bbbbbbbbbbbbbbbbbbbbbbbbbb

SPECpowerCommitteeServerEfficiencyRatingTool(SERT)TMDraftDesignDocumentStandardPerformanceEvaluationCorporationSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page2of17Copyright2006-2010SPECTableofContents1Introduction41.
1Summary.
41.
2AboutSPEC.
41.
2.
1SPECMembership41.
3TheEPA'sEnergyStarforComputerServerSpecificationandSPEC.
41.
4SPEC'sGeneralDevelopmentGuidelines55666777788888991010111111111112121212121313142.
11.
3Logfiles.
14141515151515151.
4.
1DifferencesfromConventionalBenchmarks1.
5Scope.
1.
6OverviewSummary.
1.
7Scope/Scheduletradeoffs.
2ServerEfficiencyRatingTool(SERT)2.
1Overview2.
2SPECPTDaemon2.
3EnvironmentalConditions.
2.
4TemperatureSensorSpecifications.
2.
5PowerAnalyzerSetup2.
5.
1PowerAnalyzersandTemperatureSensors.
2.
6RunTime.
2.
7TestHarness.
2.
7.
1GraphicalUserInterface.
2.
8Workload.
2.
8.
1Worklets.
2.
8.
2WorkletExamples.
2.
8.
3Workloadterms.
2.
8.
4WorkletExecution.
2.
8.
5LoadLevels.
2.
8.
6Scaling2.
9ServerOptionsandExpansioncapabilities2.
9.
1Redundancy.
2.
9.
2IOComponent.
2.
9.
3DiskIO2.
9.
4NetworkIO.
2.
10Metric/Score.
2.
11ReportingandOutputFiles.
2.
11.
1Report1:"SummaryReport"2.
11.
2Report2:"PowerandPerformanceDataSheet"2.
12TestSoftware.
2.
13Validation/Verification3Logistic3.
1Ownership.
3.
2Support.
3.
3Maintenance.
3.
4Updates.
SPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page3of17Copyright2006-2010SPEC3.
5Licenseeandprice.
1516161616161616171717173.
6Delivery3.
7SubmissionReviewProcess.
3.
8Trademark.
4SERTandEPAEnergyStarforServerVersion2.
0.
4.
1TestedasShipped4.
2FairUseofSERTinformation.
4.
2.
1FairUseRules4.
3Accredited,Independentlaboratory.
4.
4SupplyVoltagetolerance.
4.
5DCLine-Voltage.
4.
6AirflowSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page4of17Copyright2006-2010SPEC1Introduction11.
1Summary2TheEPA'sEnergyStarDevelopmentTeamiscurrentlyworkingonVersion2.
0oftheirComputerServerSpecification345678910111314151617181920212223242526272829303132333435363738404142431.
Version2.
0aimstoevolvetheprogrambyaddingameanstomeasuretheoverallefficiencyoftheserverwhileitisperformingactualcomputingworkviaanActiveModeEfficiencyRatingTool.
TheSPECpowercommitteeiscurrentlyworkingonthedesign,implementationanddeliveryoftheServerEfficiencyRatingTool(SERT)TMnextgenerationtoolsetthatwillmeasureandevaluatetheenergyefficiencyofcomputerservers.
ThisfirstpublicdraftoutlinesthedesignofSERTforreviewbyEPAstakeholdersandtheirassociates.
1.
2AboutSPEC12TheStandardPerformanceEvaluationCorporation(SPEC)wasformedbycomputerindustryparticipantsin1988toestablishindustrystandardsformeasuringcomputeperformance.
SPEChassincebecomethelargestandmostinfluentialbenchmarkconsortium.
ItsmissionistoensurethatthemarketplacehasafairandusefulsetofmetricstoanalyzethenewestgenerationofITEquipment.
TheSPECcommunityhasdevelopedmorethan30industry-standardbenchmarksforsystemperformanceevaluationinavarietyofapplicationareasandprovidedthousandsofbenchmarklicensestocompanies,resourcecenters,andeducationalinstitutionsglobally.
Organizationsusingthesebenchmarkshavepublishedmorethan20,000peer-reviewedperformancereports.
SPEChasalonghistoryofdesigning,developing,andreleasingindustry-standardcomputersystemperformancebenchmarksinarangeofindustrysegments,pluspeer-reviewingtheresultsofbenchmarkruns.
Performancebenchmarkingandthenecessaryworktodevelopandreleasenewbenchmarkscanleadtodisagreementsamongparticipants.
Therefore,SPEChasdevelopedanoperatingphilosophyandrangeofnormativebehaviorsthatencouragecooperationandfairnessamongstdiverseandcompetitiveorganizations.
Theincreasingdemandforenergy-efficientITEquipmenthasresultedintheneedforpowerandperformancebenchmarks.
Inresponse,theSPECcommunityestablishedSPECpower,aninitiativetoaugmentexistingindustrystandardbenchmarkswithapower/energymeasurement.
Leadingengineersandscientistsinthefieldsofbenchmarkdevelopmentandenergyefficiencymadeacommitmenttotacklethistask.
Thedevelopmentofthefirstindustry-standardbenchmarkthatmeasuresthepowerandperformancecharacteristicsofserver-classcomputeequipmentstartedonJanuary26th2006.
InDecemberof2007,SPECpower_ssj2008wasreleased,whichexercisestheCPUs,caches,memoryhierarchyandthescalabilityofsharedmemoryprocessorsonmultipleload-levels.
Thebenchmarkrunsonawidevarietyofoperatingsystemsandhardwarearchitectures.
Inversion1.
10,whichwasreleasedonApril15th2009,SPECaugmentedSPECpower_ssj2008withmulti-nodesupport(e.
g.
,blade-support).
1.
2.
1SPECMembership39SPECmembershipisopentoanyinterestedcompanyorentity.
OAGmembersandassociatesareentitledtolicensedcopiesofallreleasedOSGbenchmarksandunlimitedpublicationofresultsonSPEC'spublicwebsite.
Aninitiationfeeandannualfeesaredueformembers.
Nonprofitorganizationsandeducationalinstitutionshaveareducedannualfeestructure.
Furtherdetailsonmembershipinformationcanbefoundonhttp://www.
spec.
org/osg/joining.
htmlorrequestedat44info@spec.
org.
AlsoacurrentlistofSPECmemberscanbefoundhere:45http://www.
spec.
org/spec/membership.
html.
464749501.
3TheEPA'sEnergyStarforComputerServerSpecificationandSPEC48SPECapplaudstheEPAforitsgoaltodrivetowardgreaterenergyefficiencyinITEquipment,andSPECconsiderstheEPAEnergyStarProgramanindustrypartnerinthiseffort.
Thedevelopmentof1USEnvironmentalProtectionAgency–EnergyStarProgramRequirementsforComputerServershttp://www.
energystar.
gov/index.
cfmc=revisions.
computer_serversSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page5of17Copyright2006-2010SPECanActiveModeEfficiencyRatingToolisanessentialcomponentintheongoingefforttoreduceworldwideenergyconsumptionandpavesthewayforasuccessfulEnergyStarforComputerServersprogramthathasthepotentialtoharmonizeenergyefficiencyprogramsworldwide.
5152535455565759606162636465666768697071727475767778798081828384858687888990919293949596SPECwelcomesthisopportunitytoworkwiththeEPAontheServerEfficiencyRatingTool(SERT)TMinsupportoftheEnergyStarSpecificationforComputerServerandisproudlylookingforwardtocontinuingourlong-standingassociationwiththeEPAEnergyStarDevelopmentTeam.
1.
4SPEC'sGeneralDevelopmentGuidelines58SPEC'sphilosophyandstandardsofparticipationarethebasisforthedevelopmentofSERT.
Thetoolisbeingdevelopedcooperativelybyacommitteerepresentingdiverseandcompetitivecompanies.
Thefollowingguidesthecommitteeinthedevelopmentofatoolthatwillbeusefulandwidelyadoptedbytheindustry:Decisionsarereachedbyconsensus.
Motionsrequireaqualifiedmajoritytocarry.
Decisionsarebasedonreality.
Experimentalresultscarrymoreweightthanopinions.
Dataanddemonstrationoverruleassertion.
Fairbenchmarksallowcompetitionamongallindustryparticipantsinatransparentmarket.
Toolsandbenchmarksshouldbearchitecture-neutralandportable.
Allwhoarewillingtocontributemayparticipate.
Wideavailabilityofresultsontherangeofavailablesolutionsallowsmarketstodeterminewinnersandlosers.
SimilarguidelineshaveresultedinthesuccessandwideuseofSPECbenchmarksintheperformanceandpower/performanceindustryandareessentialtothesuccessofSERT.
1.
4.
1DifferencesfromConventionalBenchmarks73Performancebenchmarksandenergyefficiencybenchmarkstendtofocusoncapabilitiesofcomputerserversinspecificbusinessmodelsorapplicationareas.
SERTisfocusedonprovidingafirstorderofapproximation2ofenergyefficiencyacrossabroadrangeofapplicationenvironments.
Theabsolutescoreislessrelevantfortheenduser,becauseitwillnotreflectspecificapplicationcapabilities.
Aratingtoolthatprovidesapass-failora[Level1/Level2/Level3]pass-failratingisabetterfitforEPA'sEnergyStarEnvironmentforComputerServersthanatypicalbenchmarkresultwithmultipledigitsofprecisioninthemetric.
Marketingoftheabsolutescoreswillbedisallowed,inordertoencouragemoreparticipationintheprogramBenchmarkstendtofocusonoptimalconditions,includingtuningoptionstocustomizetheconfigurationandsoftwaretotheapplicationofthebenchmarkbusinessmodel.
Theneedtoachievecompetitivebenchmarkresultsoftencausessignificantinvestmentinthebenchmarkprocess.
SERTisdesignedtobemoreeconomicalandeasiertouse,requiringminimalequipmentandskillsthough:HighlyautomatedprocessesandleveragingexistingSPECmethodsFocusonas-shippeddefaultsettingsfortheserverFreefromsuper-tuningSERTdevelopmentisplannedtoincludeinputfromabroadspectrumofindustryexpertsusingpartnershipswithEPAIndustrystakeholderstoprovideongoingfeed-backduringthedevelopmentprocessCollaborateonworkload,metricandallotherrequirementsoftheEPA'sVersion2.
0Framework2AndrewFanara,EvanHaines,ArthurHowardhttp://www.
energystar.
gov/ia/partners/prod_development/downloads/State_of_Energy_and_Performance_Benchmarking_for_Enterprise_Servers_Final.
pdfSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page6of17Copyright2006-2010SPECThefeedbackaboutSERT'sdesignwillbecollected,sorted,prioritizedandpassedontoSPECDevelopmentteamviaEPA'sassociates(979899100101102103104106107108109110111112113114115116117118119120121122123124125127128servers@energystar.
gov)Whereabenchmarkrepresentsafixedreferencepoint,EnergyStarprogramsaredesignedtofostercontinuousimprovement,withthresholdsforsuccessrisingastheindustryprogresses.
SERTwillbedesignedtomatchthisparadigm,including:QuickadoptionofnewcomputingtechnologiesRapidturn-aroundforSERTversionupdates1.
5Scope105ThecurrentscopeofVersion2.
0EnergyStarforComputerServersincludesserverswith1-4processorsocketswithastatedgoaltoexpandtoincludebladetechnologiesofsimilarscope.
AdesigngoalofSERTistoaccommodatetheseandlargertechnologies.
However,scheduleconstraintsandthecomplexityoftestingandvalidatingthetoolforlargerconfigurationsmayprecludetheirinclusioninanEnergyStarprogram.
Amongtheissuesinvolvedwithsupportoflargersystemsaretheoverallcapacityofthesystemtocompletework,theabilitytodesignaworkloadthatscaleswiththeinclusionofadditionalprocessors,memory,networkinterfacecards,diskdrives,etc.
Differentworkloadcharacteristicsarerequiredtodemonstrateeffectivenessforeachofthesecomponents.
Providingaworkloadthatfairlyrepresentstheirpresencewhilenotunfairlyrepresentingtheirabsenceisachallenge.
Theseissuesaremoreprevalentwithlargersystemsthathavemoreexpansioncapabilitiesthansmallerservers.
FortheseareaswheretheEPAandtheirsupportingpartnersconcludethatthetooldoesnotadequatelyrepresentthevalueofacomponentcomparedtoitspowerrequirements,thetoolwillbedesignedtoaccommodatetheinclusionof"configurationpower/performanceadders".
AdesigngoalistoautomaticallyincludethisadditionalinformationinthecomputationoftheEnergyStarqualificationresults,includingdetaileddocumentationthatthiswasdone.
Notethatthesewillonlybeincludedif,throughthedatacollectionprocesstheEPAdeterminesthat"configurationadders"arenecessary.
Itisalsopossiblethattheworkloadmixthatisdefinedforsmallersystemswillnotscalewellwhenexamininglargersystems.
1.
6OverviewSummary126ThefollowingtablesummarizessomeofthedesigngoalsthatSERTwillandwillnotprovide.
ISISNOTRatingToolforoverallenergyefficiencyBenchmarkMeasuringtoolforpower,performanceandinlet-temperatureMeasuringtoolforAirflow,Airpressure,outlet-temperatureGeneralcompute-environmentmeasureSpecificapplicationbenchmarkmeasureSupportofAC-poweredserversSupportofDC-poweredservers(seealso4.
5)UsedinsingleOSinstanceperserverenvironmentsUsedinVirtualizationenvironmentEnergyStarRatingToolMarketingTool1291311321331341351361371.
7Scope/Scheduletradeoffs130SPECbenchmarksaredevelopedwiththegoaltogenerateresultswhicharedirectlycomparableformultiplehardwareandsoftwarearchitecturestotheextentthisispossible.
ThesamebasicgoaldirectsthedesignofSERTasspecifiedinthisdocument.
EventhoughSERTisdesignedwiththegoalofbeingarchitectureagnostic,codeneedstobeimplementedforeachoftheworkloadsandthetoolharnessonallsupportedarchitectures.
Furthermorethiscodemustbetestedintensivelyonallarchitecturesinordertoensureafunctionallyequivalentsetofbinaries,whichgeneratefairandcomparableresults.
SimplyusingaportableSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page7of17Copyright2006-2010SPECprogramminglanguagewillnotbesufficienttoachievethesegoals.
Consequentlysignificantcomplexityisaddedtothedevelopmentprocess.
138139140141142143144145146147148149150151152153154155156157160161162163164165167168169170171172173174175176177178179180182183184185186187188GiventhatSERTisdesignedasafirstorderapproximationratingtool,comparabilitymaybehandleddifferentlythanwithbenchmarks(secondorderapproximationtools)whichareusedforcompetitivemarketing.
Neverthelessit'sessentialtoensureaminimallevelofcomparability.
TheresourcesavailableintheSPECpowercommitteearelimitedandatimelydevelopmentofthetoolforasinglearchitecturewillbechallenging.
Supportforadditionalarchitectureswillremoveresourcesfromthedevelopmentofthebasictestroutinesbecausetheywillbeneededforportingthecode.
Furthermoreadditivetestingeffortisrequirednotonlyforthenewarchitecturesbutfortheoriginalimplementationaswellinordertoensurecomparability.
Thereforeeachextraarchitecturewilladdacurrentlyundeterminedamountoftimetotheschedule.
Theresourceandscheduleproblemsrecurwiththesupportofmultipleoperatingsystems.
SERTwillbeinitiallyimplementedonselectedOperatingSystems(OS)perHWarchitecture.
IfsupportforadditionalarchitecturesorOSsisdesired,thenactiveparticipationfromrequestingentitiesismandatory.
CompaniesdedicatingadditionalresourcestotheSPECpowercommitteefordevelopmentofSERTwouldrelaxthescheduleconstraints.
Thecombinationofseveralarchitecturesandvariousoperatingsystemswouldresultinasignificantincreaseoftestingeffortsandpostponethereleasedateconsiderably.
Anestimatedschedulecanbecreatedoncewehavedecidedonalldesigndetails.
2ServerEfficiencyRatingTool(SERT)1582.
1Overview159SERTiscomposedofmultiplecomponentsincludingatestharness,workloads,andasystemundertest(SUT).
Thetestharnesshandlesthelogisticalsideofmeasuringandrecordingpowerdataandincludesthemeasurementhardware,controlsystem,andthecontrollingsoftwareinstalledontheSUTandcontrollersystem.
TheworkloadsexercisetheSUTwhilethetestharnesscollectsthepowerdata.
2.
2SPECPTDaemon166SPEC'spower/temperaturedaemon(alsoknownasPTDaemon,PTDorptd)isusedbySERTtooffloadtheworkofcontrollingapoweranalyzerortemperaturesensorduringmeasurementintervalstoasystemotherthantheSUT.
IthidesthedetailsofdifferentpoweranalyzerinterfaceprotocolsandbehaviorsfromtheSERTsoftware,presentingacommonTCP-IP-basedinterfacethatcanbereadilyintegratedintodifferentbenchmarkharnesses.
TheSERTharnessconnectstoPTDaemonbyopeningaTCPportandusingthesimplecommandsdetailedintheAPIsectionofthisdocument.
Forlargerconfigurations,multipleIP/portcombinationscanbeusedtocontrolmultipledevices.
PTDaemoncanconnecttomultipleanalyzerandsensortypes,viaprotocolsandinterfacesspecifictoeachdevicetype.
Thedevicetypeisspecifiedbyaparameterpassedlocallyonthecommandlineoninitialinvocationofthedaemon.
ThecommunicationprotocolbetweentheSUTandPTDaemondoesnotchangeregardlessofdevicetype.
ThisallowsSERTtobedevelopedindependentlyofthedevicetypestobesupported.
2.
3EnvironmentalConditions181Powermeasurementsneedtobetakeninanenvironmentrepresentativeofthemajorityofusageenvironments.
Theintentistodiscourageextremeenvironmentsthatmayartificiallyimpactpowerconsumptionorperformanceoftheserver.
Thefollowingenvironmentalconditionsneedtobemet:Ambienttemperaturerange:20°CoraboveElevation:withindocumentedoperatingspecificationofSUTHumidity:withindocumentedoperatingspecificationofSUTSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page8of17Copyright2006-2010SPEC2.
4TemperatureSensorSpecifications189Temperaturemustbemeasurednomorethan50mminfrontof(upwindof)themainairflowinletoftheSUT.
Toensurecomparabilityandrepeatabilityoftemperaturemeasurements,SPECrequiresthefollowingattributesforthetemperaturemeasurementdeviceusedduringthebenchmark:190191192193194195196197198200201202203205Logging-Thesensormusthaveaninterfacethatallowsitsmeasurementstobereadbythebenchmarkharness.
Thereadingratesupportedbythesensormustbeatleast4samplesperminute.
Accuracy-Measurementsmustbereportedbythesensorwithanoverallaccuracyof+/-0.
5degreesCelsiusorbetterfortherangesmeasuredduringthebenchmarkrun.
2.
5PowerAnalyzerSetup199ThepoweranalyzermustbelocatedbetweentheACLineVoltageSourceandtheSUT.
NootheractivecomponentsareallowedbetweentheACLineVoltageSourceandtheSUT.
PoweranalyzerconfigurationsettingsthataresetbytheSPECPTDaemonmustnotbemanuallyoverridden.
2.
5.
1PowerAnalyzersandTemperatureSensors204SERTwillutilizeSPEC'sacceptedmeasurementdeviceslistandSPECPTDaemonupdateprocess.
SeeDeviceList(http://www.
spec.
org/power_ssj2008/docs/device-list.
html)foralistofcurrentlysupported(bytheSPECPTDaemon)andcompliant(inspecifications)poweranalyzersandtemperaturesensors.
2062072082092112122142152162.
6RunTime210Theruntimewilldependontheagreedsetofworklets.
Thetargetruntimeislessthan2hours.
2.
7TestHarness213SERTwillrequireatestharnessconfigurationverysimilartothatofSPECpower_ssj2008,withacontrollersystem,poweranalyzer,temperaturesensor,andwiththelikelyadditionofanetworkedclientsystemtogeneratedifferenttypesofworkloadtrafficfortheserverundertest(SUT).
NetworkSwitchTemperatureSensorStorageCPUCPUPSUPSUWorkloadNetworkMemoryServerunderTestCommunicationviaTCP/IPACPowerInCommunicationviaRS-232orUSBPowerAnalyzerControllerSERTControlSWLocalTCP/IPsocketcommunicationSPECPTDaemonSPECPTDaemonNetworkSwitchNetworkSwitchNetworkSwitchTemperatureSensorTemperatureSensorStorageCPUCPUPSUPSUWorkloadNetworkMemoryServerunderTestStorageStorageCPUCPUCPUCPUCPUCPUPSUPSUPSUPSUPSUPSUWorkloadWorkloadWorkloadNetworkNetworkMemoryMemoryServerunderTestCommunicationviaTCP/IPCommunicationviaTCP/IPACPowerInCommunicationviaRS-232orUSBCommunicationviaRS-232orUSBPowerAnalyzerPowerAnalyzerControllerSERTControlSWLocalTCP/IPsocketcommunicationSPECPTDaemonSPECPTDaemonControllerSERTControlSWSERTControlSWLocalTCP/IPsocketcommunicationLocalTCP/IPsocketcommunicationSPECPTDaemonSPECPTDaemonSPECPTDaemonSPECPTDaemonSPECPTDaemonSPECPTDaemon217SPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page9of17Copyright2006-2010SPEC2182202212222232242252.
7.
1GraphicalUserInterface219Oneofthestretchgoalsistheincorporationofagraphicaluserinterface(GUI)tofacilitateconfigurationandsetupoftestruns,allowreal-timemonitoringoftestrunsandtoreviewtheresults.
TheSERTGUIwillleadtheuserthroughthestepsofdetectingorenteringthehardwareandsoftwareconfiguration,settingupatrialrunoravalidtest,displayingresultsreportsandotherfunctionscommontothetestingenvironment.
226227228229230231232233234235236237238239240241242243245246247248249250251252TheSERTGUIwillincludeseveralfeaturestoenableSERTtestingwithminimaltrainingandenhancetheaccuracyofresults:EasyNavigationwithTabbedScreensHowtoUse(in-lineusageguidanceandhelp)ConfigurationDiscovery(Detectfunction)willautomaticallypopulatemostfieldsaboutSUTandControllerhardwareandsoftware.
TheGUIwilldisplay,allowentryofandstorerequiredinformationaboutthetestenvironmentoForuseinreports:e.
g.
CompanyInfo,PlatformConfig,Run-Timeparameters,etc.
oMasterandTestRuninformationfilescanbestored,enablingreuse,savingtimewithmultipleplatforms.
TestSetup,ExecutionandProgressDisplayoStartmeasurements;Choosetypeofrun(trialorfinal)oDisplayprogress,warningsanderrors.
DisplayresultsandenableprintingandcaptureofreportsProvisionsforredundantcomponentsandpowerandperformanceadders.
2.
8Workload244TheexistingSPECbenchmarksaremainlybasedontailoredversionsofrealworldapplicationsrepresentingatypicalworkloadforoneapplicationareaorasyntheticworkloadderivedfromtheanalysisofexistingserverimplementations.
Thesebenchmarksaresuitabletoevaluatedifferentsub-areasoftheoverallserverperformanceorefficiencyifpowermeasurementsareincluded.
Theyarenotdesignedtogivearepresentativeassessmentoftheoverallserverperformanceorefficiency.
ThedesigngoalfortheSERTworkloadhoweveristoincludeallmajoraspectsofserverarchitecture,thusavoidinganypreferenceforspecificarchitecturalfeatureswhichmightmakeaserverlookgoodunderoneworkloadandshowdisadvantageswithanotherworkload.
SPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page10of17Copyright2006-2010SPECTheSERTworkloadwillinsteadtakeadvantageofdifferentservercapabilitiesbyusingvariousloadpatterns,whichareintendedtostressallmajorcomponentsofaserveruniformly.
253254255256257258259260261262263265266267268269270271272273274275276277278279280281282283284285286287288290291292293294295296297298Ifsomecomponentscannotbestressedadequatelybytherespectiveloadpatternthiscanbecompensatedbyadjustingthethresholdforthesecomponents,e.
g.
increasingthepowerallowanceforadditionalcomponentswhicharenotusedbytheloadpattern.
Itishighlyunlikelythatasingleworkloadcanbedesignedwhichachievesthegoalsoutlinedabove,especiallygiventhetimeconstraintsofthescheduletargetedforEnergyStarforServersVersion2.
0bytheEPA.
ThereforetheSERTworkloadwillconsistofseveraldifferentworkletseachstressingspecificcapabilitiesofaserver.
Thisapproachfurthermoresupportsgeneratingindividualefficiencyscoresfortheservercomponentsbesidestheoverallsystemscore.
2.
8.
1Worklets264DevelopingtheworkloadinthetraditionalSPECwaybasedonrealworldapplicationswouldresultincomplextestenvironmentsandhighruntimes,especiallyfortheIOintensiveworkloads,e.
g.
manyclientsystemswouldberequiredfornetworkIOandlargedisksubsystemsfordiskIO.
Theresultingcostsforrunningsuchtestscouldbeprohibitiveforaratingtool.
ThereforetheSERTworkloadwillbeacollectionofsyntheticworkletsforavarietyofdifferentloadscenarios.
Inordertoachieveconsistentresultsfromallworkletsandabroadcoverageoftechnologiesthefollowingguidelinesneedtobeobserved:Eachworkletmustbeadjustabletodifferentperformancelevels,e.
g.
somepredefinedlevelsbetween100%(maximumload)and0%(idle)Eachworkletmustcalibratetomaximumperformancelevelbyitself,i.
e.
nodefinitionofthe100%levelbythetestuserMultipleprogramminglanguagesmaybeusedPrecompiledbinariesofthetestprogramsshouldbeusedwherepossible.
Eachworkletshouldscalewiththeavailablehardwareresources.
Moreresourcesshouldresultinahigherperformancescore,e.
g.
moreprocessor/memory/diskcapacityoradditionalprocessor/memory/diskmodulesyieldabetterresultintheperformancecomponentoftheefficiencyrating.
Inordertofollowtheseguidelinestheworkloadswillprobablybebasedonbatchesofdiscretework,whereeachbatchconstitutesatransaction.
Thedifferentloadlevelswillbeachievedbyschedulingtherequirednumberoftransactions.
Eachworkletshouldperformspecifictasksasinthefollowingexamples.
Thedefinitionofsuchbuildingblocksisnotfinalizedandneedsfurtherinvestigation.
Regardless,theyshouldcoveralargevarietyoffunctionsfromlow-levelprogramprimitivestomorecomplextransactions.
2.
8.
2WorkletExamples289CPUfloatingpointmatrixmultiply,businesslogicsimulation,datacompression/encryptionMemorythroughput,latency,read,writeNetworkIOthroughput,latency,streaming,packetread/writeStorageIOthroughput,databaseaccesspatternsimulation,fileread/writeSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page11of17Copyright2006-2010SPEC2.
8.
3Workloadterms299ExampleWorkloadAllCPU,Memory,DiskandNICworkletsatallloadlevelsWorkletDiskbandwidthtestatallloadlevelsTransactionmemoryread+modify+memorywriteOperationDiskwrite300301302303305306307308309310312313314315316318319320321322323324325326327328329330331332333334335336338339340341342343344345TheSERTworkloadconsistsofmultipleworklets.
Eachworkletwillexecuteaseriesoftransactionswhicharecomprisedoutofvariousoperations.
2.
8.
4WorkletExecution304Inthecurrentdesigntheworkletswillrunconsecutively,eachinitsownphase.
Thisallowsgeneratingindependentscoresatdifferentloadlevelswhichcanbecombinedtoanoverallmetric.
Selectedworkletsoradditionalworkletsspecificallydesignedforconcurrentexecutionmayberunsimultaneouslyinanextraphase.
Concurrentexecutionofworkletswillintroducemorerealistictask-switching,especiallyusingIOloadmodules.
2.
8.
5LoadLevels311MultipleloadlevelsareadesiredgoalofSERTandthedesignwillincludesupportformultiplelevels.
Theactiveidleloadlevelaswellasa100%workloadlevel(notmaxpower)arealreadygoodcandidates.
Prototypetestingwillshowwhichlevelswillbeincludedandifanyweightingwillbenecessary.
2.
8.
6Scaling317SincetheserverefficiencyratingofagivenserveristheprimaryobjectiveofSERT,oneofthemaindesigngoalsforSERTistobeabletoscaletheperformanceonthesysteminproportiontothesystemconfiguration.
Asmorecomponents(processors,memory,diskstorage,networkinterfaces)areaddedtotheserver,theworkloadsshouldutilizetheadditionalresourcessothattheresultantperformanceishigherwhencomparedtotheperformanceonthesameserverwithalesserconfiguration.
Similarly,foragivenserver,whenthecomponentsareupgradedwithfastercounterparts,theperformanceshouldscaleaccordingly.
Thisisaveryimportantaspectofthetoolsinceaddingandupgradingcomponentstypicallyincreasesthetotalpowerconsumedbytheserverwhichwillaffecttheoverallefficiencyresultoftheserver.
Creatingatoolthatscalesperformancebasedonthenumber/speedofCPUsismostreadilyachievable–fortheothercomponents,thecomplexityofimplementingsuchatoolincreasessubstantially.
WhileSERTwillbedesignedtoscaleperformancewithadditionalhardwareresourcesoftheSUT,ifthereareperformancebottlenecksinsystemcomponentsunrelatedtotheaddedhardwaretheSUTitselfmaynotbeabletosustainhigherperformance.
InsuchcasestheadditionofcomponentstotheSUTwillnormallyresultinhigherpowerconsumptionwithoutacommensurateincreaseinperformance.
2.
9ServerOptionsandExpansioncapabilities337Aservermayhavemanyoptionalfeaturesthataredesignedtoincreasethebreadthofapplications.
Thesefeaturesnotonlyabsorbadditionalpower,butalsorequiremorecapacityinthepowersuppliesandcoolingsystem.
SomeSERTworkloadcomponentswillbedesignedtodemonstratetheenhancedcapabilitiesthatthesefeaturesprovide.
However,whilethetoolneedstocreditthesecapabilitiesfortheexpandedworkloadsthattheywillaccommodate,itcannotpenalizeefficientserversthatarenotdesignedwithsubstantialexpansionoptions.
Abalancemustbestruckbetweenprovidingenhancedratingsforenhancedconfigurationsandavoidingeasyqualificationofserversbysimplyaddingfeaturesthatmaynotbeneededinallsituations.
SPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page12of17Copyright2006-2010SPECSERT'sgoalistoavoidunnecessarilypenalizingserversthataredesignedforlowexpandability,whilecreditingserverswithgreaterexpandability.
ForexampleaconfigurationwithfourI/OadaptersinPCIslotsmayexecutetheworkloadofthetoolmoreeffectivelythanaconfigurationwithonlyonesuchadapter.
Ontheotherhanditmayonlyruntheworkloadofthetoolaseffectivelyasaconfigurationwithtwonetworkadapters.
Becausetheconfigurationwithfouradaptersmayrunsomerealworkloadsmoreeffectivelythanconfigurationswithonlytwoadapters,theEPAmayelecttoallowforsomeformof"configurationadders"toprovidecreditforthepowerinfrastructureneededtosupporttheadditionalPCIslots.
,346347348349350351352353354355356357359360361362363365366367368369370371372373374375377378379380381383384385386387389390391392393394395396WhilethedecisiontoincludethisformofpowercreditandthevalueofanysuchcreditwillbeinthehandsoftheEPA,thetoolwillbedesignedandtestedtoensurethat,should"configurationpower-performanceadders"creditsbeincluded,thetoolwillaccommodatethem.
2.
9.
1Redundancy358Manyservershaveredundancybuiltinforpowersuppliesandcoolingfans.
Someserversincludedifferentlevelsofredundancyformemory,disk,andevenprocessors.
Adesigngoalistoincludeaccommodationforredundantcomponents,althoughnospecifictestsareplannedforenergymeasurementunderfaulttolerantconditionswhenoneofaredundantsetofcomponentsisdisabled.
2.
9.
2IOComponent364DiskandNetworkIOcomponentarestronglydesiredtoprovideabetter-roundedpictureofsystemperformanceandpowerthanaCPU-centrictest.
SPECisintheearlystagesofevaluatingIOworkloadsforSERT,sothissectionprovidesmanydiscussionpointsbutnotnecessarilyconclusions.
SPECrecognizesthatsomeoftheitemsinthenexttwosectionsmaynotbereasonableorpracticaltotestormeasureinameaningfulway.
Inthosecaseswewouldsuggesttheuseof"configurationpower-performanceadders"tocompensatefortheextrapowerdrawassociatedwithextrafunctionality.
Otheritemsunderconsiderationinclude:Differenttypes/quantitiesofIOfordifferentservercategoriesSelf-calibratingperformancemeasurementsforthediskandnetworksubsystem2.
9.
3DiskIO376IdeallythediskIOcomponentofSERTwouldgivecreditfor:HigherperformancedisksubsystemsLargercapacitydisksubsystemsReliabilityandavailabilityfeatures(RAID,batterybackedcache,etc)2.
9.
4NetworkIO382IdeallythenetworkIOcomponentofSERTwouldgivecreditfor:HigherperformanceNetworkInterfacesLargertransferspeedNetworkInterfacesReliabilityandavailabilityfeatures2.
10Metric/Score388WhileSERTisnotintendedtobeabenchmark,neverthelessasaratingtoolitmustproduceametricorscoreindicativeoftheefficiencyoftheserverundertest.
ThatmetricmustcombineboththeperformanceoftheSUTaswellasitspowerconsumptioninawaythatallowscomparisonamongallsystemssubjectedtoit.
Thedesiredoutcomeofthatcomparisonisaquantitativemeasureoftherelativepower-performanceefficienciesofthesystems.
Thesystemwhichproducesthehighermetricshouldhavegreaterpower-performanceefficiencythanthesystemwhichproducesthelowermetric.
Sincedifferentarchitecturesperformdifferentlyondifferentworkloads,SERTmaybecomposedofseveraldiscreteworkletstoensurearchitectureneutrality.
EachworkletwillproduceameasureSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page13of17Copyright2006-2010SPECrepresentingtheperformanceachievedbytheSUT,whichthenmustbecombinedwiththemeasuresproducedbytheotherworkletstoyieldametricindicativeoftheoverallperformanceoftheSUTonallworkletsusedinthetool.
SPECrecommendsthatthemultipleperformancemeasuresproducedinthismannerbecombinedintoasinglemetricasthegeometricmeanoftheindividualmeasures.
397398399400401402403404405406407408409410411412413414415416417418419420421423424425426428429430431432433434435436437438439440441442443444445446447Thegeometricmeanofindividualworkletperformancemaybeusedwhethertheindividualworkletsarerunsequentiallyorsimultaneously.
Dependingontheworkletschosenandthemagnitudesoftheirindividualmeasures,weintendindexingthemeasurestoasetofreferencescoresbeforecombiningthemintothesinglemetricasthegeometricmean.
Thesetechniqueshavetheadvantagesofrenderingthesinglemetricunit-less,andofkeepingthescaleoftheindividualmeasureswithinsimilarranges,sothataworkletwithlargemagnitudeindividualmeasuredoesnotoverwhelmtheresultfromaworkloadwithasmallermeasure.
Oncedetermined,theoverallperformancemustbecombinedwiththemeasuredpowerconsumptionoftheSUTinawaythatdemonstratesthepower-performanceefficiencyofthesystem.
ThiswillbeacomplexcalculationautomaticallyperformedbySERTtotakeintoaccountthepower-performanceefficiencyoftheSUTatdifferentutilizationlevels.
ThemetricthatisproducedbySERTisseparatefromtheEnergyStarrating.
TheEPAwilldeterminecriteriaforEnergyStaracceptanceofwhichtheSERTscoresmaybeonlyapart.
It'santicipatedthatthetop25%oftestedunitswillachieveEnergyStarqualification.
A"gold-level"EnergyStarqualificationmaybeavailableforunitsachievinginthetop5%ofresults.
AdditionallytheEUhasproposedasystemofgraduatedachievementinpower-performanceefficiencywithlevelsAthroughF,forwhichtheywilldeterminetheoverallcriteria.
ServerundertestmaybeplacedindifferentcategoriesbytheEPA.
TheEPAwilldecidehowtoapplythesecategoriesandwhetherunitsinaparticularcategorymaybecomparedtounitsinanothercategory.
2.
11ReportingandOutputFiles422SERTwillproducetworeportsandasetoflogfiles.
ThereportswillbecreatedinXMLformat,inordertoreducetheeffortforbothEPAandthepartnerindisplayingandorstoringthedesiredinformation.
Wewilltakestepsinordertoensureauthenticity(e.
g.
encryption)ofthereports.
2.
11.
1Report1:"SummaryReport"427Thisreportwillcontainaplaceholderfora"passorfail"noticeforthetestedplatform,tobeprovidedbytheEPA.
Atestrunismarkednon-compliantifthetestcompleteswithtechnicalerrors.
Insuchacase,errormessagesand/orwarningswillbeautomaticallyincludedinthereport.
Theinformationinthisreportispublicandcouldbeusedformarketingpurpose.
Itemsincludedinthisreportare:EPAPartnernameandEPAPartnerIDEPAEnergyStarCategoryofthetestedplatformTestDateandLocation(plus"Testedby")TestedPlatformManufacturerandModelNumberPlaceholderfor"Pass/Fail"WarningsorErrorNoticesifapplicableSystemConfigurationinformation(Redundantcomponentstobemarkedappropriately):oformfactoronumberandtypeofprocessorsoavailableprocessorsocketsomemorysize,type,#memoryDIMMs,#DIMMSlots,MaxMemoryCapacityoavailableexpansionslotsonumberofandmake-modelofpowersupply,outputrating,min/maxoInputpowerSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page14of17Copyright2006-2010SPECoOSsupported/OSusedfortest448449450451452453454456457458459460461462463464465466467469470471472473474475476477478479480481482483484485486487488490491492493494495onumberofandmake-modelofstoragecontrolleronumberofandmake-modelofmassstoragedevicesonumberofandmake-modelofnetworkinterfacecards(NICs)oManagementControllerorServiceProcessorInstalled[Yes/No]oOtherHardwareFeatures/Accessories2.
11.
2Report2:"PowerandPerformanceDataSheet".
455ThisreportwillcontainalltheinformationtheEPArequiresandthatisdeemednecessarybySPEC.
ThePowerandPerformanceDataSheetwillbepublic,butmarketinguseisprohibitedbySERTFairUsageRules.
TheinformationisintendedtobedeliveredtotheEPAinaformmostexpeditiousforEPAreview.
Thisreportwillcontainallthedatafromthe"SummaryReport"withthefollowingadditionaldetailsections:OverallResult/ScoreAlltargetloadlevelresultsHardwareandSoftwareConfigurationPowerMeasurementSummaryEnvironmentalinformation2.
11.
3Logfiles468Asetoflogfileswillbeproducedforeachtestrun.
Theinformationinthelogfilesisintendedtobe"non-public".
Thesefileswillbeidentifiedbyarunserialnumbersuchthatmultipleconsecutivetestrunsproducemultiplelogfilesets.
Eachlogfilewillbearecordofactionsfromthesoftwareduringthevariousphasesofthetesting,includingerrorsandwarnings.
Theintentofthelogfilesisforauditingandsupportpurpose.
oProblemsorfailurescanbemoreeasilyresolvedwiththislowleveldetailrecord.
Ifanyissuesarisewithregardtotheaccuracyorveracityofthepartnerreports,theselogfiles(potentiallyencrypted)shouldbeadequatetoresolvemostissues.
Examplesoflogfilecontentare:oHandshakevalidationmessagesamongvariouscomponentsoErrororwarningmessagesoStatechangemessages/notifications.
o'Transaction'instantaneous/periodicsummaryinformationo'Transaction'responsetimesTheEPAmayrequirethatanyoralloftheaboveoutputsbedeliveredpriortoEnergyStarqualification.
Regardless,thepartnermustcommittoarchivingalloutputfromanyresultssubmittedtotheEPA.
2.
12TestSoftware489Another"stretchgoal"ofSERTistoenablea"LiveCD"approachtotoolinstallation,forsomeenvironments–suchthattheentiretoolsuitealongwiththeunderlyingoperatingsystemcouldallberunfromasinglebootableCDorDVDwithnootheroperatingsysteminstalledontheSUT.
Thisshouldprovideincreasedeaseofinstallationandimprovetheadoptionrateofthetool.
Possibleissueswiththisapproachincludethelackofspecifichardwaredriversfornewerdevices,thepotentiallackofvendorspecificpowermanagement,licensingandavailabilityissuesforsomeSPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page15of17Copyright2006-2010SPECoperatingsystems.
Alternativesincludeallowingadditionaldriverstobeinstalledduringsetup,orprovidingseparatetestinstallerswithbinariesforusewithavendor'sownas-shippedOSinstallation.
4964974985005015025035045055065075085095105115125135145155165195205215225245255265275285305315335345355365375385395405415435445452.
13Validation/Verification499SERTsoftwarecomponentswillimplementsoftwarecheckswhereverpossibletoincreaseinformationaccuracy,verifyuserinput,monitorrun-timedatacollection,andvalidateresultswiththeintentofimprovingaccuracyandremedyingusererrors,preventinginvalidreportstotheEPA.
Whenconditionsorresultsdonotmeetspecificcriteria,warningswillbedisplayedanderrormessageswillappearintheSERTreports.
ThesefeatureswillmakeiteasyfortheEPAPartnertogeneratecompliantresultsandpreventsubmissionoferroneousreportstotheEPA.
Examplesofcompliancecheckingare:Verifyinputproperties(parameters)andrun-timedurationofloadlevels.
Temperatureoutofrangewillbereported.
PowerandTemperaturereaderrorsmustbeunderachosenthreshold.
AlltheSERTsoftwarecomponentswillperformvalidationcheckswithinthedomainoftheirfunctions,e.
g.
warningsofconnectionproblems,logmeasurementerrorsandout-of-rangeconditions,warningtheuserofmissingorincompleteinformationandcheckthevalidityofsomeentereddata.
OthernewvalidationmethodswillbeconsideredastheSERTsoftwaredesignandimplementationprogresses.
3Logistic5173.
1Ownership518SPECretainsownershipoftheSERTcodeandhasthefinalsignoffonthecode.
SPECiscommittedtoworkingwiththeEPAduringthedevelopmentofSERTtoensurethatcodemeetstheindustriesrequirements.
3.
2Support523TheendusersupportlogisticsandscaleinvolvedwithEnergyStarprogramsarebeyondwhatSPEC'svolunteerdevelopmentteamcansustainandsupport.
Thus,SPECshouldonlybeconsideredasasecondlineofsupportreservedforfundamentalSERTissuesonly.
TheEPAwillberesponsibleforprovidingdirectcustomersupportforanyEnergyStarprogramwhichutilizesSERT.
3.
3Maintenance529SPECwillmaintaintheSERTcodebase.
3.
4Updates532TheinitialversionofSERTistargetedtosupportEPAE*Tier2requirements.
Afterthereleaseoftheinitialversion,SPECwillcontinuetodevelopSERTintwoways.
Fixesfordiscoveredbugsandfeatureenhancementsoradditionswillbeprovidedasrevisionsofthecurrentversion.
LicenseesofthecurrentversionofSERTareentitledtothesebug-fixand/orenhancementrevisions.
SPECwillcreatenewversionsofSERTtosatisfytherequirementsoflaterE*Versions.
End-useruseofnewversionsrequireslicensepurchase.
SPECwillcooperatewiththeEPAonthelogisticsofdistributingupdates.
3.
5Licenseeandprice542SERTwillhaveaSPEClicenseandfeestructureforendusercompanies.
AswithotherSPECproducts,allSPECOSGmemberorganizationsareentitledtoacomplimentarySERTlicenseandfreeversionandrevisionupdatesinconsiderationoftheirdevelopmentcostsandcontributions.
SPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page16of17Copyright2006-2010SPECAdditionalpotentialEPAE*partnercompaniesareentitledtopurchasetheirinitialSERTlicenseatacostanticipatedtobeontheorderof$1000.
AsnewE*VersionsareannouncedandSERTisupdated,newversionsofSERTwhichsupportthelatestE*Versionmaybepurchasedbylicenseesoftheimmediatelyprecedingversionattheanticipatedcostof$500.
546547548549550551552554555557558559560562563565566567569570571572574575576577578579580581582583584585587588589590591592593594595596Whenavailable,bugfixorenhancementrevisionstothecurrentversionwillbeprovidedtocurrentlicenseesfreeofcharge.
3.
6Delivery553SPECwilldeliverySERTviaCD/DVDtotheEPAPartners.
3.
7SubmissionReviewProcess556TheEPAwillberesponsibleforreviewingsubmissionsforacceptancefortheEnergyStarforServerprogram.
SPECwillnotreviewEnergyStarforServersubmissions.
SPECiscommittedtoautomatingSERTextensivelywiththeendgoalofminimizingthereviewprocess.
3.
8Trademark561Productandservicenamesmentionedhereinmaybethetrademarksoftheirrespectiveowners.
4SERTandEPAEnergyStarforServerVersion2.
0564InordertoensurethatSERTisutilizedintheintendedmatter,werecommendtheinclusionofthefollowingitemsintheEnergyStarforServerSpecification.
4.
1TestedasShipped568Toprovideresultsthatarerepresentativeofacustomerenvironment,thegoalistotestsystemsinan"as-shipped"state.
Nosupertuningwouldbeallowed,butratheralimitedlistfortuning/optimizationmightbepermitted.
ThislistwouldbeagreedonbeforeSERTrelease.
4.
2FairUseofSERTinformation573AcleargoaloftheEnergyStarprogramistohavethebroadestpossibleparticipationamongvendors.
Experienceinthecomputerindustry'sperformancebenchmarkcommunitydemonstratesthatwhenperformancedetailsbecomeavailableformarketingpurposes,onlyvendorswithsuperior(atthetimeofpublication)productsareincentedtopublishresults.
Toencouragebroaderparticipationacrosstheindustry,asetofstrongrulesmustbeinplacethatwillrestrictmarketinguseofanyofthedetailedinformationgeneratedbythetool.
NodatabesidestheactualEnergyStarqualificationshouldbeutilizedinEPAPartners'marketingcollateral.
TheseruleswillbestipulatedinboththelicenseforthetoolandtheEPAPartneragreement.
Notethat,whiletheserulesarenotstrictlyapartofthetool"design",theexistenceoftheserulesarenecessarytoallowtheflexibilityofthedesignandthedeliveryofdetailedconsumerinformationthatisdesired.
4.
2.
1FairUseRules586TheonlyinformationprovidedbythetoolthatcanbeusedformarketingcollateralistheEnergyStarqualificationofaserverconfigurationorserverfamilyTheonlyinformationprovidedbythetoolthatcanbeusedforpubliccomparisonistheEnergyStarqualificationofaserverconfigurationorserverfamilyAllotherpubliclyavailableinformationfromthetoolismadeavailabletohelptoverifythatthetestswereruncorrectlyandtoallowconsumerstobetterunderstandhowwelltheconfigurationstestedmatchtheirspecificneeds.
IfthetoolisusedforresearchtogenerateinformationoutsideoftheEnergyStarprogram,theinformationmaynotbecomparedtotheEnergyStarprogramresultsandcompetitivecomparisonsmaynotbemadeusingthedatagenerated.
SPECpowerCommitteeSERTDesignDocument1stPublicDraftdate:03/16/2009page17of17Copyright2006-2010SPEC597598600601602603605606607608610611612613614616617618619620621622623624625626627628629TheEPAEnergyStarQualificationisgovernedbyEPArules.
4.
3Accredited,Independentlaboratory599Therequirementtouseaccredited,independentlaboratoriesmayplacealargeburdenonEPAEnergyStarpartners,especiallysmallercompanies.
Werecommendtheuseofanindependentlaboratoryasanoption,butnotimplementingthisasarequirement.
4.
4SupplyVoltagetolerance604Inordertouseavoltagewithina1%difference,anextravoltagesourceisneeded.
Thiswillunnecessarilyincreasethecostforthepartner,especiallysmallercompanies.
Werecommendthetolerancebesetto±5%.
4.
5DCLine-Voltage609TheSPECPTDaemondoesnotsupportDCmeasurementsandcurrentlynoresourcesaredevotedtoincludingthissupport.
ItisouropinionthatcomparingserverspoweredbyACagainstserverspoweredbyDCisnotfair,sincetheAC-DCconversionlossesarenotincludedinDC-poweredserver.
WewouldrecommendcreatingaseparatecategoryforDC-poweredservers.
4.
6Airflow615Regardingtherequestedthermalflowratesofservers(TotalPowerDissipation;DeltaTatExhaustatPeakTemperature;AirflowatMaximumFanSpeedatPeakTemperature;AirflowatMinimumFanSpeedatPeakTemperature).
SPECrecommendsdeletingthisrequirementinordertoenablesmallercompaniestobecomeanEnergyStarpartneraswell.
Webelievethataunifiedreportingofthethermalflowrateacrosstheindustrycouldhelpfacilitymanagerstoplandatacentersbetterwiththeadditionalthermalloadoftheserversinmind.
Unfortunately,thevolumetricflowdependsontheconfigurationoftheserverandtheEPA'sconceptofProductFamilycouldnotbeused.
Alsowebelievethattheguidelinedescribedisnotstrictenoughtoensureunifiedmeasurementsandreportingacrosstheindustryandtheadditionalcostforeachpartnerneedstobeconsideredaswell.
Inordertoqualifyaserver,avolumetricairflowbench(neitherasimplewindtunnelnorananemometer)isnecessary(Cost~$5000)aswellasatemperaturechamber(Cost~$5000).
Weestimatethetimespenttoproducetheseresultsisaround8hoursforeachconfiguration.
SPECrecommendsdeletingthisrequirementinordertoenablesmallercompaniestobecomeanEnergyStarpartneraswell.
"