaperstoragedisks,andothercommodityresources.Factorymanagersnowfocusonautomatingand"industrializing"production.AlthoughDeutscheTelekom'stwodivisionsdidn'tradicallychangetheirorganizationalorreportingstructures,ITgovernancenowseekstoensurethatproductandservicelevelsareconsistentacrossbusinessunitsinordertominimizecostsandtoimprovetheinfrastructure'soverallperformance.

Whatwe'veseenatDeutscheTelekomandothercompaniessuggeststhatcreatinganext-generationinfrastructureinvolvesactiononthreefronts:

segmentinguserdemand,developingproductlikeservicesacrossbusinessunits,andcreatingsharedfactoriestostreamlinethedeliveryofIT.

Segmentinguserdemand

LargeITorganizationssupportthousandsofapplications,hundredsofphysicalsites,andtensofthousandsofendusers.Allthreeoftheseelementsarecriticaldriversofinfrastructuredemand:

applicationsrequireserversandstorage,sitesneednetworkconnectivity,anduserswantaccesstodesktops,laptops,PDAs,andsoforth.Tostandardizethesesegments,anITorganizationmustfirstdevelopadeepunderstandingoftheshapeofcurrentdemandforinfrastructureservicesandhowthatdemandwillmostlikelyevolve.Thenitneedstocategorizedemandintosegments（suchasuptime,throughput,andscalability）thataremeaningfultobusinessusers.

Whengroupedinthisway,mostapplicationsfallintoarelativelysmallnumberofclusters.Apharmaceuticalmanufacturer,forinstance,foundthatmostofabusinessunit'sexistingandplannedapplicationsfellintooneoffivecategories,includingsalesforceapplicationsthatneedaround-the-clocksupportandoff-lineavailabilityandenterpriseapplicationsthatmustscaleuptothousandsofusersandhandlebatchtransactionsefficiently.

Incontrast,atypicalwholesalebank'sapplicationportfoliohasmoresegments,withawiderrangeofneeds.Someapplications—suchasderivatives,pricing,andrisk-managementtools—mustexecutecomputation-intensiveanalysesinminutesratherthanhours.Funds-transferapplicationsallowforlittleornodowntime;program-tradingapplicationsmustexecutetransactionsinmillisecondsorriskcompromisingtradingstrategies.

Althoughsimplebycomparison,theneedsofphysicalsitesandusergroupscanbecategorizedinasimilarway.Onemarketing-servicescompanythatevaluateditsnetworkarchitecture,forexample,segmenteditssitesintoofficeswithmorethan100seats,thosewith25to100,andremotebrancheswithfewerthan25.Acablesystemsoperatordivideditsusersintoseniorexecutiveswith"concierge-support"needs,professionalemployees,call-centeragents,andfieldtechnicians.

Mostcompaniesfindthatdefiningthespecificinfrastructureneedsofapplications,sites,andusersisthekeychallengeofsegmentingdemand.Majorissuesincludethetimeandfrequencyofneed,thenumberofusers,theamountofdowntimethatisacceptable,andtheimportanceofspeed,scalability,andmobility.

Standardizingproducts

Oncetheinfrastructuregrouphasassessedcurrentandfuturedemand,itcandevelopasetofproductlike,reusableservicesforthreesegments:

managementandstorageproductsforapplications,accessproductssuchasdesktopsandlaptopsforendusers,andnetwork-accessproductsforvarioussites.Foreachofthesethreeproductlines,thegroupmustthenmakeaseriesofdecisionsatboththeportfolioandtheproductlevel.

Attheportfoliolevel,ithastomakedecisionsaboutthescope,depth,andbreadthofproductofferings,withaneyetowardoptimizingresourcesandminimizingcosts.Exceptionsmustbedetailedupfront.Thegroupmaydecide,forexample,againstofferingproductstosupportapplicationswithstringentrequirements,suchasvery-low-latencyprocessing;theseapplicationsmaybebetterbuilt"byhand"and"fromthegroundup."Otherapplications,suchaslegacyones,maybebetterleftoutsidethenewmodelifthey'rerunningwellandcan'teasilybeportedtonewhardware.Thegroupshouldalsodecidehowtointroducenewtechnologiesandtomigrateexistingapplicationsthatareeasiertomove.

Attheproductlevel,thegroupmustdefinethefeatures,servicelevels,andpriceofeachproduct.Fo