Principle analysis of IP wavelength router

\ol. 44 Nu. 6SCIENCE IN CHINA (Series F)Irevrondr 2001RESEARCH NOTEPrinciple analysis of IP wavelength routerWANG Yong (王勇), YIN Hongxi (殷洪玺)，XU Anshi (徐安士)&WUDeming(吴德明)TR9| ANational Laboratory on Local Fiber-Optic Communication Networks and Advanced Optical Communication System.Department of Electonics . Peking University. Beiing 100871. ChinaReceived May 15, 2001AbstractCombining IP with WDM is an attractive direction for research. WDM will play an impor-tant role in IP network in future. Now, an urgent problem is how to introduce wavelength routing in anIP network. We solve this problem by designing IP wavelength router, implementing DPDP ( defaultpath and dedicated path) method. We prove the reasonableness and feasiblity of this design by a prin-ciple experiment. A lot of problems related to this design are also discussed.Keywords: IP over WDM，wavelength routing, optical crossconnect, IP engine.The Internel taffic is exploding . The router has been a bottleneck of developmcnt of the In-ternet . Although high performance roulters, for example，NX64000 of Luccnt, have capacity upto6.4 Tb/s, they cannot catch up with the rapid progress of WDM technology. With more andmore wavelength connecting to a rouler, the router should process every incoming packet of all thewavelengths ，despite the fact that most of the trf is not destined to it on average. This is reallya burden for IP router and in turm gives a good chance for wavelength router. Since optical wave-length channel is transparent to IP and wavelength routing in all-optical networks has once beenstudied thoroughly, being used as a‘label' or provisioned channe!，wavelengths can bypassintermediate nodes' 3- 5 Jso the network can achieve very high throughput，low delay and guaran-teed QoS.Therefore, we think that WDM will play an important role in Internet in future. We pruposeIP wavelengh router，using a new approach, called DPDP (defult path and dedicated path), tointroduce wavelength routing in Internet. We present ils conceivable architecture in detail and ex-plain its working procedure. We also carry out a principle experiment to prove reasonableness andfeasibility of the design .1 DPDP node structure, working procedure and communication protocolThe starting point of DPDP is similar to IP switching. For example, in fig. I, we assumethat a packet Alow enters at ingress A and exils al egress E. Normally, each packet of the flow isstored and forwarded by each router along the shortest path, A→C→D- E, on wavelengh λo，the default path. Because we have classified Alows into three classes: QoS urgent, QoS not urgentbut huge, and QoS not urgent and not huge, the coming flow can be lassified by a flow identifi-er. Then analyzed resuls can be reported to router A. According to the result, router A tries toassign a free wavelength to this flow if it is QoS urgent or taffec huge. The assigned wavelength iscalled a dedicated path because it will cut through some intermediate nodes or even reach the egress中国煤化工YHCNMHGNu. hPRINCIPILE ANALYSIS OF IP WAVFILENGTH ROUTFER475router directly. In the worst case, this flow has to traverse hop by hop, but if wavelengh re-sources are enough and wavelength assignment algorithm is effective and reasonable, in most cas-es, the average hop number will be signifcantly reduced, and in turm the router' s processingburden will be alleviated." [Domain 2As can be seen in fig. 2 IP wavelength routerFgress年子I is organic combination of IP router with 0XC. Therouter configures the lightpath for WDM while-点WDM provides high-speed， transparent lightpathIngrsAHackboncfor the rouler. All of the abuve is based on the as-sumption that the network is backbone and packetDomain 1)fows among backbone routers are often ‘ bugwhich means sufficient intensity and long duration .rig. 1. IP backbone nctwork.As is shown in fig. 2(a), the node structureIP engine--tForwarding cngine- Contol .Optical sutchFlowwARoutingidentifier decision"T menor:↓enterCotroLOpncal sntcha)(bFig. 2. Architeeture of DPDP. (a) Node structure; (b) IP engine .of DPDP consists of two parts: an optical crosscon-Setupnect6l and an IP enginel7l. Fig. 2(b) ilustratesthe details of the improved IP engine. Initially, a个Ap人SetulWDM DEMUX should be used to demultiplex thededicated path and the defaul path. All the defaultpaths of the optical links are gathered to IP engine,where the packel flows first enter forwarding engineCommitand flow identifier simultaneously. while the samededicated paths are also gathered to an opticalFCommitswilch. In IP engine, by defaul, the packels areforwarded to different output ports on default wave-不心不h不小名不山不不小whlength. As the∩low identifier reports that there is aFig. 3. wavelength asignnente procedure.huge or QoS urgent flow, the WA ( wavelength as-signment) decision center will determine whether to assign new dedicated path or not. If yes, aconnecting signaling, WAP ( wavelength assignment procedure) should be called to establish alightpath for the flow . This is described as below. Note that the default path is both a data chan-nel and a control channel of dedicated paths. The dedicated paths eannot be set up without thepermission of the routers. In other words, the IP engine knows whatever happens in dedicatedpath.中国煤化工YHCNMHG476SCIFENCE. 1\ CIHINA (Serits F)Vinl. 44WAP is a simple wavelength assignment procedure. Because W AP actually intends to set upa continuous lightpath as far as possible, it should run following the time sequence of fig. 3.Originally, the ingress A should declare its free wavelength resoures to nexl router C, thenafter rereiving the free wavelength information. the next router C should declare the common setof its free rsoures identical with A to its next D. Repeat this again and again until it reaches theegress router or break at certain middle point, thus a light path can be built after acknowledgingone by one .2 Experiment and resultsTo verify the reasonableness and feasibility of the designing principle，we carry out an ex-periment. Because of the insufficient equipment， there is only one node in our scheme. WAPP'rolocol is also simplified (see fig. 4). First, simulation software of the node generates an IP data flow, then. this flow is modulated to the default wavelength and converted to electrical domainafter lransmission . The electrical signal is processed by DSP . The process is flow identifying thentraditional routing. Initially, the routed packet flows need to be modulated on a certain wave-length and output through self-routed A WG. However. when DSP detects a continuous flow des-tined to another node A. the node can transmit the packets to A diretly through a wavelength de-termined by WAP. When the data to A is sent out, the node releases the resource and returmsback to default wavelength .Feedback_OEKouting algonithmWAP:lectrcalsignal1:dge node EOF DenIux>or:- +DSP bourdOpticalTunable laser.Bypass IP flow~ Routed IP packelsAWGWideband opticalspectrumF >[Tunablc fiterOXC(Wavelength sel-outed)Fig. 4. Schematics of experiment.The devices used are one A WC, one demultiplexier, two wavelength tunable lasers ( substi-tuted by wideband laser and tunable optical filter), one DSP EVM board and an IP source ( soft-ware). IP flow identifier uses simple X/Y method and the routing selection is still the longest-prefix- match. IP souree generales flows with self-similar property and the destination ditributionfollows random property， such as uniform distribution.One of the experiment resuls is observed on OSA (fig. 5). Fig. 5(a) shows that IP flowinpuls DSP via default wavelength 1549.3 nm. Fig. 5(b) shows that the IP flow is modulated on1552 .5 nm after processing and before input AWC, then output at the second port of AWC. Fig.5(c) shows the node tunes to 1550.9 nm，and input AWC and output also at the second port .3 Related problemsAlthough the node structure (fig. 4) does not seem complex，some key problems need to be中国煤化工MYHCNMHGPRINCIPIE ANAL.YSIS OF IP WAVEIENCTI ROLTFR77十1549.3 nm25-焉-30-35-10+(a15301S40 1:50 1S60 1S70 1580Waelength /mm-15一15525 nm .-20-- + 1550.9 nm-35--3-451530 1540 1550 1:60 1570 15801530 1540 15S0 1560 1570 1580Wavcelength /nmWavelength /mFig. 5. Exprimerntal resuls .further researched .The most important problem of all is that we need exactly model the characteristics of the re-alistic IP traffie . Second，low identification, an open problem of rouler design, is expected to bemore acurale. These lwo are the premise of router ”s action.The 3rd problem is related with fast optical wavelength switching. Though it seems that fastoptical switching is most difficult to be realized in current days, in fact, most routes existing intoday' s Internet backbones are extremely stable, with average lifetime of several hours or evenlays *However. because the things may nol be the same in metropolitan , consequently, DPDPmay need further modification.'The 4th problem is aboul optical buffer. DPDP cannot use FDL (fiber delay line) as bufferin wavelengh routing condition, but virtual buffering provided by wavelength conversion can alsoalleviate output port collision. But because of the limitation of experiment condition , wavelengthconversion is not introduced in our experiment ."he special difficuly is coming from the fth problem: wavelength merging. Because wewant to use the wavelength thoroughly, once there are several Aows destined l0 the same egressrouter, occupying the same links, we often think that a belter way is merging different flows intothe wavelength, but as we all know ，traffie multiplexing into one channel in pure optical domainis hardly possible ，therefore ，how to accomplish wavelength merging still needs more study. Ourmethod is dividing 2.5 Cb/s into two 1 .25 Cb/s channels or four STM-4 (622 Mb/s) channels ,中国煤化工YHCNMHG47kSCIENCE IN CHINA (Series F)Vol. 44if the nole rale is larger than 1 .25 Gb/s, then it oceupies one whole wavelength, if the node rateis larger lhan 622 Mb/s while less than I .25 Cb/s, the node obtains only one I .25 Cb/s chan-nel, then it can merge with another node owning another I .25 Gb/s channel by searching frameheader, shift and opical multiplex. Although this method is nol smart compared with electricalMPIS, it is nol so complex as the latter.The last problem is related with possible modification of IP upper layer. Because there isstill a wonder about whether or not the TCP protocol , which contains flow and congestion control,will perform well under multi-gigabit sped in wide-area networks, if TCP must be modified, theDPDP should adjust accordingly.4 ConclusionTo upgrade the IP infrastructure ，we design IP wavelengh router based on DPDP. We provethe reasonableness and feasibility of IP wavelength router by a prineiple experiment. The experi-mental results confirm that IP wavelength router is compalible with traditional IP router by DPDPmethod and easily expanded to Terabit or Petabit optica! router. We believe that our work is help-ful for the designing of 3rd generation optical router.Acknowledgements This work wus supponted by the National Natural Science Foundaion of Ching (Grant No. 69990540) .ReferencesRamaswamn, R.. Sivarajan, K. N.. Routing and wavelngth assignment in ll-optical nelwarks. IEEE Journal of Selerted Ar-pas in Conmunications, 1995. 14(5): 489- -500.2. Ramaswarni, R.. Sivarajan, K. N.. Optiral Networks: A Partiral Perspective. San Franciseca; Morgan Kaufmann Publish-rm. FPebruary 1998 .Chatng, G. K.. Elinar, G.，Meagher, B. et al.. A Profu-ol-Concept. Ulrna-low latenry Optical Label Switching TrstbedDemonstration for Next Generation Internet Nelworks, WD S-1, OFC' 2000, Rattimore: 0SA, 200Ghani. N.. larmxla labeling: A frarnrwurk for IP-Over- w DM using MPLS, Optieal helwurks Mag.. 2000, 1(2): 45- 57.Ghumi, N., Dixit, s. , (Channel prorisioning for higher layer protocols in WDM networks. SPIE Proc. on Al-pira Network.ing, 1996. Ramaswami. R.. Muliavelengh networks for computer communicaion, IEEE Commun Mag.. 1993, 31: 78- -88.. Iartnidgr. C.. Carey. P. P., Burgess. E.. A 50-Cb/s IP router, IEEE/ ACM Trans. on Networking, 1998. 6(3): 237248.8. Giwindan. R.. Reddy, A.. An analysis of internet domain topology and route stbility.y Infocom' 97. Kobe, Japan, April,Ine Alanitos: IEEE Computer Society Phess, 1997, 850- 857.--- ，中国煤化工MYHCNMHG