Technologies Paper

1
BasedonIntelCorporationinternaltesting.TheIntel®Xeon®E5-2680ApacheHadoopclusterconsistedofamasternodeandveworkernodes.AllnodeswereconguredwithdualIntelXeonE5-2680processorsrunningat2.7GHz,128GB
ofRAM(8x16GBDDR31600Mhz),3IntelSSDDCS3700SeriesSSDs(400GB,2.5inSATA6Gb/s,25nm,MLC,Seq.Read:500MB/s,Seq.Write460MB/s),3IntelSSD710SeriesSSDs(200GB,2.5inSATA3Gb/s,25nm,MLC,Seq.Read:
270MB/s,Seq.Write210MB/s),2IntelSSD710SeriesSSDs(300GB,2.5inSATA3Gb/s,25nm,MLC,Seq.Read:270MB/s,Seq.Write210MB/s,anIntelX25-EExtremesolidstatedrivefortheoperatingsystem(SATA2,2.5inch,SLC,High
Performance),CentOSRelease6.2withkernel2.6.32-358.el6.x86_64,ApacheHadoop1.0.4,Intel®82599EB10gigabitEthernetcontroller,andanLSILogicSAS2308PCI-ExpressFusion-MPTSAS-2RAIDcontroller.Theestimatedstreet
priceforeachnodeis$15,000.
TheIntel®Xeon®E7-4890v2ApacheHadoopclusterconsistedofamasternodeandtwoworkernodes.AllnodeswereconguredwithfourIntelXeonE7-4890v2processorsrunningat2.8GHz,512GBofRAM(32x16GBDDR31600MHz),
12IntelSSDDCS3700SeriesSSDs(400GB,2.5inSATA6Gb/s,25nm,MLC,Seq.Read:500MB/s,Seq.Write460MB/s),anIntelX25-EExtreme32GBsolidstatedrivefortheoperatingsystem(SATA2,2.5inch,SLC,HighPerformance),
CentOSRelease6.2withkernel2.6.32-358.el6.x86_64,ApacheHadoop1.0.4,Intel®82599EB10gigabitEthernetcontroller,andanIntelRAIDControllerRS2BL040.Theclusterwithserversconguredwith1.5TBofRAMusedthesame
congurationexceptfor96x16GBDDR31600MHzRAMmodules.Theestimatedstreetpriceforeachnodeconguredwith512GBofRAMis$65,000,and$80,000forthenodeswith1.5TBofRAM.Eachnodeinbothclustersusedthefol-
lowingsettings:BIOS(Intel®TurboBoostenabled,Intel®SpeedStepdisabled),ApacheHadoop(DFSreplicationsetto1,JVMreusesettoenabled),operatingsystem(‘deadlinediskscheduler’,transparentlargepagesdisabled,IPV6disabled,
iptablesdisabled,andtimesynchronizedwithNTP).
2
TheresultsshowninTable1weregeneratedusinganIntelinternaltotalcostofownershiptool(theTool)asofDecember2013,normalizingdatabasetransactionalperformancebetweenthetwooptions.Theseresultsandthescenarios
describedareestimatesandshouldonlybeusedasaguidetoevaluatethecost/benetorfeasibilityofafuturepurchaseofsystemswithIntelXeonProcessors.PriortopurchasinganysystemswithIntelXeonProcessorsyoushouldconsult
withqualiedITprofessionalstoensurethatanysuchpurchasestyourparticularcircumstances.InteldoesnotandcannotguaranteetheaccuracyorreliabilityoftheresultsgeneratedbytheToolorthatyouwillactuallyrealizeanycost
savingsasforecastbytheToolandshowninthetable.
Thisanalysisassumesanaverageofthree(3)timesperformanceincreaseintestedworkloadsbetweenclustersequippedwithtwo-socketserversbasedonIntel®Xeon®processorE5-2680and4-socketserversbasedonIntelXeon
processorE7-4890v2processors.Calculationsincludeanalysisbasedonperformance,power,cooling,electricityrates,rackspaceandnetworkmaintenance.Assumptionsinclude42Uracks,$0.10perkWh,coolingcosts2xaverage
serverpowerconsumptioncosts,Alinean*assumptionsof$500perservermaintenanceand$30perservernetworkingcosts,averagerealestatecostperyearfromVMware*Cost-Per-ApplicationCalculatorat$310persq.foot*10sq.
feetperrackdividedbythenumberofserversperrack,60%CPUutilizationandPUEof2.0.Clustercostsavingsbasedoncomparingtheperformanceofdeploying1,000two-socketserversbasedonIntel®Xeon®processorE5-2680
comparedtoequivalentperformanceoffour-socketserversbasedonIntelXeonprocessorE7-4890v2.FormoreinformationabouttheVMwareCost-Per-ApplicationCalculator,gotohttp://www.vmware.com/les/pdf/vmware-cost-per-
application-calculator-methodology.pdf.
3
AmasternodecanalsohostDataNodeandTaskTracker.
4
AvailableonselectIntel®Core™processors.RequiresanIntel®Hyper-ThreadingTechnology-enabledsystem;consultwithyourPCmanufacturer.Performancewillvarydependingonthespecichardwareandsoftwareused.Formore
information,includingdetailsonwhichprocessorssupportIntelHTTechnology,visithttp://www.intel.com/info/hyperthreading.
5
Nocomputersystemcanprovideabsolutereliability,availabilityorserviceability.RequiresanIntel®Xeon®processorE7-8800/4800/2800v2productfamiliesorIntel®Itanium®9500series-basedsystem(orfollow-ongenerationsofeither.)
Built-inreliabilityfeaturesavailableonselectIntel®processorsmayrequireadditionalsoftware,hardware,servicesand/oraninternetconnection.Resultsmayvarydependinguponconguration.Consultyoursystemmanufacturerformore
details.
6
PCIeLiveErrorRecoveryimplementationsarespecictoanOEM’simplementation.ConsultyourOEMforspecicrecoverycapabilities.
7
BasedonIntelCorporationinternaltesting.TheApacheHadoopclusterconsistedofamasternodeandtwoworkernodes.ThemasternodewasconguredwithdualIntel®Xeon®X5570processorsrunningat2.93GHzandanIntel82574L
gigabitEthernetadapter.TheworkernodeswereconguredwithdualIntelXeonE5-2680processorsrunningat2.7GHz,128gigabytesofDDR3RAM,16hardwarethreads(IntelHyper-Threadingdisabled),andanIntel82598EB10gigabit
Ethernetadapter.ThesoftwarecongurationconsistedofCentOS6.364-bitwithkernel2.6.32-279.19.1.el6.x86_64,JavaHotspot™1.7.0_1364-bitserverVM(build23.7-b01,ApacheHadoop1.0.4withperformancetuningapplied,and
compressionprovidedbySnappy1.1.ThetestsfocusedonI/O-intensivesymmetricworkloads.ThetestsconsistedofaSorttestusing250GBofseeddata,andaTerasorttestwithhighlycompressibledata(1TBofdatacompressedto160GB
withzlib).NotethatApacheHadooprequiresadvancedtuninginordertotakefulladvantageofSSDthroughput.Seehttp://software.intel.com/sites/default/les/Optimizing%20Hadoop%20Deployments.pdf for more details.
8
BasedonIntelCorporationinternaltesting.PerformancecomparisonusinggeometricmeanofSPECint*_rate_base2006,STREAM*_MPTriad,andLinpack*benchmarkresults.Baselinegeometricmeansscoreof166.75onpriorgeneration
2SIntel®Xeon®processorX5690platformbasedonbestpublishedSPECrate*scorestowww.spec.organdbestIntelinternalmeasurementsonSTREAM_MPTriadandLinpackasof5December2011.Newgeometricmeanscoreof306.74
basedonIntelinternalmeasuredestimatesusinganIntel®RoseCityplatformwithtwoIntel®Xeon®processorE5-2690,TurboandEISTEnabled,withHyper-Threading,128GBRAM,RedHatEnterpriseLinux*Server6.1betaforx86_6,
Intel®Compiler12.1,THPdisabledforSPECfp_rate_base2006andenabledforSPECint_rate_base2006.
ResultshavebeenestimatedbasedoninternalIntelanalysisandareprovidedforinformationalpurposesonly.Anydifferenceinsystemhardwareorsoftwaredesignorcongurationmayaffectactualperformance.
InteldoesnotcontrolorauditthedesignorimplementationofthirdpartybenchmarkdataorWebsitesreferencedinthisdocument.IntelencouragesallofitscustomerstovisitthereferencedWebsitesorotherswheresimilarperformance
benchmarkdataarereportedandconrmwhetherthereferencedbenchmarkdataareaccurateandreectperformanceofsystemsavailableforpurchase.
Intel’scompilersmayormaynotoptimizetothesamedegreefornon-IntelmicroprocessorsforoptimizationsthatarenotuniquetoIntelmicroprocessors.Theseoptimizations
includeSSE2,SSE3,andSSE3instructionsetsandotheroptimizations.Inteldoesnotguaranteetheavailability,functionality,oreffectivenessofanyoptimizationon
microprocessorsnotmanufacturedbyIntel.Microprocessor-dependentoptimizationsinthisproductareintendedforusewithIntelmicroprocessors.Certainoptimizationsnot
specictoIntelmicroarchitecturearereservedforIntelmicroprocessors.PleaserefertotheapplicableproductUserandReferenceGuidesformoreinformationregardingthe
specicinstructionsetscoveredbythisnotice.Noticerevision#20110804
SoftwareandworkloadsusedinperformancetestsmayhavebeenoptimizedforperformanceonlyonIntelmicroprocessors.Performancetests,suchasSYSmarkand
MobileMark,aremeasuredusingspeciccomputersystems,components,software,operationsandfunctions.Anychangetoanyofthosefactorsmaycausetheresultstovary.
Youshouldconsultotherinformationandperformanceteststoassistyouinfullyevaluatingyourcontemplatedpurchases,includingtheperformanceofthatproductwhen
combinedwithotherproducts.
INFORMATIONINTHISDOCUMENTISPROVIDEDINCONNECTIONWITHINTELPRODUCTS.NOLICENSE,EXPRESSORIMPLIED,BYESTOPPELOROTHERWISE,
TOANYINTELLECTUALPROPERTYRIGHTSISGRANTEDBYTHISDOCUMENT.EXCEPTASPROVIDEDININTELSTERMSANDCONDITIONSOFSALEFORSUCH
PRODUCTS,INTELASSUMESNOLIABILITYWHATSOEVERANDINTELDISCLAIMSANYEXPRESSORIMPLIEDWARRANTY,RELATINGTOSALEAND/ORUSEOF
INTELPRODUCTSINCLUDINGLIABILITYORWARRANTIESRELATINGTOFITNESSFORAPARTICULARPURPOSE,MERCHANTABILITY,ORINFRINGEMENTOF
ANYPATENT,COPYRIGHTOROTHERINTELLECTUALPROPERTYRIGHT.
Intelprocessornumbersarenotameasureofperformance.Processornumbersdifferentiatefeatureswithineachprocessorfamily,notacrossdifferent
processorfamilies.Seewww.intel.com/products/processor_number for details.
Copyright©2014IntelCorporation.Allrightsreserved.Intel,theIntellogo,andXeonaretrademarksofIntelCorporationintheU.S.and/orothercountries.
*Othernamesandbrandsmaybeclaimedasthepropertyofothers. PrintedinUSA 0114/TA/PRW/PDF PleaseRecycle 330050-001US
Accelerate Big Data Analysis with Intel® Technologies