外文翻译-CAN总线系统的实时对象调用

0外文原文InvocationofReal-TimeObjectsinaCANBus-SystemAbstractThepaperfocusesonmethodinvocationofrealtimeobjectsinaCANbaseddistributedrealtimesystem.Asimpleobjectmodelisintroduced,whichallowstheconvenientmodellingofhardwareandsoftwarecomponents.Relatedtotheobjectmodel,twoissuesarediscussed.Firstly,amodelisintroducedwhichallowstoformandaddressobjectgroups.Thisreflectsabasicneedinarealtimesystemtodistributeinformationtomultipleclientsefficiently.Secondly,wediscussanapproachtoexpresstimingrequirementsforobjectinvocations.Toachievedistributedconsensusoncommunicationresourceaccess,anEDFlikeapproachisintroduced,whichtakesadvantageofknowledgeaboutdeadlines,thenumberofremainingcommunicationactivities,andtheremainingworstcaseexecutiontimefortheinvokedmethodateachpointoftime.1.IntroductionFuturecomputersystemswill,toalargeextent,monitorandcontrolrealworldprocesses.Thisresultsinaninevitabledemandfortimelinessandreliability.Distributedsystemswhichinherentlyprovideimmunityagainstsinglefailuresareanadequatearchitecturetomeetthegoalofreliability.Additionally,becauserealworldapplicationsoftenrequirethespatiallydistributedcontrolofelectromechanicalcomponents,adistributedsystemarchitecturebringscomputingpowertothepointswhereitisneeded.Theavailabilityofinexpensiveyetpowerfulmicrocontrollerssupportsadistributedsolution.Thisresultsinasystemmodelwhichiscomposedfromsmartcomponentsimplementingtheinstrumentationinterfacetotherealworldcomprisingsensorsandactuators.Aconvenientwaytomodelsuchanenvironmentistouseanobjectbasedapproachwhereobjectsencapsulateallkindsofentitiesnecessarytocontrolaphysicalprocess.1Wecanidentifyatleasttwolayersinsystemsinteractingwiththerealworld.Alowerresponsivesystemleveltightlycontrollingthesensor/actuatorinterfacetotheworldandahighersystemlevelresponsibleforinterpretationandevaluationoftheperceptionoftheworldascommunicatedbytheinstrumentationinterface.Inthislayer,higherleveldecisionsareperformed.Althoughthesedifferentlevelshavedifferentrequirementsconcerningresponsiveness,theyaresometimesnotseparatedclearlysimplybecauseofthehardwarearchitecture.Weconcentrateonthelowersystemlevelanddescribeanobjectmodelwhichismotivatedbyasystemstructurewhichreliesonamodulardesignandindependentcomputingresourcesforthecomponentsoftheinstrumentationinterface.Ratherthanhavingacentralpowerfulprocessorinwhichelectromechanicalcomponentsareonlymodelledasobjectsfromwhichcontrolsignalsaregenerated,theelectromechanicalcomponentsareobjectsthemselvespoweredbyinexpensivemicrocontrollers.Theobjectsofthislayerhaveawelldefinedinterfacetobeeasilycontrolledbyahigherlevelinstancewhichcanexploitthisabstractionratherthandealingwithlowlevelcontrolsignals.Thehardwaremodulesareconnectedbyafieldbus.WechoosetheCANbus(CAN:ControllerAreaNetwork),developedbyBOSCH28becauseitprovidesadvancedtechnicalfeaturesandrepresentsanemergingstandardwithawideapplicability.Moreover,popularmicrocontrollersareavailablewithanonchipCANcontroller.CANisasharedbusdesignedtoconnectcontrolsystemsinaspatiallyrestrictedarealikecars,robots,toolmachines,andotherautomotiveorindustrialautomationapplications.Itistargetedtooperateinanoisyenvironmentwithspeedsofupto1Mbit/sec,exchangingsmallrealtimecontrolmessages.WeusetheCANmessageformattouniformlyinvokemethodsonourobjects.Thisallowsatransparentobjectinvocationinthesensethataninvokingobjectmustnotknowwheretheinvokedobjectresidesinthesystem.Additionally,wecanaddressgroupsofobjectswithasingleCANmessage.Asacommonresource,thecommunicationmediumhastobesharedbyallcomputingnodes.Accesstothemediumhastobescheduledinawaythatdistributedcomputationsmeettheirdeadlinesinspiteofcompetitionforthecommunicationline.Sincetheschedulingofthebusc2annotbebasedonlocaldecisions,adistributedconsensusaboutthebusreservationhastobeachieved.Clearly,thisisonlyaspecialcaseofthemoregeneralproblemtoscheduleacooperativedistributedcomputation.Thereexistseveralalternativeapproachestosolvethisproblembasedontheassumptionsaboutthebehaviourofthesystemandtheenvironment.Thefirstapproach,knownastimetriggeredapproach18,21,19assumesacompleteknowledgeofallfutureactionsofthesystem.Hence,duringoperationthesesystemsexhibitminimaloverheadpairedwithamaximumofpredictabilityandarehighlyappropriateforsafetycriticalapplicationswhichcanbemodelledbyaperiodicbehaviour.Amoreadaptableandeconomicwaytomanagesystemresourceshastobebasedonruntimeinformation.Theserealtimesystemshavetoplantheiroperationatruntime,atleasttoacertainextent.Althoughthetradeoffgenerallyisbetweenadaptabilityandpredictability,thesesystemsarescalableinthesensethattheyallowacoexistenceofactivitieswithdifferentrealtimerequirements.Thesesystemsrelyonruntimeguarantees30,11ratherthanonapreplannedfeasibilityguaranteeonly.Aformofcalendarbasedscheduling22,30isanapproachtoruntimeguaranteeswherehardrealtimeactivitieshavethepossibilitytoreserveresourcesinadvance.Thefreeresourcescanbeusedbylessimportanttasks.Itshouldbenotedthatschedulingpolicieslikee.g.earliestdeadlinefirst(EDF),leastlaxityfirst(LLF)orratemonotonicschedulingonlyguaranteeoptimality,i.e.ifascheduleexists,itwillbefound.Butinthedynamicsystem,theremaybeconflictingrequestsforwhichtheseschedulingpolicieswillnotfindasolution.Inthiscasetheyfailtoguaranteeanything.Acalendar,ontheotherhand,isamechanismtoguaranteeresourceavailabilityforreservationswhichhavealreadybeenmade.2.TheSystemModelOursystemmodelisinfluencedbyouranticipatedhardwarebasis.Weassumeanumberofdifferentmicrocontrollerswithdifferentperformanceattributesrangingfrom8Bitto32Bitarchitectures.AllmicrocontrollersareequippedwithaCANBusinterface.Weexploitthe3differentperformanceandpricecharacteristicsofthemicrocontrollerstostructuretheoveralltaskofthesystemintosmallpackages.Sincethesimplemicrocontrollersareintherangeofafew$s,itmustnolongerbearareresource.Inourapplicationexample,asimpleactivesuspensionsystem(Fig.21),weuseasimplemicrocontrollerforeachservoorcomplexsensor.Aservoiscomposedfromamotorandaninternalsensorwhichprovidesfeedbackonthecurrentposition.Thissensor/actuatorsystemisconnectedtotheCANbustocooperatewithothersimilarobjectsandreceivecontrolcommandsthroughCANmessages.Nodirect,lowlevelcontrolsignalsarevisibleoutsidethemodule.Thus,encapsulationandawelldefinedinterfacearesupported.Onthelogicallevel,thesensor/actuatorblockisseenasanobjectonwhichcertainmethodscanbeinvoked,e.g.topositionitselftoaspecificangel,tocommunicatetheangel,etc.Asensor/actuatorconstitutesthelowestlevelinthecontrolhierarchy,comparablewithasimplereflexloopinabiologicalsystem.Highercontrolinstancesareavailabletocontrolgroupsofobjectsoreventuallytheentiresystemusingmorepowerfulmicroprocessors.Thus,thehighercontrolinstanceshaveawelldefinedinstrumentationinterfacewhichiscomposedfromobjects,eachencapsulatingacertainfunctionality.Anobjecthasauniquenameandasetofassociatedoperations.Theuniquenameoftheobjectistranslatedtoashortformsystemnameduringruntimeandmaintainedbyaconfigurationservice.Objectsinsuchanenvironmentmusthavesomeextensionstothesequential,passivemodelknownfromthefieldofprogramminglanguages.Particularlyourmodelincludes:1).Active,autonomousobjects.Sinceobjectsmayhaveadedicatedprocessor,itisstraightforwardtoassumeamodelofautonomous,activeobjects.Anobjectischaracterizedbyanameandalistofmethodswhichcanbeinvokedontheobject.Becausetheobjectisactive,itcanalsoexportinformation.Thisisequivalenttosendinganinvocationtoanotherobjectortoagroupofobjectswithoutapreviousrequest.Asdescribedinchapter3,weprovideatransparentgroupcommunicationmechanism,i.e.anobjectitselfusuallydoesnotknowwhetheritcommunicateswithanobjectonaremoteoronthelocalnode.Thisgroupcommunicationmechanismisabasisforremotemethodinvocationofobjectgroups.Infact,communicationrelationscandynamicallybe4definedduringruntime.So,objectsmayjoinorleaveagroup,andareautonomousindecidingwhentoprocessarequest.TheconceptofTMO(timetriggeredmessagetriggeredobject)16derivedfromtheRTO.K(Real-TimeObject)17conceptenhancestheconventionalobjectmodelbymechanismstospecifythetemporalbehaviourofanobject.Itprovidesaframeworktospecifyacomplexrealtimesystemtogetherwithitsenvironmentinthesameuniformrepresentation.Theaimisanintegrateddesignofadistributedrealtimesystemandsimulatorsgeneratingrealtimeinputmodellingthetargetapplications.ATMOexhibitsactive,spontaneousbehaviouraswellasprovidingservicesrequestedbyamethodcall.ThreeexecutionmodelsarederivedforexecutingTMOS17.Amongthem,theexecutionmodelwhichprovidesadedicatedprocessorforeachobject(TypeII)isverysimilartoourbasicsystemstructure.Thedifferenceontheconceptuallevelisthenotionoftheobjectgroupandthefactthatwedonotrelatespontaneousbehaviourtoclockevents.However,ifthespontaneousactivityofanobjectishardrealtime,wealsohavetomapittosomereservedtimeslot.CHAOS11demonstratesthattheobjectmodelcanbetailoredtomeettheefficiencyrequirementsofrealtimeapplications.Themotivationalsocomesfromaroboticsapplicationwhereelectromechanicalcomponentsareencapsulatedandcontrolledbyobjects.CHAOSprovidesaspectrumofobjectswithdifferentweightrangingfromlightweightpassiveobjectstoobjectsincludingseveralprocessesprovidingschedulingandsynchronizationfacilities.Theconceptistargetedtoratherpowerfulparallelmachines.Theactiveobjectsaretooheavyweighttoimplementthemwithoursimplehardwarebasis.InAlpha,12,objectsarepassiveabstractdatatypesinwhichanumberofprocessescanexecuteconcurrently.MARUTI22providesmodulesinthedesignprocess.However,thesemodulesarebrokendownintoelementaryunits(EUS)whichconstitutetheatomicentitiesofsequentialexecution.Also,EUSresembletoobjects,theyarenotuserdefinedbutderivedduringthecompilationprocess25.2).Objectgroups.Inconventionalobjectorientedlanguagesamethodinvocationissynchronousanddirectedtoasingleobject.Itisusuallynotpossibletoexpressarequesttoagroupofobjects.How5ever,inarealtimecontrolsystemitisbeneficialtoprovidegroupsofobjectsandtouseasynchronousmulticaststoinvokemethodsofthesegroups.Themotivationrangesfromsimpleandfastdistributionofinformation,likesensordataandalarmmessages,toreplicatedobjectsformingagrouptoachievefaulttolerance.Becausethecommunicationmedium,theCANBus,supportsconsistentmulticastsonthelowsystemlevel,itisbeneficialtoexploitthisfeature.ItallowstoaddressagroupofNobjectswithonemessageratherthansendingNindividualmessages.Toacertainextent,CANprovidesatomicityofmessagetransfer,i.e.eitheralloperationalnodescorrectlyreceiveamessageornoneofthem.TheconsistentviewofthemessagestatusbetweenallnodesisoneofthemostimportantfeaturesoftheCANbus(cf.chapter3).Itisachievedonthehardwarelevelbyasynchronousbittransmissionapproach.AllreceiversofaCANmessage(includingthesender)scanthebusandanalyzethecurrentmessagestatusduringthetransmissionofeverytransmittedbit.Ifanodelocallydetectsatransmissionfailure,itimmediatelyinvalidatesthecurrentmessagebyintentionallyproducingandetectableerroronthebus.Everynodeincludingthesendingnodenowisawarethatthemessagetransferfailed.Automatically,thesenderwilleventuallyretransmitthemessage.Ifnotransmissionerrorisdetetcted,alloperationalnodeshavereceivedandacceptedthemessage.Amoredetaileddescriptionofthisfeatureanditsexploitationforatomicgroupcommunicationisgivenin1428.ThebasicinterobjectcommunicationinsystemslikeJAVAandCORBAfocusonapointtopointcommunication.,therearesomeonetomanycommunicationmethodsinJAVAliketheobservable/observerclass1orrecently,theiBus24.Bothmechanismsarebasedonaexplicitregistrationofclientsattheserverwhichdifferssignificantlyfromageneralgroupcommunicationmechanismwhichisamany-to-manycommunicationparadigm.Thetransparentcommunicationmechanismusede.g.inMARUTI25enablestheobjectstouseasendprimitivewithoutspecifyingthereceiver(s).ThisisdonebyspecifyingthecommunicationchannelbindingsintheMARUTIconfigurationlanguage(MCL).Thegoalistoseparatefunctionalandnonfunctionalissues.However,thebindingisperformedduringcompiletime.Nodynamicchangesofthecommunicationrelationsarepossible.Thecommunicationmodelwhichweadoptissimilartothatofautonomousdecentralizedsystems(ADS)626.Inthismodel,softwaresubsystemsautonomouslymanagethemselvesandcoordinatetheiractivitieswithothersubsystems.Thiscoordinationisachievedbythedatafield,whichrepresentsglobalinformation.Theautonomousentitiescanextractrelevantinformationfromthedatafield.Basedonthismechanism,groupsofobjects,sharingasubsetoftheglobalinformation,canbeconstructed.Inordertosupportrealtimeobjectgroupswithconsistentinformation,thegroupcommunicationprotocolmustdeliverrealtimemessagestoallmembersofagroupbothtimely,andinaconsistentorder.Insection4wedescribehowtoguaranteetimelymulticastdeliverybyexploitingtheCANbusarbitrationmechanism.Giventheguaranteeoftimelymessagetransmission,wehavealsoshownin15thatthedeadlinecanbeusedtoconsistentlyorderhardreal-timemulticastmessages.3).Useoftimeinformation.Objectsmusthavesomenotionoftimetoexpressexecutiontimesoftheirmethods,deadlines,andslacktime.Inadynamicrealtimesystem,onlineschedulingdecisionsarebasedontheseassumptionswhichthereforemustbeavailableatruntime.Thetemporalbehaviouroftheobjectmustbespecified,verified,andmonitored.Anapproachtobindtimeinformationtoobjectclassesandmonitorthetemporalbehaviourofthesystemisdescribedin10.Itisplannedtousethesemechanismstospecifyandmonitorthetemporalbehaviourofthesystem.Inthispaper,however,weonlydescribetheuseoftimeinformationforthedynamicreservationofthebusresource.7中文翻译CAN总线系统的实时对象调用摘要本文方法是在CAN-based分布式实时系统中调用一个简单的实时对象,从而方便允许硬件和软件组成相关的对象模型的造型，共讨论了两个问题。首先，一个模型，可以形成和地址对象组，这反映了一种基本实时系统需要多客户有效分配信息。其次，我们讨论表达定时要求为对象invocations实现分布式共识的方法,通信资源访问EDF-like算法，利用知识的剩余期限、交流活动，剩下的时间为调用方法用在每点的时间。1、介绍未来的计算机系统,在很大程度上,这个结果必然要求监测和控制实际过程,及时性和可靠性分布式系统的固有免疫，对抗单一故障提供适当的架构去提高这个目标的可靠性。此外,因为realworld往往需要空间应用机电元件、分布式控制系统结构构建一个分布式计算能力的平台便宜但有力的支持分布式解决方案。至此构成了系统模型从智能组件的仪器设备接口,实现真实世界由传感器与执行器这样一种便捷的方法，是使用环境object-based方法在各种各样的实体对象封装需要控制的物理过程。我们可以辨别至少两层系统相互作用。利用一个较低的水平紧紧控制响应系统传感器/驱动器接口,在这一级系统和更高的系统负责解释和评价传感器所传达的仪器接口信号,来进行更高层次的决定。虽然不同层次有不同的要求,他们有时是简单的相关性与清晰的硬件结构,因为我们把注意力集中在较低的系统级和描述对象模型,该模型是由一个系统结构,靠模块化设计和独立的计算资源的仪器接口元件而不是一个中央处理器,机电组件强大的功能,控制信号的生成、机电组件对象本身是由廉价微控制器，这一层的对象有良好定义接口，从而可以利用由上级实例实现这个抽象控制，而不是处理低水平的控制信号。我们选择了由CAN总线连接的硬件模块应用(CAN:控制器局域网,由博世28,因为它提供先进的技术特点和代表了一种新兴的标准和适用性，而且新兴的单片微控制器可与CAN控制共享总线控制系统，设计连接一个空间限制区域,如汽车、机床、机器人及其他汽车或工业自动化应用。它是针对在嘈杂环境中运行的速度可以达到1Mbit/秒的系统,交换小实时控制信息。在某种意义上我们用能报文格式统一调用方法对我们的对象进行一个透明的对象调用,此外在这个系统，一个调用对象必须随机从CAN调用对象,我们可以解决组对象单一CAN通信。作为一种资源,交流媒介必须分享计算的节点。访问中已经被安排在一种分布式的计8算，以满足他们的最后时间,尽管自从调度的汽车无法立足于本地决策的通信线路竞争共识,显然一个分布式的公共汽车的定座已达到,这仅仅是一个特殊的例子，更一般的安排一个合作的分布式解决计算的问题。基于假定系统的行为模式和环境，CAN选择几种可供的途径来解决这个问题,。第一种方法,称为time-triggeredd,m.北京:21,m.北京:19的假设未来的一个完整的知识行为的系统。因此,这些系统运行过程中,有轻微的开销用于最大的可预见性和高度适合紧急安全程序,可以由一个周期解决。在一定程度上，这些实时系统运行计划运行时，一个更强的适应性和经济的方式来管理系统资源必须基于实时信息。虽然通常是平衡的，但这些系统的适应性和可预见性是可伸缩的,也就是说它们允许共存的活动与不同的实时要求。这些系统依靠实时保证30,11,而不是在只有一种calendar-based调度的可行性保证22,30是一种在即时活动运行时保证有可能提前预定的资源。应该指出可以用较少的重要任务调度政策，如最早的期限第(EDF),至少松弛第一(LLF)或率，只有保证最优、初始调度计划:如果存在,它将会被发现。但是在这个动态系统,可以有矛盾的要求,在这种情况下这些调度的政策将找不到解决办法,他们无法保证任何东西都在一个时间内,另一方面,已经预订一种机制来保障资源可用了。2、这个系统的模型我们的系统模型是受我们的预期的硬件基础限定的，我们假设许多具有不同的性能和特征的不同的微控制器,从8Bit32位架构所有的微控制器配备应用界面，我们利用不同的性能和价格的微控制器结构特点,在简单的微控制器的一些$s系统的总体任务被分成小块,在我们的应用实例它必须不再是一种稀有资源,简单的主动悬架系统(图1-2),我们用一个简单的单片机为每个伺服或复杂的传感器伺服电机组件，从内部传感器对当前位置提供反馈。该传感器/制动器的应用系统连接合作和其他类似的对象,通过控制指令能收到消息没有直接的、低水平的控制信号,以及CAN以外的模块。因此,在某种方法下封装和定义的接口，在逻辑层次、传感器/驱动器块被视为同一个物体时可被调用，例如把自己定位到一个特定的天使,天使、交流等构成一个最低水平的传感器/驱动器,可控制的层次类似一个简单的反映在生物系统的循环。因此更高的控制情况下，可用来控制群体的系统或最终整个系统更强大的微处理器,在控制情况下有很明确的仪器接口组成的系统,各自代表某个功能。每个对象都有一个独特的名字和一组相关的操作及独特的名称相对应，在运行时转化为一个短的形式系统的名字,保持在一个配置服务内，在这样的环境下系统必须有一些从被动模式的编程语言延伸到顺序，尤其是我们的模型，包括:(1)主动、自动控制类自从信号拥有专用处理器,它便是直接假定一个模型的自主,激活一个对象的特点，即一名和一系列的方法,但可被调用的对象应该是灵活的，它也可以发出信号，这相当于从一个对象调用到另一个对象，或发送到一组没有先前要求的对象。在第3章，我们提供一个透明的群体沟通机制，即被控对象本身通常不知道它是否与系统在一个遥远的地方通信。事实上这个小组远程调用沟通机制是系统关系协调的基础，沟通关系可以被定义在动态运行时间里，所以信号可以加