Singleapnic
apnic 时间:2021-01-10 阅读:()
AristaR3-Series-MultipleGenerationsofInnovationsThegrowingdemandforhigherbandwidthinacceleratedtimeframesisdrivenbyhyper-scalecloudnetworksandServiceProviders.
Theimminentavailabilityof5GandEdgecomputingwithhigherbandwidthdevices,IoTdata,newmodelsforvideoconsumptionandongoingCyber-securitychallengesraisestheexpectationofintelligentnetworkservices.
Customerrequirementsrarelycallforanentirelynewnetworkarchitecture.
Newtechnology,howeverinnovative,mustpreserveexistinginvestments,providebackwardandforwardcompatibility,andenablemigration.
Manysystemvendorsseethispaceofchangeasafundamentalchallengethatshouldbeaddressedwithnewandradicallydifferentforwardingarchitectures,pipelinesoroperatingmodelsthatdisruptcustomers,ratherthanfacilitatingtheevolutionaryapproach.
AristatakesanevolutionaryapproachwhenitcomestofacingthesechallengeswiththenewR3generation.
TheAristaR3-SeriescomprisesasetofproductsinthelatestgenerationofthehighlysuccessfulArista7500and7280Seriesplatforms.
TheArista7500Serieswasfirstreleasedin2010asahighdensity10Gsystemandoverthelast10yearshasevolvedwitheachnewgenerationofferinghigherperformanceandmorecapabilitiesforrouting,telemetryandnetworksecurity,withinvestmentprotection.
Thelatestgenerationoftheseplatformsarethehighcapacity7800R3,7500R3,andthe7280R3,offeringabreadthofformfactor,performanceanddensitythatscalesfrom4Tbpstoover460Tbps,withallsystemssharingcommonandconsistentcapabilities.
ThispaperintroducesthelatestgenerationoftheR-Series,outlinesthecapabilitiesandinnovationsthataddresstheconstantlyevolvingnetworkchallenges,andtheintroductionof400Gnetworks.
Figure1:AristaR3-Seriesfrom4Tbpsto460TbpsCommonEOSAristaEOSInnovationsand10YearsofExperienceSince2010,whentheveryfirst10Gbased7500Seriesand1Gbased7048Serieswerelaunched,eachnewgenerationoffixedandmodularsystemshasintroducedasetofinnovationsdrivenbyAristaEOS.
Thefirstgenerationofdeepbuffermodularandfixedsystemsdroveadvancedtwo-tierdesignsforOilandGasresearch,HighPerformanceCompute,Onlineretailers,bigdataandsearchthatsetthestandardforthenext10years.
The7500SerieswassowellreceivedthatitwasawardedNetworkWorldBestofInteropin2010.
The2013launchofthe7500Eand7280SEplatformsexpandedSpineandLeafnetworksto40Gand100G,andintroducedaseriesofinnovationsincludingTapAggregation,latencymonitoring,Coherent100GDWDMwithintegratedMACsecencryptionforDCIandsupportforVXLANforoverlaynetworks.
In2013the7500ESerieswasawardedasecondbestofInterop,recognizingtheinvestmentprotection,coupledwiththeseinnovationsand3Xperformanceincrease.
The2016releaseofthe7500Rand7280RUniversalSpinebroughtsupportforCloud-graderoutingforInterneteXchange(IX)providers,CloudWAN,nextgenerationcableoperatorsandOverTheTop(OTT)servicesandContentDeliveryNetworks(CDN).
TheR-SeriesofscalableplatformssupportinternetscaleroutingwithFlexRoute,EOSNetDBinnovations,modernprotocolsandopenautomation,whichallowedforthetransformationofroutingarchitecturesandreplacedlegacyrouters.
Onthe7500R2and7280R2Series,AlgoMatchandAcceleratedsFlowprovideunprecedentedlevelsofsecureaccesscontrolanddeepvisibilityintohighcapacity100GnetworksandMACsecwhichwasexpandedforDataCenterInterconnect(DCI)solutions.
ThenextgenerationoftheR-Seriesisavailableinthe7800R3,7500R3andthe7280R3SerieswithsupportformoreEOSdriveninnovations.
ProvenArchitectureInvestmentProtectionCustomersrarelygetacleansheetofpaperwithwhichtodesignnetworks.
Newsystemshavetocoexistwithpreviousnetworksdesigns,andofferinplaceupgrades.
TheR3-Seriesprovidecontinuedinvestmentprotection,fromtheoriginal7500Seriesthroughtothe7500R3Series.
Interoperabilityofthe7500R3Serieswiththepreviousgenerationoflinecardsandfabricsallowsmixinggenerationstopreserveexistinginfrastructure.
ReuseofcommonarchitectureacrosstheR3-Seriessystemsprovidesacommonsetofbuildingblocks,leveragedtoacceleratetimetoserviceofnewsystems.
Figure2:Arista7500Series10yearsofcontinuousinnovationsandexperienceSingleOSacrosstheentireportfolioaddressesthechallengeofqualifyingnewproducts,whereexistingtools,processesandfeaturesworkasexpected.
Thisisincomparisontothechallengeofintroducinganallnewproductwithadifferentoperatingsystemanddissimilarfeatures.
FeatureRichbuildingonthecodebaseofEOSandleveraging1000'sofsoftwaredevelopmentyearswithcommonplatformsallowsforfasterqualificationwithouthavingtowaitforfeaturestobereleased.
AristaEOSisamodernoperatingsystemandcontinuestoevolve,withongoingrefactoringofmajoraspectsofthecode,addingscale,newfeaturesandsupportfornewsilicon,allwithoutsacrificingsupportforexistingfeaturesandcodequality.
ConsistentVoQnon-blockingfairdeliveryTheAristaR3-Seriesleveragesadeepbuffervirtualoutputqueue(VOQ)architecturewithacellbasedfabricthatis100%efficientandprovidesconsistentlylowlatencyandjitter,eliminateshead-of-line(HOL)blockingandvirtuallyeliminatespacketdropseveninthemostcongestednetworkscenarios.
Anadvancedtrafficschedulerfairlyallocatesbandwidthbetweenallvirtualoutputqueueswhileaccuratelyfollowingqueuedisciplinesincludingweightedfairqueuing,fixedpriorityorhybridschemes.
Asaresult,theArista7280R3,7500R3and7800R3Seriescanhandlethemostdemandingdatacenterrequirementswithease,includingmixedtrafficloadsofreal-time,multicast,andstoragetrafficwhilestilldeliveringlowlatency.
TheadvantagesoftheVoQarchitecturearethattrafficisnotforwardedtoadestinationwithoutacredit,andthatthecellbasedfabricusesallpathstotheegresspacketprocessortoavoidcreatinghot-spots.
Thecellbasedfabricleveragesavariablecellsizeandhandlesmixedinterfacespeedswithnoimpactonsystemperformance.
Figure3:CellBasedFabricandIngressVoQArchitectureTable1:RealInvestmentProtectionCharacteristicWhyitMattersH/WInvestmentProtectionCompleteinteroperability3rd/4thgenerationExistinglinecards,fabricmodules,chassis&interfacesworkas-isTrueHardwareInvestmentProtectionSignificantCapExsavingsNoforkliftupgradeCommonArchitectureDeepbufferIngressVoQCellbasedFabric100%efficientHandleunpredictableworkloadsNohotspotsOneOSSingleOSacrossentireportfolioReducesOpExExistingtools&processesworkas-isProvenPlatformFeature-richdayoneHighqualityondayoneLeverage1000sofSWdevelopmentyearsFasterqualificationEnhancementsfortheR3-Seriesincludemulticastexpansionintheingress,fabricoregress,toreducedropsorqueueing.
Intheeventofadrop,itcanbemirroredtoadestinationforanalysis.
DeeppacketbufferswithHBM2TheR3-SeriesisconsistentwithpreviousgenerationsoftheR-Serieswithflexiblepacketbufferscombiningdedicatedon-chipresourceswithhighbandwidthaccessto8GBofHBM2memoryperpacketprocessor.
Thebufferaccessspeedhasincreasedatafactoroffiveoverthepreviousgeneration,usingadvancedmemorytechnologycommonlyfoundingraphicsandAIsystems,alignedtoa5Xincreaseinthepacketprocessorperformance.
HighBandwidthMemory-2,orHBM2,whenintegratedtothepacketprocessorpackage,enablesbothhighbandwidthandgreatercapacity,requiring43%lesspowerthantheequivalentGDDRmemory.
IntegratedHBM2memoryallowstheR3-Seriestodeliverahighnumberofinterfaces,increasessystemreliability,andrequiressubstantiallylesspowerthancomparablehighspeedmemoryoptionssuchasHMC,whichlackthepackageintegrationbenefits.
HBM2isavailablefrommultiplesources,isbackedbytheJEDECconsortium(1),incommonuseongraphicsprocessorsandhasatechnologyroadmaptohigherbandwidthandperformance.
The32MBofon-chipbuffersareusedfornon-congestedforwardingwithseamlessexpansiontotheHBM2basedoninstantaneousorsustainedcongestionwithnopacketdrops.
BuffersareassignedperVoQandclassofservice,withnorequirementtotuneforoptimumbehavior.
MemorymanagementisasimportantascapacityandintheR3-Seriesthepacketbuffermemoryisoptimizedformodernnetworks,thatarenaturallybursty.
Avoidingwastefromsmallframes,memoryisallocatedin4Kbytesegmentstoaccommodatemultipleframes,whicharequeueddirectlyfrommemorytothedestinationpacketprocessor,andarenotreturnedviatheOCB.
PowerEfficient100Gand400GEthernetTheR3-Seriesintroducessupportforhighdensity100Gand400GEthernetinavarietyofplatformsfrom24portsof400Gin1Utoamodularsystemwith576ports,andincreaseperformanceby3Xoverthepreviousgeneration.
Figure4:HBM2andPacketProcessorPackagingFigure5:SeamlesspacketbufferoptimizationwithOCBandHMB2Asingleportof400Gcansupport1x400G,2x200Gor4x100Gmodeswithupto36portsinasinglelinecard,equivalentto144portsof100G.
Whiledemandfor400GisdrivenbythelargestscalenetworkstheR3-Series100Gsystemsprovidehigherbandwidthat35%lowerpowerpergigabit,enablingtherapidadoptionofnextgenerationsystems.
Tobridgethemigrationfrom100Gto400GtheR3-Series400Gand100Gsystemsprovideflexibleportspeedoptionsandmediatypeswithbreakoutcablesandparalleloptics.
FlexRouteImprovementsInternetRouteScaleAllR3-Seriesproductssupportfullinternetroutescalewithacapacityof1.
3MroutesleveragingacommonsetofforwardingtableresourcesfortheL2andL3tables.
TheR3K-Serieswithexpandedsizetablescansupportupto2.
5MIPv4andIPv6routes,allowingformultipleoverlappingfulltablesorformanyyearsofgrowthoftheInternet.
Routesareheldonthepacketprocessor,withouttheneedforexternalTCAMsoradditionalmemory,affectingperformance,powerefficiencyoroverallsystemdensity.
GeoffHuston,ChiefScientistatAPNIC,theAsiaPacificRegionalInternetRegistryhasbeenprovidingresearch,analysisandcommentaryontheglobalinternetroutingtableforoveradecade.
Thepublishedprojectionforthenext5yearsisshownbelow.
AsIPv6addressestakemorespacethanasingleIPv4it'slikelythatcapacityfor1.
3Mrouteswillberequiredtoholdafulltable,inthenearfuture,andthat2.
5Misrequiredforprojectedgrowthbeyond2024.
Source:https://labs.
apnic.
net/p=1195Figure6:ImprovedpowerperbandwidthTable2:HistoricgrowthofIPv4andIPv6announcementsMetricJan2016(Actual)Jan2017(Actual)Jan2018(Actual)Jan2019(Actual)Jan2020(Prediction)Jan2021(Prediction)Jan2022(Prediction)Jan2023(Prediction)Jan2024(Prediction)IPv4Prefixes587,000646,000699,000755,000810,000864,000919,000974,0001,028,000(%Increase)9%8%7%7%6%6%6%5%IPv6Prefixes(L)27,00037,00045,00062,00075,00089,000102,000116,000130,000(%Increase)27%18%27%17%16%13%12%11%IPv6Prefixes(E)27,00037,00045,00062,00083,000109,000145,000192,000255,000(%Increase)27%18%27%25%24%25%24%25%Total(linearIPv6)614,000683,000744,000817,000885,000953,0001,021,0001,090,0001,158,000(%Increase)6%6%6%6%6%6%6%6%Total(exponentialIPv6)614,000683,000744,000817,000893,000973,0001,064,0001,166,0001,283,000(%Increase)10%8%9%9%8%9%9%9%NextGenerationEdgewithEVPNandSegmentRoutingTheR3-SeriesexpandsonSegmentRoutingandEVPNsupportwithmoreflexibleoptionsforunderlayandoverlaytopologies.
Theysupportexpandedlabeldepth,newtunneltypesincludingv6GREandVXLAN-v6andimprovedloadbalancingondeeptunneledpackets.
Asmultipletechnologytransitionsevolveatthenetworkedgee.
g.
multicloud,workloadmovementbetweenprivateandpubliccloud,contentcachesnearcustomerproximityanddistributedNFVservicesattheedge,connectivityandsegmentationarekeytoagileservicedelivery.
TheR3-Seriesprovidestheflexibilitytodelivercostoptimizedandhighdensityinterconnectsolutions,toaccelerateNFVedgeserviceswithEVPN,andSegmentRoutingsolutionsfortheWAN(2).
ProgrammabilityProgrammableProcessingwithElasticPipelineTheR3-Seriesdeliversadvancedprogrammablepacketprocessingwithanewelasticpipeline.
Complexfeaturesrequiringadditionallookupcyclesareimplementedthroughapoolofstagesinthepipeline.
Existingadvancedfeatures,includingAcceleratedsFlow,areimplementedintheprogrammablepipelineavoidingexternaloffloadengines,andincreasingflexibility.
FlexiblePacketProcessorTablesInlargescalenetworksthetotalscalabilityisboundbyoneormoreresources:Routes,Hosts,MACaddressesorMPLSLabels.
Intheseenvironmentstheabilitytoincreasethecapacitybyflexiblyreallocatingtablesallowsformoreefficientdesigns.
TheR3-Seriesleverageacommoncentraldatabaseofforwardingandlookupresources,availabletoallingressandegressstagesforthecommonforwardingentries.
Theseresourcesareallocatedusingforwardingprofilesthatensuretheoptimalallocationtodifferenttables,forawiderangeofuse-cases.
TheL3optimizedprofileexpandstheroutingandnext-hoptablestoaddressnetworkswherelargerroutetablecapacityisrequired,whiletheBalancedprofileallocatestheresourcesacrossalltablesandissuitedforleafandspinedatacenterapplications.
TheR3K-Seriessystemsandlinecardsincludesupportfordoublethecapacityofforwardingtablesforenvironmentsthatrequireresourcestobeexpandedfurther.
TheR3Kpacketprocessorshave2xlargerintegratedforwardingtables,andtheprofilesoptimizeforPeering,DCIandIP/MPLSapplicationswithupto2.
5Mroutesor2.
8MMPLSLabels.
Figure7:NextGenerationTelcoEdgewithEVPNandSegmentRoutingFigure8:R3-SeriesPacketProcessorFlexibleTablesAdvancedNetworkTelemetryAcceleratedsFlowwith1:100at400GTheR-SeriessupportforAcceleratedsFlowenabledsupportforhighratesamplingacrossasystemwithdedicatedsFlowengines.
TheR3-SeriesexpandsAcceleratedsFlowtobesupportedonallsystems,witheachpacketprocessorleveragingtheprogrammablepipelineforhighsamplingratesat400Gand100Gonallports,withtheadditionalmetricsofAS-PathandNext-Hop.
Telemetryinformationontrafficflowsisvaluablewhendeterminingthesourcesofrogueandmisbehavingapplicationswithinthenetwork,togetherwithflowanalysisandtrafficengineeringofpaths.
In-bandNetworkTelemetryforflowandpacketforensicsIn-bandnetworktelemetry,orINT,isastandardsapproachtoprovidingdeepvisibilityintotrafficinreal-time,withnoimpactonswitchperformance.
INTisavailableontheR3-Seriesandprovidesper-flowmonitoringofdrops,latency,congestionandcanreconstructthenetworktopologyandpaths.
INTinformationcanbeexportedinIPFIXorsFlowformatstoamanagementsystemorcollectorsuchasAristaCloudVision,forpredictiveanalyticsanddeepforensics.
Figure9:R3K-SeriesPacketProcessorFlexibleTablesFigure10:AcceleratedsFlowwithstatestreamingFigure11:InbandNetworkTelemetryinActionCopyright2019AristaNetworks,Inc.
Allrightsreserved.
CloudVision,andEOSareregisteredtrademarksandAristaNetworksisatrademarkofAristaNetworks,Inc.
Allothercompanynamesaretrademarksoftheirrespectiveholders.
Informationinthisdocumentissubjecttochangewithoutnotice.
Certainfeaturesmaynotyetbeavailable.
AristaNetworks,Inc.
assumesnoresponsibilityforanyerrorsthatmayappearinthisdocument.
April6,201902-0082-02AlgorithmicACLsTheR3-seriesprovidesupportforalgorithmicACLsonallsystemsthatincreaseboththescaleandflexibilityleveragingtheprogrammablepipeline.
Withsupportforover100KrulestheR3-SeriesaresuitedforinternetedgeandcybersecurityroleswithstatefulACLsandrichcountersidentifyingandfilteringwithnolossinperformance.
SummaryTheAristaR3-Seriesexpandstheindustryleadingspineandleafsystems,preservinginvestmentwithaconsistentarchitecture,EOSfeaturesandtheintroductionofsupportfor400Galongwithnewinnovationsforprogrammability,telemetry,routingandsecurity.
(1)https://www.
jedec.
org/standards-documents/docs/jesd235a(2)EANTCTestreportsvalidatingEVPNandSegmentRoutingfeatures(3)APNIChttps://labs.
apnic.
net/p=1195Figure12:AlgorithmicACLscombinesoftwareandhardwareinnovation
Theimminentavailabilityof5GandEdgecomputingwithhigherbandwidthdevices,IoTdata,newmodelsforvideoconsumptionandongoingCyber-securitychallengesraisestheexpectationofintelligentnetworkservices.
Customerrequirementsrarelycallforanentirelynewnetworkarchitecture.
Newtechnology,howeverinnovative,mustpreserveexistinginvestments,providebackwardandforwardcompatibility,andenablemigration.
Manysystemvendorsseethispaceofchangeasafundamentalchallengethatshouldbeaddressedwithnewandradicallydifferentforwardingarchitectures,pipelinesoroperatingmodelsthatdisruptcustomers,ratherthanfacilitatingtheevolutionaryapproach.
AristatakesanevolutionaryapproachwhenitcomestofacingthesechallengeswiththenewR3generation.
TheAristaR3-SeriescomprisesasetofproductsinthelatestgenerationofthehighlysuccessfulArista7500and7280Seriesplatforms.
TheArista7500Serieswasfirstreleasedin2010asahighdensity10Gsystemandoverthelast10yearshasevolvedwitheachnewgenerationofferinghigherperformanceandmorecapabilitiesforrouting,telemetryandnetworksecurity,withinvestmentprotection.
Thelatestgenerationoftheseplatformsarethehighcapacity7800R3,7500R3,andthe7280R3,offeringabreadthofformfactor,performanceanddensitythatscalesfrom4Tbpstoover460Tbps,withallsystemssharingcommonandconsistentcapabilities.
ThispaperintroducesthelatestgenerationoftheR-Series,outlinesthecapabilitiesandinnovationsthataddresstheconstantlyevolvingnetworkchallenges,andtheintroductionof400Gnetworks.
Figure1:AristaR3-Seriesfrom4Tbpsto460TbpsCommonEOSAristaEOSInnovationsand10YearsofExperienceSince2010,whentheveryfirst10Gbased7500Seriesand1Gbased7048Serieswerelaunched,eachnewgenerationoffixedandmodularsystemshasintroducedasetofinnovationsdrivenbyAristaEOS.
Thefirstgenerationofdeepbuffermodularandfixedsystemsdroveadvancedtwo-tierdesignsforOilandGasresearch,HighPerformanceCompute,Onlineretailers,bigdataandsearchthatsetthestandardforthenext10years.
The7500SerieswassowellreceivedthatitwasawardedNetworkWorldBestofInteropin2010.
The2013launchofthe7500Eand7280SEplatformsexpandedSpineandLeafnetworksto40Gand100G,andintroducedaseriesofinnovationsincludingTapAggregation,latencymonitoring,Coherent100GDWDMwithintegratedMACsecencryptionforDCIandsupportforVXLANforoverlaynetworks.
In2013the7500ESerieswasawardedasecondbestofInterop,recognizingtheinvestmentprotection,coupledwiththeseinnovationsand3Xperformanceincrease.
The2016releaseofthe7500Rand7280RUniversalSpinebroughtsupportforCloud-graderoutingforInterneteXchange(IX)providers,CloudWAN,nextgenerationcableoperatorsandOverTheTop(OTT)servicesandContentDeliveryNetworks(CDN).
TheR-SeriesofscalableplatformssupportinternetscaleroutingwithFlexRoute,EOSNetDBinnovations,modernprotocolsandopenautomation,whichallowedforthetransformationofroutingarchitecturesandreplacedlegacyrouters.
Onthe7500R2and7280R2Series,AlgoMatchandAcceleratedsFlowprovideunprecedentedlevelsofsecureaccesscontrolanddeepvisibilityintohighcapacity100GnetworksandMACsecwhichwasexpandedforDataCenterInterconnect(DCI)solutions.
ThenextgenerationoftheR-Seriesisavailableinthe7800R3,7500R3andthe7280R3SerieswithsupportformoreEOSdriveninnovations.
ProvenArchitectureInvestmentProtectionCustomersrarelygetacleansheetofpaperwithwhichtodesignnetworks.
Newsystemshavetocoexistwithpreviousnetworksdesigns,andofferinplaceupgrades.
TheR3-Seriesprovidecontinuedinvestmentprotection,fromtheoriginal7500Seriesthroughtothe7500R3Series.
Interoperabilityofthe7500R3Serieswiththepreviousgenerationoflinecardsandfabricsallowsmixinggenerationstopreserveexistinginfrastructure.
ReuseofcommonarchitectureacrosstheR3-Seriessystemsprovidesacommonsetofbuildingblocks,leveragedtoacceleratetimetoserviceofnewsystems.
Figure2:Arista7500Series10yearsofcontinuousinnovationsandexperienceSingleOSacrosstheentireportfolioaddressesthechallengeofqualifyingnewproducts,whereexistingtools,processesandfeaturesworkasexpected.
Thisisincomparisontothechallengeofintroducinganallnewproductwithadifferentoperatingsystemanddissimilarfeatures.
FeatureRichbuildingonthecodebaseofEOSandleveraging1000'sofsoftwaredevelopmentyearswithcommonplatformsallowsforfasterqualificationwithouthavingtowaitforfeaturestobereleased.
AristaEOSisamodernoperatingsystemandcontinuestoevolve,withongoingrefactoringofmajoraspectsofthecode,addingscale,newfeaturesandsupportfornewsilicon,allwithoutsacrificingsupportforexistingfeaturesandcodequality.
ConsistentVoQnon-blockingfairdeliveryTheAristaR3-Seriesleveragesadeepbuffervirtualoutputqueue(VOQ)architecturewithacellbasedfabricthatis100%efficientandprovidesconsistentlylowlatencyandjitter,eliminateshead-of-line(HOL)blockingandvirtuallyeliminatespacketdropseveninthemostcongestednetworkscenarios.
Anadvancedtrafficschedulerfairlyallocatesbandwidthbetweenallvirtualoutputqueueswhileaccuratelyfollowingqueuedisciplinesincludingweightedfairqueuing,fixedpriorityorhybridschemes.
Asaresult,theArista7280R3,7500R3and7800R3Seriescanhandlethemostdemandingdatacenterrequirementswithease,includingmixedtrafficloadsofreal-time,multicast,andstoragetrafficwhilestilldeliveringlowlatency.
TheadvantagesoftheVoQarchitecturearethattrafficisnotforwardedtoadestinationwithoutacredit,andthatthecellbasedfabricusesallpathstotheegresspacketprocessortoavoidcreatinghot-spots.
Thecellbasedfabricleveragesavariablecellsizeandhandlesmixedinterfacespeedswithnoimpactonsystemperformance.
Figure3:CellBasedFabricandIngressVoQArchitectureTable1:RealInvestmentProtectionCharacteristicWhyitMattersH/WInvestmentProtectionCompleteinteroperability3rd/4thgenerationExistinglinecards,fabricmodules,chassis&interfacesworkas-isTrueHardwareInvestmentProtectionSignificantCapExsavingsNoforkliftupgradeCommonArchitectureDeepbufferIngressVoQCellbasedFabric100%efficientHandleunpredictableworkloadsNohotspotsOneOSSingleOSacrossentireportfolioReducesOpExExistingtools&processesworkas-isProvenPlatformFeature-richdayoneHighqualityondayoneLeverage1000sofSWdevelopmentyearsFasterqualificationEnhancementsfortheR3-Seriesincludemulticastexpansionintheingress,fabricoregress,toreducedropsorqueueing.
Intheeventofadrop,itcanbemirroredtoadestinationforanalysis.
DeeppacketbufferswithHBM2TheR3-SeriesisconsistentwithpreviousgenerationsoftheR-Serieswithflexiblepacketbufferscombiningdedicatedon-chipresourceswithhighbandwidthaccessto8GBofHBM2memoryperpacketprocessor.
Thebufferaccessspeedhasincreasedatafactoroffiveoverthepreviousgeneration,usingadvancedmemorytechnologycommonlyfoundingraphicsandAIsystems,alignedtoa5Xincreaseinthepacketprocessorperformance.
HighBandwidthMemory-2,orHBM2,whenintegratedtothepacketprocessorpackage,enablesbothhighbandwidthandgreatercapacity,requiring43%lesspowerthantheequivalentGDDRmemory.
IntegratedHBM2memoryallowstheR3-Seriestodeliverahighnumberofinterfaces,increasessystemreliability,andrequiressubstantiallylesspowerthancomparablehighspeedmemoryoptionssuchasHMC,whichlackthepackageintegrationbenefits.
HBM2isavailablefrommultiplesources,isbackedbytheJEDECconsortium(1),incommonuseongraphicsprocessorsandhasatechnologyroadmaptohigherbandwidthandperformance.
The32MBofon-chipbuffersareusedfornon-congestedforwardingwithseamlessexpansiontotheHBM2basedoninstantaneousorsustainedcongestionwithnopacketdrops.
BuffersareassignedperVoQandclassofservice,withnorequirementtotuneforoptimumbehavior.
MemorymanagementisasimportantascapacityandintheR3-Seriesthepacketbuffermemoryisoptimizedformodernnetworks,thatarenaturallybursty.
Avoidingwastefromsmallframes,memoryisallocatedin4Kbytesegmentstoaccommodatemultipleframes,whicharequeueddirectlyfrommemorytothedestinationpacketprocessor,andarenotreturnedviatheOCB.
PowerEfficient100Gand400GEthernetTheR3-Seriesintroducessupportforhighdensity100Gand400GEthernetinavarietyofplatformsfrom24portsof400Gin1Utoamodularsystemwith576ports,andincreaseperformanceby3Xoverthepreviousgeneration.
Figure4:HBM2andPacketProcessorPackagingFigure5:SeamlesspacketbufferoptimizationwithOCBandHMB2Asingleportof400Gcansupport1x400G,2x200Gor4x100Gmodeswithupto36portsinasinglelinecard,equivalentto144portsof100G.
Whiledemandfor400GisdrivenbythelargestscalenetworkstheR3-Series100Gsystemsprovidehigherbandwidthat35%lowerpowerpergigabit,enablingtherapidadoptionofnextgenerationsystems.
Tobridgethemigrationfrom100Gto400GtheR3-Series400Gand100Gsystemsprovideflexibleportspeedoptionsandmediatypeswithbreakoutcablesandparalleloptics.
FlexRouteImprovementsInternetRouteScaleAllR3-Seriesproductssupportfullinternetroutescalewithacapacityof1.
3MroutesleveragingacommonsetofforwardingtableresourcesfortheL2andL3tables.
TheR3K-Serieswithexpandedsizetablescansupportupto2.
5MIPv4andIPv6routes,allowingformultipleoverlappingfulltablesorformanyyearsofgrowthoftheInternet.
Routesareheldonthepacketprocessor,withouttheneedforexternalTCAMsoradditionalmemory,affectingperformance,powerefficiencyoroverallsystemdensity.
GeoffHuston,ChiefScientistatAPNIC,theAsiaPacificRegionalInternetRegistryhasbeenprovidingresearch,analysisandcommentaryontheglobalinternetroutingtableforoveradecade.
Thepublishedprojectionforthenext5yearsisshownbelow.
AsIPv6addressestakemorespacethanasingleIPv4it'slikelythatcapacityfor1.
3Mrouteswillberequiredtoholdafulltable,inthenearfuture,andthat2.
5Misrequiredforprojectedgrowthbeyond2024.
Source:https://labs.
apnic.
net/p=1195Figure6:ImprovedpowerperbandwidthTable2:HistoricgrowthofIPv4andIPv6announcementsMetricJan2016(Actual)Jan2017(Actual)Jan2018(Actual)Jan2019(Actual)Jan2020(Prediction)Jan2021(Prediction)Jan2022(Prediction)Jan2023(Prediction)Jan2024(Prediction)IPv4Prefixes587,000646,000699,000755,000810,000864,000919,000974,0001,028,000(%Increase)9%8%7%7%6%6%6%5%IPv6Prefixes(L)27,00037,00045,00062,00075,00089,000102,000116,000130,000(%Increase)27%18%27%17%16%13%12%11%IPv6Prefixes(E)27,00037,00045,00062,00083,000109,000145,000192,000255,000(%Increase)27%18%27%25%24%25%24%25%Total(linearIPv6)614,000683,000744,000817,000885,000953,0001,021,0001,090,0001,158,000(%Increase)6%6%6%6%6%6%6%6%Total(exponentialIPv6)614,000683,000744,000817,000893,000973,0001,064,0001,166,0001,283,000(%Increase)10%8%9%9%8%9%9%9%NextGenerationEdgewithEVPNandSegmentRoutingTheR3-SeriesexpandsonSegmentRoutingandEVPNsupportwithmoreflexibleoptionsforunderlayandoverlaytopologies.
Theysupportexpandedlabeldepth,newtunneltypesincludingv6GREandVXLAN-v6andimprovedloadbalancingondeeptunneledpackets.
Asmultipletechnologytransitionsevolveatthenetworkedgee.
g.
multicloud,workloadmovementbetweenprivateandpubliccloud,contentcachesnearcustomerproximityanddistributedNFVservicesattheedge,connectivityandsegmentationarekeytoagileservicedelivery.
TheR3-Seriesprovidestheflexibilitytodelivercostoptimizedandhighdensityinterconnectsolutions,toaccelerateNFVedgeserviceswithEVPN,andSegmentRoutingsolutionsfortheWAN(2).
ProgrammabilityProgrammableProcessingwithElasticPipelineTheR3-Seriesdeliversadvancedprogrammablepacketprocessingwithanewelasticpipeline.
Complexfeaturesrequiringadditionallookupcyclesareimplementedthroughapoolofstagesinthepipeline.
Existingadvancedfeatures,includingAcceleratedsFlow,areimplementedintheprogrammablepipelineavoidingexternaloffloadengines,andincreasingflexibility.
FlexiblePacketProcessorTablesInlargescalenetworksthetotalscalabilityisboundbyoneormoreresources:Routes,Hosts,MACaddressesorMPLSLabels.
Intheseenvironmentstheabilitytoincreasethecapacitybyflexiblyreallocatingtablesallowsformoreefficientdesigns.
TheR3-Seriesleverageacommoncentraldatabaseofforwardingandlookupresources,availabletoallingressandegressstagesforthecommonforwardingentries.
Theseresourcesareallocatedusingforwardingprofilesthatensuretheoptimalallocationtodifferenttables,forawiderangeofuse-cases.
TheL3optimizedprofileexpandstheroutingandnext-hoptablestoaddressnetworkswherelargerroutetablecapacityisrequired,whiletheBalancedprofileallocatestheresourcesacrossalltablesandissuitedforleafandspinedatacenterapplications.
TheR3K-Seriessystemsandlinecardsincludesupportfordoublethecapacityofforwardingtablesforenvironmentsthatrequireresourcestobeexpandedfurther.
TheR3Kpacketprocessorshave2xlargerintegratedforwardingtables,andtheprofilesoptimizeforPeering,DCIandIP/MPLSapplicationswithupto2.
5Mroutesor2.
8MMPLSLabels.
Figure7:NextGenerationTelcoEdgewithEVPNandSegmentRoutingFigure8:R3-SeriesPacketProcessorFlexibleTablesAdvancedNetworkTelemetryAcceleratedsFlowwith1:100at400GTheR-SeriessupportforAcceleratedsFlowenabledsupportforhighratesamplingacrossasystemwithdedicatedsFlowengines.
TheR3-SeriesexpandsAcceleratedsFlowtobesupportedonallsystems,witheachpacketprocessorleveragingtheprogrammablepipelineforhighsamplingratesat400Gand100Gonallports,withtheadditionalmetricsofAS-PathandNext-Hop.
Telemetryinformationontrafficflowsisvaluablewhendeterminingthesourcesofrogueandmisbehavingapplicationswithinthenetwork,togetherwithflowanalysisandtrafficengineeringofpaths.
In-bandNetworkTelemetryforflowandpacketforensicsIn-bandnetworktelemetry,orINT,isastandardsapproachtoprovidingdeepvisibilityintotrafficinreal-time,withnoimpactonswitchperformance.
INTisavailableontheR3-Seriesandprovidesper-flowmonitoringofdrops,latency,congestionandcanreconstructthenetworktopologyandpaths.
INTinformationcanbeexportedinIPFIXorsFlowformatstoamanagementsystemorcollectorsuchasAristaCloudVision,forpredictiveanalyticsanddeepforensics.
Figure9:R3K-SeriesPacketProcessorFlexibleTablesFigure10:AcceleratedsFlowwithstatestreamingFigure11:InbandNetworkTelemetryinActionCopyright2019AristaNetworks,Inc.
Allrightsreserved.
CloudVision,andEOSareregisteredtrademarksandAristaNetworksisatrademarkofAristaNetworks,Inc.
Allothercompanynamesaretrademarksoftheirrespectiveholders.
Informationinthisdocumentissubjecttochangewithoutnotice.
Certainfeaturesmaynotyetbeavailable.
AristaNetworks,Inc.
assumesnoresponsibilityforanyerrorsthatmayappearinthisdocument.
April6,201902-0082-02AlgorithmicACLsTheR3-seriesprovidesupportforalgorithmicACLsonallsystemsthatincreaseboththescaleandflexibilityleveragingtheprogrammablepipeline.
Withsupportforover100KrulestheR3-SeriesaresuitedforinternetedgeandcybersecurityroleswithstatefulACLsandrichcountersidentifyingandfilteringwithnolossinperformance.
SummaryTheAristaR3-Seriesexpandstheindustryleadingspineandleafsystems,preservinginvestmentwithaconsistentarchitecture,EOSfeaturesandtheintroductionofsupportfor400Galongwithnewinnovationsforprogrammability,telemetry,routingandsecurity.
(1)https://www.
jedec.
org/standards-documents/docs/jesd235a(2)EANTCTestreportsvalidatingEVPNandSegmentRoutingfeatures(3)APNIChttps://labs.
apnic.
net/p=1195Figure12:AlgorithmicACLscombinesoftwareandhardwareinnovation
- Singleapnic相关文档
- "IPv4AddressAllocationsbyCountry"
- Miseapnic
- 中国apnic
- singleapnic
- probeapnic
- authenticatedapnic
阿里云金秋上云季,云服务器秒杀2C2G5M年付60元起
阿里云(aliyun)在这个月又推出了一个金秋上云季活动，到9月30日前，每天两场秒杀活动，包括轻量应用服务器、云服务器、云数据库、短信包、存储包、CDN流量包等等产品，其中Aliyun轻量云服务器最低60元/年起，还可以99元续费3次！活动针对新用户和没有购买过他们的产品的老用户均可参与，每人限购1件。关于阿里云不用多说了，国内首屈一指的云服务器商家，无论建站还是学习都是相当靠谱的。活动地址：h...
DiyVM独立服务器：香港沙田服务器,5M带宽CN2线路,L5630*2/16G内存/120G SSD硬盘,499元/月
diyvm怎么样?diyvm商家VPS主机均2GB内存起步，三个地区机房可选，使用优惠码后每月69元起;DiyVM独立服务器开设在香港沙田电信机房，CN2线路，5M带宽，自动化开通上架，最低499元/月，配置是L5630*2/16G内存/120G SSD硬盘。DiyVM是一家成立于2009年的国人主机商，提供的产品包括VPS主机、独立服务器租用等，产品数据中心包括中国香港、日本大阪和美国洛杉矶等，...
易探云美国云服务器评测，主机低至33元/月,336元/年
美国服务器哪家平台好?美国服务器无需备案，即开即用，上线快。美国服务器多数带防御，且有时候项目运营的时候，防御能力是用户考虑的重点，特别是网站容易受到攻击的行业。现在有那么多美国一年服务器，哪家的美国云服务器好呢？美国服务器用哪家好?这里推荐易探云，有美国BGP、美国CN2、美国高防、美国GIA等云服务器，线路优化的不错。易探云刚好就是做香港及美国云服务器的主要商家之一，我们来看一下易探云美国云服...
apnic为你推荐
-
虚拟主机服务器虚拟主机和站点服务器什么区别美国vps主机美国VPS好？还是香港VPS好？台湾vps虚拟主机，VPS，服务器，其中哪个流量最大？虚拟空间哪个好国内哪个空间商（虚拟主机）最好虚拟空间哪个好虚拟空间哪个好网站空间免备案哪有不用备案的网站空间？香港虚拟主机香港虚拟主机多少钱一年呢？上海虚拟主机上海虚拟主机哪家好啊？北京虚拟主机北京的虚拟主机提供商哪个经济实惠？云南虚拟主机云南虚拟主机，公司网站用本地客户，云南数据港怎么样？