《电子与通信工程专业英语》课件第4章

Unit4ApproachestoOpticalInternetPacketSwitchingNEWWORDSANDPHRASES

NOTES

EXERCISES

参考译文

EXTENSIVETEXT

1．Introduction

Theconvergenceoftelecommunicationsanddatacommunicationshascausedaparadigmshiftinthenetworkingenvironment.ThemassiveexplosionintrafficgeneratedbytheInternethasdriventhecurrenttrend,withtheInternetProtocol(IP)becomingthedominantprotocolfordatacommunicationsaswellasrepresenting,inthelongerterm,averystrongcandidatefortheconvergenceofdatacommunicationswithtelecommunications.Intandem,thedevelopmentofwavelength-divisionmultiplexed(WDM)techniquesonpoint-to-pointlinkshavebeguntoutilizethemassiveopticalbandwidthofinstalledsingle-modeopticalfibersmoreefficiently.WDMwasinitiallyarapidsolutiontothesevererouteexhaustionproblemsbroughtaboutbyexponentiallyincreasingtraffic.Thefuturedeploymentofopticalcross-connectedWDMtransportnetworks,initiallyforprotectionandbypassbutultimatelymanagingopticallightpathsdynamicallyinmultipleringormesharchitectures,willpotentiallymodifytheroleofthenetworkfunctionalitiesprovidedbythesynchronousopticalnetwork/synchronousdigitalhierarchy(SONET/SDH)layer[1].Thus,IPoverWDMhasbecomeaveryimportantareaofstudy,encompassingawiderangeofsolutionstosupportingpredominantlyIPtrafficoverWDMopticalpaths.MuchresearcheffortfocusesondevelopinganelegantsolutiontothemismatchbetweenthetransmissioncapacitiesofferedbytheWDMopticallayerandtheprocessingpowerofrouters.IProutersperformfourmaintasks:

(1)

Routing:providingnetworkconnectivityinformationthroughroutingtables.

(2)

Forwarding:definingtheoutputofeachincomingpacket(basedontheroutingtables).

(3)

Switching:directingeachpackettotheproperoutput(definedbytheforwardingprocess).

(4)

Buffering:resolvingcontentionbystoringpacketswhenmorethanonewishestogotothesameoutputatonceduetotheunschedulednatureoftheirarrival.

Currently,theforwardingprocessimpliesmajorthroughputlimitations,withthesizeoftheroutingtablesandfrequencyoftheirupdatesbeingmajorissues.Suchproblemsarecurrentlyaddressedandmanaged,butthetimeneededfortablelookupsetsafundamentallimitonrouterthroughput.Muchworkhasconcentratedonthedevelopmentofdatastructuresandalgorithmsforminimizingthelookuptimegivenroutingtableandmemoryspaceconstraints.

Thereisnodoubtthatmassivestrideshavebeenachievedinhigh-throughputrouterdesigns(>1

Tb/s).Nevertheless,despitetheseimpressiveadvances,thereisstillafearthatelectronicswitchingsystemsexhibitlimitedupgradeflexibility.GiventhatWDMallowscheapandeasyincrementalincreasesofthetransmissionbandwidth,frequentupgradesofthetransport-layertransmissioncapacitycanbeenvisagedtomatchincreasingdemand,inturnplacingheavydemandsontheswitchingprocess[2].

2．Opticalpacketswitching

Anumberofotherapproachestoobviateoramelioratetheforwardingbottleneckarebeingresearched,utilizingopticalpacketorbursttechniques,perhapswithelectronicbuffering,toimplementIPoverWDM.However,thestrategydetailedhereimplementscontentionresolution(buffering)directlyintheopticaldomaintoyieldWDMopticalpacketswitching.Theobjectiveistoshiftthebulkoftheswitchingburdenintotheopticaldomain,permittingcompatiblescalingoftheswitchingcapabilitywithWDMtransmissioncapacity.Thusfar,thisstrategyhasassumedahybridsolution,achievingdecouplingbetweenthethroughputandtherouting/forwardingprocesses.Transmissionandswitchingareexecutedintheopticaldomain,whileroutingandforwardingarecarriedoutelectronically,wheretherelativelycomplexpacketheaderprocessingoccursindependentoftheopticalpayload[3].ThisdecouplingeffectivelypermitstheopticalpacketlayertosupportarangeofnetworkprotocolswhileharnessingthepowerofWDMtransmission.However,itmustbenotedthatthisisalsochanging,withtherecentdemonstrationofrudimentaryheaderprocessingfunctionsdirectlyintheopticaldomain.Theserelievesomeoftheburdenplacedonelectronicprocessing,therebyreducingcontrolsignalsetuptimeandmanaginglatencymoreeffectively.

Withanextensiveopticalpacketlayer,theinterfacetoIPandotherprotocolsiscrucial.Encapsulation,theadditionofdeliveryinformationtothedatabytheopticalpacketlayer,willoccuratinterworkingunits(IWUs)ateachinterfacetotheelectronicclientlayer.EncapsulationpermitsarangeofprotocolssuchasIPandasynchronoustransfermode(ATM)tobemappedintotheopticalpayloads,whichmaybeofeitherfixedorvariableduration.IPhidesthecomplexityofthephysicallayer(includingopticalpacketswitching),providingaunifiedinterfacetohigherlayers,regardlessoftheunderlyingnetworktype.Inadditiontoencapsulation,theIWUscreateheadersforproperroutingwithintheopticalpackerlayer,andmultiplextrafficfromdifferentinputlinksforonwardtransmissioninopticalpacketsforthesamedestination,ensuringanentirelyopticalend-to-endconnectionpath.Opticalpacketsprovideafurthermultiplexingtier,allowingtheaggregationoftrafficflowspriortotransmissionovertheopticallayer,andalsopotentiallyobviatingtheneedforSDHasanadaptationlayerforIPtrafficonWDMlinks.Opticalpacketnetworkthereforeoffersapotentialsolutiontoprovidingbothconnectionlessandconnection-orientednetworkingcapacities,flexibleintermsofbandwidthmanagementandfuture-proofwithregardtobandwidthgrowth.NostandardsexistyetformappingprotocolssuchasIPandATMintotheopticalpacketlayer.

Ashintedabove,therearetwoprincipalapproachestoopticalpacketswitching,bothwithapplicationstotheInternet:

(1)

Employingfixed-lengthopticalpackets,withmanycorrespondingtooneIPdatagram,requiringIWUstofragmentandreassemblethepacketseitherattheedgesofthelayerorontheinputsandoutputsoftheswitch.

(2)

Employingavariable-lengthopticalpacketforeachIPdatagram.

Mostreportedresearchtodatausesfixed-durationopticalpackets.Hence,forthepurposesofthisarticle(inordertoillustratethefunctionalityofanopticalpacket-switchedlayer),thebulkofthesubsequentmaterialwillbeconfinedtofixed-lengthpacketswhereboththeheaderandpayloadareencodedonthesamewavelength.Itisassumedthatthedestinationswitchoutputforeachpacketisderivedfromtheheaderafteropto-electronicconversion;theheadermaythusbeatalowerbitratetoallowitselectronicmanipulation.Duetothenatureofopticalbuffering,thepayloaddurationisfixed,whateveritscontent;thenetworkthroughputisproportionaltopayloadbitratewhichmayvaryfrom10Gb/sandup,witheasyupgradecapability.Aswillalsobecomeapparent,thewavelengthdimensioniscrucialnotonlyfortransmissioncapacitybutalsoinexecutingpracticalcontentionresolution.Thearticlewillendwithsomeforward-lookingconceptsaddressingtherequirementstoswitchandbuffervariable-lengthopticalpackets.

3．Thedesignofopticalpacketswitches

Agenericopticalpacket-switchednodestructureconsistsofthreesubblocks(Fig4.1):

(1)

Aninputinterfaceconsistingofan(optical)synchronizerwhichalignsincomingpacketsinrealtimeagainstaclock.

(2)

Aswitchingcorewhichroutesthepacketstotheirproperoutputsandexecutescontentionresolution.

(3)

Anoutputinterfacewhichinsertsanewheaderandmayhavetoregeneratethedata.

Fig4.1Agenericopticalpacket-switchednodestructure

Packetformatisafundamentalconsiderationinanypackettransmissionsystem,definedbytherequirementsofthelayerwithrespecttonetworkfunctionalities,andisalsocrucialforopticaldomainimplementations[4].Forexample,considerthepacketformatdefinedbytheKEOPSproject,uponconsiderationofthedelay-throughputperformanceoftheopticalpacket-switchedlayerunderdifferenttrafficflowsaswellasconsiderationofthenodeandnetworkroutingrequirements(Fig4.2).Throughoutitisassumedthattimeisdividedintoequaltimeslots,eachcontainingoneopticalpacket,andthepayloadmaycontaindatafrom622Mb/sto10Gb/s.

Fig4.2Anexampleofanopticalpacketformat(asdefinedbyKEOPS)

Insummary,theheaderfieldis14bytes:8forroutinginformation;3foridentificationofpayloadtype;flowcontrolinformation,packetnumberingforsequenceintegritypreservation,andheadererrorchecking.Theformatalsoindirectlyindicatestheneedforopticalsynchronizersastheinputinterfacetothenode.Itisassumedthatwhenpacketsentertheswitch,theirboundariesarealignedsothateachpacketisalignedwithitstimeslot.Suchsynchronizationisgenerallyarequirementforcorrectswitchoperation,achievinglowpacketloss;approachestoachievingpacketsynchronismattheinputstotheswitchesconstituteaseparatetopicandwillnotbediscussedfurther.Thus,2bytesarerequiredtoaidinthesynchronizationprocessas“tag”uniquelyidentifyingthestartofthepayload.Guardbandsaccountforswitchingtimesoftheconstituentopto-electronicdevicesaswellaspayloadpositionjitter.Thepayloaddurationresultsfromatrade-offbetweentransmissionefficiency(thelongerthepayloadduration,thehighertheefficiency)andthepracticallimitationsonoveralllengthoftheopticalfiberdelaylinebuffers.Atthistime,delaylinesaretheonlyviableapproachtoimplementingcontentionresolutiondirectlyintheopticaldomain.

Anopticalpacketswitchhasthreeprincipalfunctions:switching,buffering,and,optionally,headertranslation.InIP-orientedsystems,thelatterfunctionmaybereplacedbyroutingandforwarding,whichwerediscussedearlier.Switchingensuresthateachpacketemergesatthecorrectoutput,dependingontheinformationcontainedinthepacketheader.Althoughfixed-lengthpacketsarrivingontheinputsmustbesynchronized,thereisnocoordinationbetweenpacketstreamsarrivingondifferentinputs.Henceoneormorepacketsmayarriveduringthesametimeslotondifferentinputswishingtogotothesameoutput.Forthisreason,bufferingisrequired,whereoneormorepacketsarestoredwhileothersaretransmittedtothedesiredoutput.

HeadertranslationrepresentsacentralprocessintheATMtransmissionsystemsstrategy,while,asdiscussedpreciously,routingandforwardingareequallyimportantforIPsystems.Althoughtheseschemesoffergreatfunctionalityandflexibility,theyarenotusedineveryopticalpacket-switchingsystemsincedirectopticalheadertranslationisintheearlystagesofdevelopment.Thus,theheaderisusuallyatalowerbitratethanthepayloadtofacilitateelectronicdecodingandinterpretationofheaderinformation.

NEWWORDSANDPHRASES

convergence n. 会合，会聚

exhaustion n. 耗竭；耗尽；耗损；衰竭

exponentially adv. 指数地

mesh n. 网状，网眼

elegant adj. 优良的，雅致的；简洁的

stride n. 进展，进步；步幅

envisaged vt. 正视

obviatevt. 消除，排除(危险、障碍等)，回避，预防

ameliorate vt. 改善，改进

bottleneck n. 瓶颈

contention n. 争夺，争论

hybrid adj. 杂种的，混合的

decouple vt. 解耦，去耦合

harness vt. 治理，利用

rudimentary adj. 根本的，低级的

latency n. 等待时间；执行时间

unified adj. 同一的，统一的

future-proof 经得起未来考验的

manipulation n. 操作，处理

withrespectto 关于，至于

NOTES

[1]Thefuturedeploymentofopticalcross-connectedWDMtransportnetworks,initiallyforprotectionandbypassbutultimatelymanagingopticallightpathsdynamicallyinmultipleringormesharchitectures,willpotentiallymodifytheroleofthenetworkfunctionalitiesprovidedbythesynchronousopticalnetwork/synchronousdigitalhierarchy(SONET/SDH)layer.“initiallyforprotectionandbypassbutultimatelymanagingopticallightpathsdynamicallyinmultipleringormesharchitectures,”是原因状语，用来解释前面网络使用的目的。“providedbythesynchronousopticalnetwork/synchronousdigitalhierarchy(SONET/SDH)layer.”修饰“networkfunctionalities”，它是省略了whichare的后置定语从句。

本句可译为：光交叉互联WDM传输网络最初是为了保护和旁路多环网或网状网，但最后(演变为)动态管理光路，它可能改变网络功能的角色，这些网络功能是由同步网络/同步数字体系层提供的。

[2]GiventhatWDMallowscheapandeasyincrementalincreasesofthetransmissionbandwidth,frequentupgradesofthetransport-layertransmissioncapacitycanbeenvisagedtomatchincreasingdemand,inturnplacingheavydemandsontheswitchingprocess.

“Giventhat…”意为“假设……则……”，表示某种假设情况，后面是前面假设条件成立后的结果。“Match”意为“匹配”。“demandson”意为“对……的要求”。

本句可译为：假设WDM允许传输带宽轻易、频繁地不断增长，传输层传输容量的频繁升级则可看做是匹配日益增长的容量需求，进而就会对交换过程施加繁重的需求压力。

[3]Transmissionandswitchingareexecutedintheopticaldomain,whileroutingandforwardingarecarriedoutelectronically,wheretherelativelycomplexpacketheaderprocessingoccursindependentoftheopticalpayload.“while”引导并列从句，和前面的“Transmissionandswitchingareexecutedintheopticaldomain,”是并列从句；“where”引导定语从句，修饰“electronically”。

本句可译为：传输和交换在光域实现，而路由和转发在电域实现，相对复杂的分组头独立于光载荷在电域进行处理。

[4]Packetformatisafundamentalconsiderationinanypackettransmissionsystem,definedbytherequirementsofthelayerwithrespecttonetworkfunctionalities,andisalsocrucialforopticaldomainimplementations.“providingtheleastsignalattenuationandthehighesttransmissionspeedsofanyfibercabletype.”为分词短语作结果状语，可以理解为whichcanprovidetheleastsignalattenuationandthehighesttransmissionspeedsofanyfibercabletype的缩写。

本句可译为：分组的格式在任何分组传输系统中都是一个基本的、需要考虑的事项，对于光域的应用也是至关重要的。分组的格式是由某层根据网络功能的需求决定的。

EXERCISES

I．Translatethefollowingwordsorphases.

IP WDM IWU

SONET/SDH hybridsolution opticalpayload

光分组交换 光电转换 光纤延迟线

传输容量

II．TranslatethefollowingparagraphsintoChinese.

(1)

Withanextensiveopticalpacketlayer,theinterfacetoIPandotherprotocolsiscrucial.Encapsulation,theadditionofdeliveryinformationtothedatabytheopticalpacketlayer,willoccuratinterworkingunits(IWUs)ateachinterfacetotheelectronicclientlayer.

(2)

Intandem,thedevelopmentofwavelength-divisionmultiplexed(WDM)techniquesonpoint-to-pointlinkshavebeguntoutilizethemassiveopticalbandwidthofinstalledsingle-modeopticalfibersmoreefficiently.WDMwasinitiallyarapidsolutiontothesevererouteexhaustionproblemsbroughtaboutbyexponentiallyincreasingtraffic.

(3)

Thus,2bytesarerequiredtoaidinthesynchronizationprocessas“tag”uniquelyidentifyingthestartofthepayload.Guardbandsaccountforswitchingtimesoftheconstituentopto-electronicdevicesaswellaspayloadpositionjitter.Thepayloaddurationresultsfromatrade-offbetweentransmissionefficiency(thelongerthepayloadduration,thehighertheefficiency)andthepracticallimitationsonoveralllengthoftheopticalfiberdelaylinebuffers.Atthistime,delaylinesaretheonlyviableapproachtoimplementingcontentionresolutiondirectlyintheopticaldomain.

(4)

Switchingensuresthateachpacketemergesatthecorrectoutput,dependingontheinformationcontainedinthepacketheader.Althoughfixed-lengthpacketsarrivingontheinputsmustbesynchronized,thereisnocoordinationbetweenpacketstreamsarrivingondifferentinputs.Henceoneormorepacketsmayarriveduringthesametimeslotondifferentinputswishingtogotothesameoutput.Forthisreason,bufferingisrequired,whereoneormorepacketsarestoredwhileothersaretransmittedtothedesiredoutput.

参考译文

第四单元光互联网分组交换方法

1．简介

无线电通信和数据通信的融合导致了网络环境的模式变迁，Internet产生的大量业务推动了这种趋势。随着IP协议正在成为数据通信的主要协议，在较长的时期内，IP协议也是代表无线通信和数据通信融合的强大的支持方。一前一后地，点对点链路的WDM(波分复用技术)的发展已经开始利用已有的单模光纤的大带宽使其利用率更高。最初，WDM为呈指数增长的业务的严重路由耗尽问题提供了快速解决的方法。光交叉互联WDM传输网络最初是为了保护和旁路多环网或网状网，但最后(演变为)动态管理光路，它可能改变网络功能的角色，这些网络功能是由同步网络/同步数字体系层提供的。因此，IPoverWDM已经成为一个重要的研究领域，它包括了支持主要的IP业务在WDM光链路上传输的许多方法。许多研究工作集中在开发一种优良的解决办法，这种办法用来解决WDM光层的传输能力和路由器的处理能力之间的不匹配问题。IP路由器完成四个主要的功能：

(1)路由：通过路由表提供网络连接信息。

(2)转发：定义每个输入分组的输出端口(基于路由表)。

(3)交换：把每个分组输出到正确的输出端口(由转发过程定义)。

(4)缓存：由于分组到来的随机性，当多个分组想要同时通过同一出口时，可通过存储分组解决冲突。

目前，随着路由表尺寸的变大和更新频率问题，转发过程成为了限制吞吐量的主要问题。这些问题目前已得到解决，但是路由表的吞吐量(依然)基本受限于路由表查询时间。大量的工作集中在开发数据结构和算法，这些数据结构和算法(的作用)是为了最小化路由表的查询时间和存储空间限制。

毫无疑问，在大吞吐量路由器设计上已经取得了巨大的进展(大于1

Tb/s)。不过，尽管取得了这些惊人的进步，但人们仍然对电子交换系统有限的升级灵活性存有疑虑。假设WDM允许传输带宽轻易、频繁地不断增长，传输层传输容量的频繁升级则可看做是匹配日益增长的容量需求，进而就会对交换过程施加繁重的需求压力。

2．光分组交换

许多其他的解决或消除转发瓶颈的方法正在进行研究。利用光分组或突发技术，或者与电缓存一起来实现IPoverWDM。但是，这里讲述的策略是直接在光域解决冲突(缓存)来让步于WDM光分组交换。其目的是把交换负担转移到光域，允许交换能力和WDM传输容量之间进行有比例的协调。因此这种策略可认为是一种混合解决方法，它解除了吞吐量和路由/转发过程之间的耦合关系。传输和交换在光域实现，而路由和转发在电域实现，相对复杂的分组头独立于光载荷在电域进行处理。这种去耦合方式使得光分组层能有效地支持很多网络协议，同时也利用了WDM的传输能力。但是需要注意的是：(这种情形)随着最近的一些演示情况在发生变化，这些演示中，有一些分组头头部处理功能可初步直接在光域进行，这减轻了电处理的负担，因而也有效地减少了控制信号的建链时间和管理时间。

有了一个扩展的光分组层，则它与IP协议和其他协议的接口就是至关重要的。在每个互连单元(IWU)与电客户层的接口处光分组层进行传输信息的叠加，即封装。封装允许大量的协议数据比如IP和ATM数据映射到光载荷中，这些光载荷的持续时间可能是固定或可变的。IP隐藏了物理层的复杂性(包括光分组交换)，对高层提供统一的接口，而不管底层网络是什么类型的。除了封装功能，互连单元(IWU)也为(光分组)在光分组层的正确路由生成头部，同时为了转发而复用来自不同输入链路的、去往同一目的地址的分组业务，从而保证了一个全光的端到端的链路(的正确运行)。光分组提供了一个更高的复用等级，它允许业务流在光层传输之前进行聚合，潜在地消除了对SDH的需求(SDH是IP业务在WDM链路上传输时需要的适配层)。因此，光分组网络提供了一种潜在的解决办法：能传输无连接和面向连接业务，并且具有灵活的带宽管理能力和经得起未来考验的带宽增长能力。目前还没有制定IP和ATM数据映射到光分组层的标准。

正如上面所提到的，有两种主要的近似于光分组交换的方法，都可以应用到Internet中：

(1)使用固定长度的光分组，很多个这样的分组对应一个IP数据报，需要IWU在层间或交换机的输入和输出端进行分段和重组。

(2)每个IP数据报使用一个不定长度的光分组。

大部分已经报道的研究都使用固定长度的光分组。因此，为了说明光分组层的功能，下面的资料都限于固定长度分组，分组的头部和载荷都使用同样的波长来编码。假设每个分组的目的交换端口是光/电转换后从头部获取的，那么，为了使得电操作(成为)可能，其头部的速率应当较低。由于光缓存的天然特性，载荷的持续时间是固定的，不管它的内容是什么，网络的吞吐量和负载比特率是成比例的。负载比特率的范围可从10Gb/s到更高，具有很大的上升空间。显而易见，波长的个数不但对传输容量是至关重要的，而且对在实际运用时的冲突解决也是至关重要的。为了交换和缓存变长光分组，本文将提出一些有前瞻性的概念来分析解决这些需求，并以这些概念结尾。

3．光分组交换机的设计

一个通用的光分组交换机的节点结构包括三个子部分(见图4.1)：

(1)输入接口，包括一个用来对准实时时钟的输入分组(光的)同步装置。

(2)交换核心，用来把分组路由到它们正确的输出端口，并执行冲突解决。

(3)输出接口，它插入新的分组头，并且可再生数据。

图4.1通用光分组交换机节点结构分组的格式在任何分组传输系统中都是一个基本的、需要考虑的事项，对于光域的应用也是至关重要的。分组的格式是由某层根据网络功能的需求决定的。例如，KEOPS工程定义的分组格式既考虑了不同业务流情况下的光分组交换层的延迟-吞吐量性能，也考虑了节点和网络路由的需要(见图4.2)。自始至终它都假设时间被分成相等的时隙，每个时隙容纳一个光分组，并且有效载荷可能包含从622Mb/s到10Gb/s速率的数据。

图4.2光分组交换格式例子(KEOPS的定义)概要地说，头域有14个字节：8个字节是路由信息，3个字节是有效载荷的类型标识、流量控制信息、分组序号完整性保护编码和头部错误校验。分组格式间接地表明了对光同步装置作为节点输入接口的需要。假设当分组进入交换机时，它们的边界是对准的，那么每个分组和它的时隙也是对准的。一般来说，这样的同步过程对于正确的交换操作和低的分组丢失率是必需的。在输入端到交换机之间同步的方法可以构成一个单独的命题，这里不作深入讨论。因此，在同步过程中，使用2个字节作为“标签”特别标识有效载荷的开始来帮助达到同步。保护频带负责组成光电设备的交换时间以及有效载荷位置的抖动。有效载荷持续时间源于传输效率(持续时间越长，效率越高)和光纤延迟线缓存总长的实际限制之间的折中平衡。此时，光纤延迟线是唯一能直接在光域解决冲突的可行办法。

光分组交换有三个主要功能：交换、缓存和头部翻译(可选)。在基于IP的系统中，后面的功能(头部翻译)可能被路由和转发所替换，这个问题前面讨论过了。交换依靠包含在分组头中的信息来保证每个分组能能被传送到正确的输出端。尽管到达输入端的定长分组一定是同步过的，但是来自不同输入端的分组流也达不到(完全)一致。因此，一个或多个分组可能在同一时隙到达不同输入端，并去往相同的输出端。由于这个原因，缓存是必需的，在这里一个或多个分组被存储起来，而其他的分组被传送到希望的输出端口。

头部翻译功能在ATM传输策略中是一个重要的过程，而路由和转发功能在IP系统中也是同样重要的，这点前面已讨论过了。尽管这些机制提供了重要的功能和灵活性，但是并没有体现在每个光分组交换系统中，因为直接的光分组头部翻译功能还处在发展的初级阶段，还有很长的路要走。因此，为了方便电解码和头部信息的翻译，头部的比特率通常比有效载荷的比特率低。

EXTENSIVETEXT

OpticalBurstSwitchingfortheNextGenerationInternet

Currently,ourdemandsfornetworkbandwidthincreasedaily.WeaccesstheInternettocheckemail,readthenews,checkstockquotes,takeclassesonline,finddrivingdirections,playgamesonline,buyflowersandthislistgoesonandon.Inordertomeetourever-increasingnetworkbandwidthneeds,onesolutionistodesignandbuildtheNextGenerationInternetwithanopticalcoreinfrastructure,withfastconnectionprovisioningtimeandunprecedentedhighdataratesof100terabitspersecondandhigher.Anopticalnetworkisbuiltbyinterconnectingvariousopticalswitcheswithwavelength-divisionmultiplexing(WDM)fibers,i.e.,fibersthatcansimultaneouslytransmitdataoverdifferentwavelengths.Manyoftoday’scommercialopticalnetworks,however,donotutilizetheWDMtechnologyefficiently.Oneexampleisthepopularsynchronousopticalnetwork(SONET)rings,whichconsistofSONETadd/dropmultiplexers(ADM)linkedwithWDMfiberssoastoformaring.Inthiscase,WDMissimplyusedasatransportmedium.Thatis,thedataistransmittedoptically,buttheopticalsignalisterminatedateveryADM.ThesignalisthentranslatedtoelectronicsforprocessingandthentranslatedbacktoanopticalsignalforforwardingtothenextADM.TheproblemisthemismatchbetweentheelectronicprocessingspeedsandtheopticaltransmissionratesthatleadtoanundesirablebottleneckateachSONETnode.Toresolvethebottleneckproblem,anall-opticalWDMnetworkwherethedatatravelsfromthesourcetothedestinationentirelyintheopticaldomainhasbeenproposed.Withrespecttothecurrentstateofthetechnology,anOpticalBurstSwitched(OBS)Networkisoneofthemostpromisingall-opticalarchitecturesfortheNextGenerationInternet.Itefficientlysupportsthetransmissionofburstytrafficoveranall-opticalinfrastructure.OBSisstillbeingdevelopedandithasnotbeenstandardizedyet.ThisarticledescribesthemainfeaturesofanOBSnetwork,itsbenefitsaswellasitschallenges.

1．TheOBSbasics

AnOBSnetworkconsistsofcorenodesandend-devicesinterconnectedbyWDMfibersasshowninFig4.3.AnOBScorenodeconsistsofanopticalcrossconnect(OXC),anelectronicswitchcontrolunit,androutingandsignalingprocessors.AnOXCisanon-blockingswitchthatcanswitchanopticalsignalfromaninputporttoanoutputportwithoutconvertingthesignaltoelectronics.TheOBSenddevicesareelectronicInternetProtocol(IP)routers,asynchronoustransfermode(ATM)switches,orframerelayswitches,equippedwithanOBSinterface.EachOBSend-deviceisconnectedtoaningressOBScorenode.Theend-devicecollectstrafficfromvariouselectronicnetworks(suchasATM,IPandframerelay).ItsortsthetrafficperdestinationOBSend-deviceaddressandassemblesitintolargervariable-sizeunits,calledbursts.Foreachburst,theend-devicealsoconstructsacontrolpacket,whichcontainsinformationabouttheburst,suchastheburstlengthandtheburstdestinationaddress.Thiscontrolpacketisimmediatelysentalongtherouteoftheburstanditiselectronicallyprocessedateachnode.

Fig4.3TheOBSnetworkarchitecture

Thefunctionofthecontrolpacketistoinformthenodesoftheimpendingdataburstandtosetupanend-to-endopticalpathbetweenthesourceandthedestination.Afteradelaytime,knownastheoffset,theend-devicealsotransmitstheburstitself.Thebursttravelsasanopticalsignalovertheend-to-endopticalpathsetupbyitscontrolpacket.Thisopticalpathistorndownafterthebursttransmissioniscompleted.ThisseparationofthecontrolinformationandtheburstdataisoneofthemainadvantagesofOBS.Itfacilitatesefficientelectroniccontrolwhileitallowsforagreatflexibilityintheformatandtransmissionrateoftheuserdata.Thisisbecausetheburstsaretransmittedentirelyasanopticalsignal,whichremainstransparentthroughoutthenetwork.Ingeneral,thetimeittakesthecontrolpackettoreachthedestinationend-deviceisequaltotheend-to-endpropagationdelayplusthesumofalltheprocessingdelaysatalltheintermediatecorenodes.Ontheotherhand,thetimeittakesforabursttoreachthedestinationend-deviceisonlyequaltotheend-to-endpropagationdelay.ThereasonisthattheburstistransmittedasanopticalsignalthattraversestheOBSswitcheswithoutanyprocessingorbufferingdelays.Inviewofthis,thetransmissionofaburstisdelayedbyanoffset.ThiswayitalwaysarrivesatanOBSnode,afteritsswitchcontrolunithashadthechancetoprocessthecontrolpacketassociatedwiththeburstandconfiguretheopticalswitchfabric.Theoffset,therefore,isafunctionofthenumberofnodesthatthecontrolpackethastotraverseendtoend.

2．Burstaggregationalgorithm

Theburstaggregationalgorithmattheend-devicescangreatlyimpacttheoverallOBSnetworkoperationbecauseitsetstheburstcharacteristicsand,therefore,itshapestheburstarrivaltraffic.Thealgorithmhastoconsiderthefollowingparameters:apre-settimer,amaximumburstlengthandaminimumburstlength.Thetimerdetermineswhentheend-deviceistoassembleitscollectedtrafficintoanewburst.Themaximumandtheminimumburstlengthparametersshapethesizeofthebursts.Weneedtosetamaximumburstlengthsinceverylongburstsholdontotheresourcesofthenetworkforalongtime.Thus,theycausetheunfairlossofotherbursts.Ontheotherhand,theminimumburstlengthisalsonecessarybecauseveryshortburstsmaygiverisetotoomanycontrolpackets.ThissituationcanoverloadthecontrolunitoftheOBSnode.Theburstaggregationalgorithmmayusebit-paddingifthereisnotenoughdatatoassembleaminimumsizeburst.OnewaytoprovideclassesoftrafficinOBSistoimplementpriorityqueuesattheedgeofthenetworkduringtheburstaggregation.Basedontheclassofservice,theend-devicessorttheupperlayertrafficintodifferentqueues.Asaresult,eachend-devicewillhaveC*Npriorityqueues,whereCisthenumberofserviceclassesandNisthenumberofpossibledestinations.Anappropriateschedulingalgorithmguaranteesthatthesequeuesareservedaccordingtotheirpriority.

3．Signaling,routingandwavelengthallocation

Signalingisanimportantaspectinanynetwork.Itspecifieshowtheconnectionsareestablishedanditdetermineswhetherornottheresourcesareutilizedefficiently.InmostOBSvariants,thesignalingofconnectionsisaccomplishedusingaone-waysignalingscheme:theburstistransmittedafteranoffsetwithoutanyknowledgeofwhethertheopticalpathhasbeensuccessfullyestablishedend-to-end,asillustratedinFig4.4.

Fig4.4One-waysignaling

End-devicesAandBareconnectedviatwocoreOBSnodes.TheverticallinesinFig4.4representatimelinesoastoshowtheactionstakenbyeachnode.End-deviceAtransmitsacontrolpackettoitsingressOBSnode.Thecontrolpacketisprocessedattheingressnode.Iftheconnectioncanbeaccepted,itisforwardedtothenextnode.Thecontrolpacketisreceivedby