基于事件驱动架构的实时企业(Richard)

12[Problem]LotsmoredatathaneverbeforeTwitter:90milliontweetsperday,8terabytesperdayOnacrosscountryflight,thesensorsinaBoeing737generateabout240terabytesofdataRFIDhasthepotentialtogenerate100to1000timesthedatavolumeofconventionalbarcodetechnology.In2010,morethan1200exabytesofdigitalinfowascreated.SingleExabyteisequaltoabout1trillionbooks.datacreatedbyindividuals,butlotsfromincreasing#ofsensorsaswellGartnersaysenterprisedatainallformswillgrow650%overthenextfiveyearswithIDCclaimingentireworld’sdatadoublesevery18months30%ofbizleaderssaytheycan’tgetthedatatheyneedatthespeedtheyneedit,and61%wantfasteraccess1in3wantMOREdataToomuchdataoverwhelmingourcomputingcapabilityandnewmodelneededHowdoweknowmore,faster?;Continuousintelligence3[Wherewewanttobe]EDAbetweensystems,inanalytics,etcAsopposedtoETL,polling-basedsolutions,point-to-pointDatainmotionvs.dataatrestUseeveninIS-centricsolutions(e.g.monitoring)MonitorSLAsinrealtimeandexceptionsfromthenormWelldesignedEDAapplicationisanattentionamplifier**acriticalandincreasinglyscareresourceisinterruptedtime**Gettingjustalittlebitoftherightinfojustaheadofwhenit’sneededisalotmorevaluablethanalltheinfointheworldamonthordaylaterNewcompetitiveadvantagewillbeanabilitytoanticipateeventsbasedoninfoaboutwhat’shappeningrightnowSometimesyoucannotactquickly,soMUSTbepredictive(thinkmakingaproductwithlongleadtime)4[KeyPoint1]•Improvesreactiontime•reducesbizprocessduration•improvesavailabilityofinfo(dataconsistency,situationalawareness)•Valueofeventprocessingliesinrecognizingthesignificanceofaneventfromabusinesscontext,andidentifyingtherightresponsestoassociatewiththatevent**eventdrivengoodforunpredictablefactors,situationsandtiming**;timedrivenbehavioroccurswhennatureandtimingofeventcanbeplannedinadvancewhilerequest-drivenisappropriatewhennatureofactivityisunderstoodandagreedto,buttimingisnotpredictableUseEDAforappsthatmustrespondquicklytosituationsthatchangerapidlyandasyncandwhereinteractionsdoNOTneedtobetransactional5Eventcanbereceivingcustomerorder,makingbankpayment,changingcustomeraddress,hiringanemployee,detectingattemptedfraud,changeincompetitorprice,detectingnon-compliance,durationofbusinessprocessWhatisanevent?aneventisachangeinthestateofacomponentofanenterpriseoritsenvironmentStatechange:RFIDtagHappening:war,buyahouse(typicallyacomplexevent)Detectablecondition:GPSmessageoftruckatstandstill**absenceofmessagesconveysinformationineventdriveninteraction**Now,everyEVENT(notjusttransactions)canbecomeabitofdigitalinformationLogintosite,abandoncartDatabasescan’tprocessnon-events.IfsomethingDIDN’Thappen,thedatadoesn’tgetcreated6Continuousprocessingsystemshavebenefitofdistributingworkmoreevenlytoeliminatebottlenecks,peaksandtroughsEDAisanarchstyleinwhichoneormoreofthecomponentsinasoftwaresystemareeventdrivenandminimallycoupled(onewaytransferofeventobjects)EDAif(firstthreetimeliness,lasttwodistribution):reportcurrenteventsashappen(couldbebundledforefficiency)pushesnotifications(eventproducerdecides)respondsimmediatelyoneway(fireandforget)freeofcommands(areport,notacommand)7Casesinwhichasingleeventwillnottriggeranactionperformedbyaconsumer,butinsteadtheactionistriggeredbyacomplexcompositionofeventshappeningatdifferenttimesandindifferentcontextsCEPisawayofdistillingtheinformationvaluefromanumberofsimplebusinesseventsintoafewmoreuseful,summarylevelcomplexevents**Termcomplexmaybeoff-putting,butinreality,complexeventsactuallysimplifydatabysummarizingandabstractingwhatishappeningSiftthroughcountlessindividualevents(e.g.calls)orprovidessums/averagesManualCEPisoftentimedriven(whendoyouacquiredata)andreliesonlowdatavolumes,availablepeopletodoanalysisandslowformingdecisionsDifffromtraditionaldatabasedrivenapplicationsQueryismoreorlesspermanent,dataisephemeral,8continuouslyarrivingandthendisappearingasitbecomesoutofdateRatherthananapplicationrepeatedlycompilingaquery,submittingittoadatabaseandwaitingforaresult,applicationsusingCEPsubmitaqueryonce.Databaseshavemajorhandicaps.Inherentlyfocusedonthepast.Analyzewhathasalreadyhappened,notpredictwhat’sabouttohappen.CEPisusefulwhencharacterizationofanomaliesarecomplexandwhendetectionrequiresanalysisgatheredfrommultipleagentsacrosstimeWanttechthatcanconstantlywatchevents,runthroughpredictivemodelbuiltonmemoriesofpastpatterns,andcontinuouslylookforthenextmove8ThemiddlewareoftenembodiestheeventcloudEventsources:middleware,databases,devices,agents,businessprocesses9•Eventsarejustdata,notoperationstoo•Nothingspecialaboutdatainevent;uniquenatureofeventprocessingderivesfromWAYinwhichdataisexchangedandprocessed(5principles)•Whatdefinesanevent;whatgoesinthebody•Tagindicatingtypeofevent,uniqueID,creationtimestamp,start/endtimestamps,producer,priority•Before/after(idempotence)•Maysupporta“callback”forthereceivertogogetthechangeddata•Eventsmayhaveadurationcomponent(life)**Eventisnotjustadatarecordwithatimestamp**StaticviewofeventsusedprimarilyforarchivingandposteventprocessinglikeBIDataDERIVEDfromeventsDatareallythei