Elementsoracle数据库视频

OracleDatabase:BenchmarkingaTrillionEdgesRDFGraphOracleWhitePaperJune,2020Copyright2020,Oracleand/oritsaffiliatesPublic2WHITEPAPER|OracleDatabase:BenchmarkingATrillionEdgesRDFGraphCopyright2020,Oracleand/oritsaffiliates|PublicPURPOSESTATEMENTThisdocumentprovidesanoverviewofhowOracleconductedaonetrillionedgesLUBMbenchmark(LUBM4400k)inSeptember2014withRDFGraphinOracleDatabaseonanOracleExadataDatabaseMachine.
DISCLAIMERThisdocumentinanyform,softwareorprintedmatter,containsproprietaryinformationthatistheexclusivepropertyofOracle.
YouraccesstoanduseofthisconfidentialmaterialissubjecttothetermsandconditionsofyourOraclesoftwarelicenseandserviceagreement,whichhasbeenexecutedandwithwhichyouagreetocomply.
Thisdocumentandinformationcontainedhereinmaynotbedisclosed,copied,reproducedordistributedtoanyoneoutsideOraclewithoutpriorwrittenconsentofOracle.
ThisdocumentisnotpartofyourlicenseagreementnorcanitbeincorporatedintoanycontractualagreementwithOracleoritssubsidiariesoraffiliates.
Thisdocumentisforinformationalpurposesonlyandisintendedsolelytoassistyouinplanningfortheimplementationandupgradeoftheproductfeaturesdescribed.
Itisnotacommitmenttodeliveranymaterial,code,orfunctionality,andshouldnotberelieduponinmakingpurchasingdecisions.
Thedevelopment,release,andtimingofanyfeaturesorfunctionalitydescribedinthisdocumentremainsatthesolediscretionofOracle.
Duetothenatureoftheproductarchitecture,itmaynotbepossibletosafelyincludeallfeaturesdescribedinthisdocumentwithoutriskingsignificantdestabilizationofthecode.
3WHITEPAPER|OracleDatabase:BenchmarkingATrillionEdgesRDFGraphCopyright2020,Oracleand/oritsaffiliates|PublicTABLEOFCONTENTSPurposeStatement2Disclaimer2Introduction4ATrillionEdgesRDFGraphBenchmarkonOracleDatabase5TheResults5TheConfiguration5BestPracticesUsed6Databasesettings:6Tuning:6Conclusion64WHITEPAPER|OracleDatabase:BenchmarkingATrillionEdgesRDFGraphCopyright2020,Oracleand/oritsaffiliates|PublicINTRODUCTIONOnetrillionisareallybignumber.
Whatcouldyoustorewithonetrillionfacts1000tweetsforeveryoneofthe1BillionTwitterusers.
770factsabouteveryoneofthe1.
3BillionFacebookusers.
10factsfrom107Billionsensors,locatedsomewhereontheplanet.
400metabolicreadingsforeachofthe2.
5Billionheartbeatsoveranaveragehumanlifetime.
12factsabouteveryoneofthe86Billionneuronsinthehumanbrain.
5factsabouteveryoneofthe200BillionstarsintheMilkyWayGalaxy.
7factsabouteveryoneofthe150Billiongalaxiesintheuniverse.
6,350factsabouteachofthe158MillionbooksintheLibraryofCongress,thelargestintheworld.
10factsabouteachofthe107BillionpeoplewhoeverlivedResourceDescriptionFramework(RDF)graphsandtheanalyticstheypermitarebecomingcentraltobigdataapplicationsforsocialnetworksandlinkeddata.
Theseapplicationsareoftenfoundinpublicsector,healthcareandlifesciences,finance,media,andintelligencecommunities.
TheWorldWideWebConsortium(W3C)1definesRDFandtheWebOntologyLanguage(OWL)graphstandardsforrepresentinganddefiningsemanticdataandrules,andSPARQL,apatternmatchingquerylanguagedesignedspecificallyforgraphanalysis.
ThebasicnatureofanRDFgraphfacilitatesidentification,integration,anddiscovery:RDFdataelementsaregloballyunique.
TheyaredefinedusingUniformResourceIdentifiers(URIs)thatenableaconsistentmetadatalayerforintegrationofdisparatedatasources.
RDFdataelementsarelinkedtoformagraph.
Elementsareusedtomakestatementsintheformofsubject-predicate-objecttriples.
Predicates(edges)linkthesubjectandobject(nodes)andcandescribeanyrelationshiporproperty.
Theobjectcanbeanothersubjecttolinktriplestogethertoformagraphoraliteralthatisanattributeofthesubject.
Thetriplescanbefurtherqualifiedwithafourthnamedgraphcomponent,whicharereferredtoasRDFquads.
TheRDFmodelallowseasy,dynamicschemaevolution.
Addinganewschemaelementisaseasyasinsertingatriplewithanewpredicate.
RDFandSPARQLsupportadhocqueries.
Queriesmaynotbeknownwhentheschemaisdesigned.
TheRDFmodelmakesanOpenWorldAssumptionthatcanfacilitatediscovery.
Itassumesthatwhatisunknownisundefined,ratherthanfalse,asisthecasewithrelationaltechnology.
Italsohastechnologiesthathelpdiscovermissingresults.
RDFembedssemantics(meaning)directlyinthedata.
Entitiesarecategorizedwithclasses,predicatesarepropertiesorrelationships,andtheyareallpartofthedata,unlikecolumnheaders,foreignkeys,orconstraintsinrelationaldata.
RDFsupportsmachine-driveninferencingfordiscovery.
TheOWLsemanticlanguageandrulesusedtodefinethepredicatesintriplesarebasedonformalDescriptionLogicsthatenableautomaticdiscovery,suchasidentifying"same-as"relationshipsbetweendifferenttermswiththesamemeaningintwoapplications.
Thesetofinferredtriples(conclusionsthatcanbedrawn)isreferredtoasanentailment.
TheOWLlanguagecanunifyanenterprise'sdictionaries,vocabularies,andtaxonomies.
Allofthetermsusedbytheapplicationsinanenterprisecanberelatedtoeachotherandformconcepts.
Conceptsaremanagedasoneormoredomain-specificontologiesandstoredinRDFgraphs.
Ontologiesarelinkedtotheassertedinstancedataingraphsandusedforinferencingandquerying.
Thisisanothercapabilitythatfacilitatescreatingaconsistentmetadatalayerfordataintegration.
1http://www.
w3.
org/RDF/5WHITEPAPER|OracleDatabase:BenchmarkingATrillionEdgesRDFGraphCopyright2020,Oracleand/oritsaffiliates|PublicTheLehighUniversityBenchmark2(LUBM)isadefactoindustrystandardbenchmarkforevaluatingRDFgraphstoreproductperformance.
ItisusedbyRDFgraphstorevendorstocharacterizetheload,inference,andqueryperformanceoftheirproduct.
VendorspostresultsontheW3CLargeTripleStorespage3.
EndusersuseLUBMbenchmarkresultsaspartoftheirevaluationofanRDFGraphstoreproduct.
ThebenchmarkincludesaW3COWL-baseduniversityontology,adatageneratortocreateagraphofanysize,andfourteentestqueries.
OracleconductedaonetrillionedgesLUBMbenchmark(LUBM4400k)inSeptember2014withOracleDatabase12.
1.
0.
1standardinstallationonExadataDatabaseMachineandachievedtworecord-settingaccomplishments:OraclebelievesitsbenchmarkisthelargestcompleteLUBMbenchmarkintheindustrytodate.
Thecombinedload,inference,andqueryresultsarethefastestRDFgraphperformancenumbersreported;thisisespeciallysignificantforabenchmarkofthisscaleandcomplexity.
Thedetailsforthisbenchmark,includingresults,configuration,andbestpracticesarediscussedinthenextsectionofthispaper.
ATRILLIONEDGESRDFGRAPHBENCHMARKONORACLEDATABASEAsbigdatagraphsgrowfrombillionstotrillionsofrelationshipsitbecomesincreasinglyimportanttocharacterizeproductperformance.
OracleconductedanRDFgraphLUBM4400kbenchmark.
Itinvolvedloading,inferencing,andqueryingoveronetrillionedgeswithRDFGraphinOracleDatabaseonanOracleExadataDatabaseMachine.
TheLUBMenvironmentwasusedtogeneratedataaboutuniversitiesandtheirdepartments.
Thedatawascreatedandorderedinto4.
4millionnamedgraphsbyexpandingthetriplesintoquads.
Therewasonenamedgraphperuniversity.
Theoverallgraphincluded605.
4billionuniqueassertedquadsandanentailmentofanother475.
6+billionquads.
TheResultsTheRDFGraphLUBM4400kbenchmarkonOracleDatabaseachievedthefollowingresults:DataLoadingPerformance:1.
420millionQuadsLoadedandIndexedperSecond.
605.
4BillionQuadswereloadedandtwoindexeswerecreatedin115.
2hours.
Note:GraphloadinginOracleDatabaseisuniqueintheindustryforcheckingthatquadsarewellformedandforremovingduplicates.
InferencePerformance:1.
527millionTriplesInferredandIndexedperSecond.
475.
6BillionTriplesandtwoindexeswerecreatedin86.
5hours.
SPARQLQueryPerformance:1.
130MillionQueryResultsperSecond.
92.
5BillionAnswersweregeneratedin22.
5hours.
ATrilliontriplesgraphAssertedInferredTotalAnswers605.
4BillionQuads475.
6BillionTriples1.
081TrillionQuads92.
5BillionTheConfigurationThemarket-leadingperformanceofthisbenchmarkwasduetothecombinationofthenativeRDFgraphstorecapabilitiesofRDFGraphinOracleDatabaseonthebalancedconfigurationofanOracleExadataDatabaseMachineX4-2.
TheuniquecapabilitiesoftheExadataDatabaseMachinethatassistedbenchmarkqueryperformanceinclude:SmartScanthatreducesdatamovementbetweenstorageservers(cells)anddatabaseserverbypushingqueriesdowntothestoragecell,storageindexesusedbythestoragecelltoreadonlyregionsofstoragethathaverelevantdata,andInfiniBandfabricthatprovidesfasttransfer(40Gb/second)ofrelevantbytesbacktothedatabaseservertocompletetheexecutionofaquery.
2http://swat.
cse.
lehigh.
edu/projects/lubm/3https://www.
w3.
org/wiki/LargeTripleStores6WHITEPAPER|OracleDatabase:BenchmarkingATrillionEdgesRDFGraphCopyright2020,Oracleand/oritsaffiliates|PublicTheOracleExadataDatabaseMachineX4-2Highcapacityfullrackwasconfiguredasfollows:8databasenodesand14storagenodesforatotalof168CPUcores2TBtotalRAMand44.
8TBFlashCacheZS3-2storagewith2controllersand8traysofdisksSoftware:OracleDatabase12.
1.
0.
1standardinstallationonExadataDatabaseMachine.
BestPracticesUsedThebestpracticesfallintotwocategories,databasesettingsandtuning.
Databasesettings:SGA_TARGET=132GBPGA_AGGREGATE_TARGET=100GOpencursors=1000Processes=100032KblocksizegiventoallgraphtablespacesaTEMPgroupcreatedwith3bigfiletablespacesUseoftheauto-allocateoptionforallocationoftablespaceextentscoupledwithalarge,8millionbytesextentsize.
ThisreducedthenumberofwaitscausedbyHVenqueuecontention;thatis,waitsonalockthatisusedtoalterthehigh-watermarkinatablespace.
Asaresult,contentionamongmultipleprocessesrequestingtablespaceexpansioncouldbeavoided.
DOPsettings(296,256,192)forautomaticdegreesofparallelismusedinloading,inferencing,andquerying.
UseofadditionalcompressionbeyondbasictablecompressionduringinferencingprovidedbytheHybridColumnarCompressionfeatureofOracleExadataDatabaseMachine.
Tuning:OracleEnterpriseManagerprovidedspecificperformanceinsightsintooperationsfortuning.
ThemethodologyusedisdocumentedintheOracleDatabasePerformanceTuningGuide.
4CONCLUSIONRDFgraphsprovideunique,standards-based,bigdatacapabilitiesformetadataintegration,anddiscoverytosupportsocialnetworksandlinkeddataapplicationsinavarietyofindustries.
RDFGraphdemonstratedindustry-leadingscalabilityandperformanceforloading,inference,andqueryingaonetrillionedgesRDFgraphmanagedinOracleDatabase.
TheLUBM4400kRDFgraphbenchmarkbenefitedfromthebalancedhardwareconfigurationofanOracleExadataDatabaseMachineX4-2.
Thebestpracticessettingsusedtoachievethesebenchmarkresultsarealsogenerallyapplicabletoreal-worldapplicationsonOracleExadataDatabaseMachineandotherbalancedhardwareconfigurations.
4http://docs.
oracle.
com/database/121/TGDBA/toc.
htm7WHITEPAPER|OracleDatabase:BenchmarkingATrillionEdgesRDFGraphCopyright2020,Oracleand/oritsaffiliates|PublicCONNECTWITHUSCall+1.
800.
ORACLE1orvisitoracle.
com.
OutsideNorthAmerica,findyourlocalofficeatoracle.
com/contact.
blogs.
oracle.
comfacebook.
com/oracletwitter.
com/oracleCopyright2020,Oracleand/oritsaffiliates.
Allrightsreserved.
Thisdocumentisprovidedforinformationpurposesonly,andthecontentshereofaresubjecttochangewithoutnotice.
Thisdocumentisnotwarrantedtobeerror-free,norsubjecttoanyotherwarrantiesorconditions,whetherexpressedorallyorimpliedinlaw,includingimpliedwarrantiesandconditionsofmerchantabilityorfitnessforaparticularpurpose.
Wespecificallydisclaimanyliabilitywithrespecttothisdocument,andnocontractualobligationsareformedeitherdirectlyorindirectlybythisdocument.
Thisdocumentmaynotbereproducedortransmittedinanyformorbyanymeans,electronicormechanical,foranypurpose,withoutourpriorwrittenpermission.
ThisdevicehasnotbeenauthorizedasrequiredbytherulesoftheFederalCommunicationsCommission.
Thisdeviceisnot,andmaynotbe,offeredforsaleorlease,orsoldorleased,untilauthorizationisobtained.
OracleandJavaareregisteredtrademarksofOracleand/oritsaffiliates.
Othernamesmaybetrademarksoftheirrespectiveowners.
IntelandIntelXeonaretrademarksorregisteredtrademarksofIntelCorporation.
AllSPARCtrademarksareusedunderlicenseandaretrademarksorregisteredtrademarksofSPARCInternational,Inc.
AMD,Opteron,theAMDlogo,andtheAMDOpteronlogoaretrademarksorregisteredtrademarksofAdvancedMicroDevices.
UNIXisaregisteredtrademarkofTheOpenGroup.
0120OracleDatabase:BenchmarkingaTrillionEdgesRDFGraphJune2020