Packet-selective photonic add/drop multiplexer and its application to ultrahigh-speed optical data networkings in LAN and MAN

Transcription

1 Packe-selecive phoonic add/drop muliplexer and is applicaion o ulrahigh-speed opical daa neworkings in LAN and MAN Ken-ichi Kiayama 1) and Masayuki Muraa 2) Osaka Universiy 1) Deparmen of Elecronics and Informaion Sysems, 2-1 Yamadaoka, Suia, Osaka , Japan, ( kiayama@comm.eng.osaka-u.ac.jp) 2) Cyberm edia Cener, 1-30 Machikaneyama, Toyonaka, Osaka , Japan, ( muraa@ics.es.osaka-u.ac.jp) Key words: packe ransmission, asynchronous access, opical code correlaion, perfor m- ance modelling, opical implemenaion Absrac: We propose packe-selecive phoonic add/drop muliplexing (P-ADM), which can deermine he source and de sinaion addresses in he packe header based upon opical code correlaion and is novel applicaion o ulrahigh-speed opical daa neworking in LAN and MAN. I is disinc from convenional wavelengh ADM (λ-adm) in he raffic granulariy, in ha P-ADM can individually handle he packes while λ-adm handles he whole raffic ogeher on a single wavelengh. As an applicaion of P-ADM o opical daa neworking, a rae-conrolled asynchronous access is sudied, and he performance analysis and numerical examples are presened. The opical implemenaion of P-ADM is also presened, in which he phoonic label selecor, srucured wih a fiber Bragg graing (FBG), is a key o he ulrahigh -speed processing capabiliy up o a few ens of Gpps (packe per second). 1. INTRODUCTION There is a growing demand for high-bandwidh and differeniaed daa services. Given he inherenly bursy naure of daa raffic, he fixed-bandwidh TDM links is no an efficien soluion. Opical daa neworking, which uses 1

2 2 Ken-ichi Kiayama and Masayuki Muraa WDM links as he opical channels [1, 2], is an alernaive approach o he TDM-based daa neworking. Opical daa neworking eliminaes unnecessary proocol sack in each nework layer, hus drasically reducing he cos and complexiy of neworking [3, 4]. In opical daa neworking, he opical add/drop muliplexer (ADM), locaed a access nodes, will be key device. A wavelengh add/drop mul i- plexer (λ-adm) enables each wavelengh o eiher add/drop and elecron i- cally processes a he node or o opically bypasses he node elecronics (Fig.1(a)). This capabiliy of opically bypassing a node s elecronic layer can be exploied o reduce overall elecronic processing, resuling in he reducions of boh he ransfer delay and he number of elecronic digial crossconnec swiches and rouers. λ-adm can be adoped for differen ypes of raffics such as saic or dynamic circui swiching, packe swiching, and flow swiching [5]. However, λ-adm is of granulariy of wavelengh, and his imposes a limi ha i can only handle raffic on an opical channel, ha is, a wavelengh pah, and i is unable o handle raffic on a packe-bypacke basis. This may wase he wavelengh resource and limi he usage of opical channels. On he oher hand, convenional SONET ADM has a finer granulariy for handling raffic. I picks ou low-speed sreams from a highspeed sream and likewise, adds low-speed sreams o a high-speed sream [6]. However, he speed of elecronic processing will evenually become a boleneck o he overall processing capabiliy as he bi rae of he WDM link goes higher. Hence, here would be a growing demand for high-speed muli-funcional ADM in he fuure opical daa neworking, which can handle raffic in finer granulariy.

3 Packe-selecive phoonic and is applicaion... 3 Node λ 2 λ-adm λ 2 λ 1 λ 1 λ 1 Drop Add Fig. 1(a) Wavelengh add/drop muliplexer (λ-adm) There have been a variey of opical packe swiching schemes, in which he packe label are mapped in opical domain and processed opoelecronically [7-12]. Very recenly, phoonic rouer for Inerne Proocol, IP packes has been proposed and demonsraed [13-15]. In he phoonic IP rouer, he phoonic labels are mapped ono a family of opical codes, which have been uilized opical code division muliplexing (OCDM) echnique [16-18]. OCDM is a class of ransmission and muliplexing echniques oher han TDM, WDM, and space division muliplexing (SDM). OCDM inheris unique feaures from he wireless code division muliple access (CDMA) [19]. I was confirmed ha as he encoding and decoding are performed in opical domain, he packe processing capabiliy exceeds an order of Gpps. In his paper, packe-selecive phoonic add/drop muliplexing (P-ADM), which can deermine he source and desinaion addresses in he packe header based upon opical code correlaion, and is novel applicaion o ulrahigh-speed opical daa neworking in LAN and MAN are proposed. I is disinc from convenional wavelengh ADM (λ-adm) in he raffic granulariy, in ha P-ADM can individually handle he packes. On he conrary, λ-adm handles he whole raffic ogeher on a single wavelengh. In he applicaion of P-ADM o opical daa neworking, a rae-conrolled asynchronous access is sudied, and he performance analysis and numerical ex-

4 4 Ken-ichi Kiayama and Masayuki Muraa amples are presened. The opical implemenaion of P-ADM is also presened, in which he phoonic label selecor, srucured wih a fiber Bragg graing (FBG), is a key o he ulrahigh-speed processing capabiliy up o a few ens of Gpps (packe per second). 2. CONCEPT OF PACKET-SELECTIVE PHOTONIC ADD/DROP MULTIPLEXER (P-ADM) 2.1 Principle of operaion A packe-selecive P-ADM is locaed a a node, and i drops o he node or bypasses he node he arriving packes or adds he generaed packes from he node (Fig.1(b)). I handles raffic on a single WDM link on a packe-by packe-basis. The basic archiecure is illusraed in Fig.2. The fundamenal funcions of P-ADM include phoonic label selecion and opical swiching. The inpu and oupu pors are conneced o a single WDM link. The phoonic label in he header of arriving packe is processed, and according o he label he opical swich is driven by he conrol signal o direc he packe o he ougoing por eiher for dropping or bypassing.

5 Packe-selecive phoonic and is applicaion... 5 λ 1 P-ADM λ 1 Drop Add To node From node Fig. 1 (b) Proposed packe -selecive phoonic ADM (P -ADM)

6 6 Ken-ichi Kiayama and Masayuki Muraa P-ADM Phoonic label selecor Backbone WDM link Inpu packes 2x2 opical swich Bypassed and added packes Add Drop Fig.2 Basic archiecure of packe -selecive phoonic ADM (P-ADM) module 2.2 Phoonic label selecor The opical packe is srucured as shown in Fig.3. The phoonic label in he header is simply a relaively shor, fixed-lengh idenifier ha bears he informaion of he desinaion address and source address. A he end of he packe a flag ha indicaes he end mus be aached. The oher overhead informaion could be included in he label upon he reques. The remaining nework conrol informaion will be encapsulaed in he frame of higher layers. As shown in Fig.4 each node is uniquely assigned wih a label, and each label is mapped ono an opical code [13]. The opical code is a sequence of opical pulses, so-called chip pulses. The chip iself is a shor pulse, and he ime duraion of he sequence has o be wihin a bi ime duraion. The phoonic label selecion is based upon he opical correlaion beween he opical code of arriving packe and he assigned code of he node. The opical correlaion is performed by mached filering in ime domain. For example, a family of opical codes of 4-chip long codes and heir auo-correlaion waveforms are illusraed in Fig.4. The code is bipolar, in which he phase of opical carrier of individual chip pulse akes wo saes of eiher 0 or π, represening binary value of 1 or -1, respecively. As increasing he num-

7 Packe-selecive phoonic and is applicaion... 7 ber of nodes, he longer code will be required. Unique o he label selecor is ha no opical logic operaion is involved, and his is a key o he ulrahighspeed operaion of P-ADM. Header Desinaion/source Phoonic labels Oher informaion Payload End flag Guard ime Fig.3 Opical packe

8 8 Ken-ichi Kiayama and Masayuki Muraa I J K Auo-correlaion Opical code i Opical code j Opical code k Cross-correlaion Node I Node J 0π0π Auo-correlaion Opical code i Opical code j Cross-correlaion Cross-correlaion Node K Opical code k Auo-correlaion Fig.4 Address assignmen o each node. A family of 4-chip long, bipolar opical codes and heir auo- and cross-correlaion funcions 2.3 Key building block: P-ADM module We will show he building blocks of P-ADM shown in Fig.2. The fundamenal funcions of P-ADM, including phoonic label selecor and opical swiching can be implemened by using opoelecronic devices. Firs, he phoonic label selecor is srucured wih an opical correlaor, a ime-gae, an O/E conversion, and hresholding (Fig.5(a)). Promising device for he opical correlaor is a fiber Bragg graing (FBG) (Fig.5(b)). The idenical opical device mus be used for he opical encoder. An advanage is ha a single FBG can selec he phoonic label as well as he wavelengh, if he graing period is uned o a desired wavelengh. A super-srucured FBG can generae bipolar opical codes [20]. Several high-speed off-he-shelf opical space swiches have now become available. The swiching speed of semic onducor opical amplifier (SOA) gae swich is in he sub-nanosecond regime, which is well beyond he requiremen for he swiching speed. Unforunaely, here is no random access opical memory, equivalen o elecronic RAM. The opical fiber delay line is only an opical device ha emporarily buffers an opical packe.

9 Packe-selecive phoonic and is applicaion... 9 (a) Inpu header Opical correlaor Time - gae O / E Time-spreading 0 π π 0 Time-despread By mached filering Time gaing τ 0 T 0 T 0 T 0 T T: One bi ime duraion (b) Inpu (λ 1,OC1) (λ 1,OC2) (λ 2,OC1) (λ 2,OC2) FBG Λ B =λ 1 /2n (λ 1,OC1) Fig.5 (a) Configuraion of phoonic code selecor. (b) FBG-based opical correlaor 3. APPLICATION TO OPTICAL DATA NETWORK- INGS IN LAN AND MAN: LINE ACCESS BY RATE-CONTROLLED ASYN CHRONOUS PACKET TRANSMISSIONS 3.1 Operaion mechanisms For daa ransmissions we wan o allow he bursy ransmission of packes in LAN and MAN, by which a flexible use of bandwidh is expeced. For his purpose, we will propose a rae-conrolled line access mehod, and we will implemen his by using P-ADM. We use a leaky-bucke mechanism [21] o conrol he packe emission ono he line a each node. The leaky-bucke is a virual device o conrol he bursiness of he flow. As shown in Fig.6, okens arrive a he oken pool wih rae ρ and are buffered in he oken pool if he oken pool is no full. If an arriving packe finds he oken pool nonempy, i depars he sysem insananeously, and a he same ime one oken is removed from he oken pool. The packe is dis-

10 deermines 10 Ken-ichi Kiayama and Masayuki Muraa carded if an arriving packe finds he oken pool empy. When a oken is generaed, and he packe buffer is nonempy, he packe on he op of queue depars he buffer immediaely by removing one oken. The deph of he oken pool ÿ he bursiness of he flow; he larger size of he oken pool allows more bursy packe ransmissions. Token wih rae ρ Bucke dephs Token pool Buffer (size K) Packes Fig.6 Logical srucure of leaky bucke Rae conrol using leaky-bucke For a realisic scenario of he rae-conrolled line access, we need o ake accoun of he dynamics of he leaky-bucke. The leaky-bucke parameers should be deermined for a given disribuion of packe arrival rae. Since he packe buffer is finie, he packe loss probabiliy should be esimaed when deermining he leaky-bucke parameers. For his purpose, we inroduce N and C ha represen he line capaciy and he number of erminals conneced o he line, respecively. Assuming ha he packe lengh is fixed, he following procedure hen deermines h e leaky-bucke parameers. 1. Assume ha he raffic load disribuion is known. Le he mean rae be λ i,j N 1 for he packe flow from node i io node j, and le λ i = λ i, j. j=i+1

11 Packe-selecive phoonic and is applicaion The admissible raffic load a each node is hen deermined by aking ino accoun of he relaive use of he line capaciy. In he case of he balanced raffic, i is simply deermined as C λ i N 2 / 4 (1) where N 2 /4 is he number of connecions on he mos congesed link. Noe ha oher bandwidh allocaion can also be considered by, e.g., aking ino accoun he fair allocaion of packe loss probabiliies among nodes. For his purpose, however, we need o formulae some opimizaion problem. 3. The admissible load given by Eq.(1) deermines he leaky-bucke parameer ρ i for node i. 4. Analyze he leaky-bucke o deermine he packe loss probabiliy and waiing ime a he packe buffer for given packe arrival disribuions. Noe ha we need more sric argumens in performing above procedures. Those will be described in he nex subsecion by considering he case where he packe arrivals follow a Poisson disribuion. In he above, we sill have a freedom in deermining he burs size specificaion σ i. By larger σ i, he packe loss probabiliy would be decreased by allowing more packes, which leads o he performance improvemen a ha node. However, he upsream nodes are also allowed he bursy packe ransmission if hose nodes are assigned larger σ i (j < i). Then, bursy arr i- vals of packes from he upsream nodes inhibi he packe ransmission a node i, which migh cause he performance degradaion for node i packes. Once we are given a permissible packe loss probabiliy a each node, however, we can deermine he bucke size and hen obain he maximum packe delay a he node in order Analysis of leaky-bucke In his subsecion, we presen an analysis for he leaky-bucke a each node, which is a sligh exension of he approach shown in [21]. Packes arrive a he packe buffer according o a Poisson process wih rae λ. Noe ha we will omi he subscrip represening he node number in he below.

12 12 Ken-ichi Kiayama and Masayuki Muraa Node 0 Node 1 Node 2 Node N-1 From upsream nodes Token wih rae ρ 2 To downsream nodes Bucke size σ 2 From local nodes (rae λ 2 ) Packe buffer (size K) Fig.7 Flow enforcemen by leaky-bucke See Fig.7. Consider a sloed ime axis where each slo is of lengh 1/ρ, and a new oken is generaed a each slo boundary. A newly arriving oken joins he oken pool if he pool conains fewer han σ okens; oherwise i is discarded. We will derive he seady sae join probabiliy disribuion of he queue size and he oken pool occupancy a slo boundaries, jus prior o he generaion of a new oken. Noe ha we neglec an asynchronous operaion of our sysem, bu i is only for modelling purpose. We inroduce P(m, i) he seady sae probabiliy of having m okens in he oken pool and i packes in he packe buffer. The number of okens canno exceed σ, and he packe should wai in he buffer if and only if here exis no okens in he oken pool. Thus, we have P( m,i)= 0, m > σ ; 1 m σ and i 1. (2) For convenience, we inroduce p i as p i = P( σ i,0), 0 i σ (3) P( 0,i σ), σ i

13 Packe-selecive phoonic and is applicaion As described in he above, packe arrivals a he local node are assumed o follow a Poisson process. We furher assume ha arrivals of ransi packes from he upsream nodes desined for he downsream nodes also follow a Poisson process wih rae λ u, which is given by i 1 N 1 λ u = λ j,k (4) j=0 k=i+1 for node i. We noe here ha in acual, we need o ake accoun of he deparure processes from he leaky-bucke a he upsream nodes, which is reaed as an inpu process a he curren node. However, since he variance of he deparure process ends o become smaller han he inpu Poisson process by he leaky-bucke, we can esimae he wors case by he above assumpion. We inroduce a i and b i o represen he probabiliies of i arrivals during a slo from he local node and from he upsream nodes, respecively, i.e., a i = e λ/ρ ( λ / ρ) i i! ; b i = e λ /ρ u ( λ u / ρ) i i! Furher, we inroduce a i and b i as i a i = 1 a j ; b i = 1 b j (6) j= 0 i j=0. (5) Since we consider he finie capaciy of he packe buffer (denoed by K), a packe ha arrives and finds he buffer full is discarded. The ransi packes are given higher prioriy since we assume o have no means o sore he packes in an opical domain. Accordingly, we consider ha he ransi packe ges he oken from he oken pool a he local node and leaves he node. The seady sae equaions are hen given by i i j p i = p 0 b k a i k + p j+1 b k a i j k, k= 0 i j=0 k=0 0 i σ + K 1. (7) The above equaions can be solved by firs assuming a value for p 0, and recursively compuing p i s (1< i <σ+k) using

14 14 Ken-ichi Kiayama and Masayuki Muraa p i = 1 i 1 i 2 p a 0 b i 1 p 0 b k a i 1 k p j+1 0 k =0 All quaniies are finally normalized wih σ +K p i i=0 = 1. (9) j=0 i 1 j The hroughpu a he designaed node is hen obained as k=0 b k a i 1 j k. (8) T = p 0 σ +K + p j j=1 σ +K 1 b k k=0 σ + K 1 b k k= 0 σ +K k b k i=1 + ( σ + K k)a σ +K k ia + ( σ + K j k + 1)a i σ+k j k+1 i=1 σ + K j k+1 (10) Finally, we obain he packe loss probabiliy a he node as P loss = 1 T λ / ρ. (11) We can also derive waiing imes for he packe sored a he buffer by applying he approach aken in [21], alhough we omi he resul. In numer i- cal examples, we will only show he maximum waiing ime of he packes ha are sored in he buffer wihou los. I is simply given as W max = σ / ρ. (12) We las noe ha we can also consider he bursy raffic such as an ON OFF process, bu for ha purpose, we need a more exension. See, e.g., [22]. The numerical examples for he packe loss probabiliy will be shown in Subsecion Opical implemenaion The proocol for accessing he line is shown in Fig.8. If he arriving packe a node j has he desinaion label of node j, he packe is dropped from swich 1. If he arriving packe has a differen desinaion from node j, and if he oken pool is nonempy, i is bypassed o downsream nodes. However, he arriving packe having a differen desinaion from node j is

15 Packe-selecive phoonic and is applicaion discarded when he oken pool is empy. We will pu a higher prioriy on he bypassing raffic han he raffic o be added from he node k, herefore, he packe ransmission from node j is awaied during he burs from node i (<j) passes by node j. A node j Desinaion label j? Yes No Leaky-bucke conroller sw1 Token nonempy? Drop Yes No sw2 sw3 Packe from is own node j Add < Prioriy Bypass Discarded Fig.8 Access proocol of asynchronous rae-conrolled packe ransmission The implemenaion of he P-ADM is shown in Fig.9. I consiss of a phoonic label selecor, hree 1x2 opical swiches, wo opical delay lines, and an elecronic leaky-bucke conroller. The phoonic label selecor direcly conrols opical swich 1, and he conrol signal from i is also fed o he leaky-bucke conroller. The leaky-bucke conroller conrols opical swiches 2 and 3. The delay ime τ of he opical delay line has o be se equal o he processing ime required for phoonic label selecor, while, he delay ime τ1 is se equal o he processing ime required for he leaky-bucke conroller. Opical swich 1 drops he arriving packe when i urns o he cross-sae, oherwise i says in he bar-sae o bypass he packe. Opical swich 3 says in he bar-sae o bypass he arriving packe, and i urns o he cross-sae o discard he arriving packe. Opical swich 2, conrolled by he leaky-bucke conroller, is allowed o urns o he cross-sae o add he packe from he buffer only when here is no bypassing packe, oherwise i says bar-sae o bypass he packe. To avoid he collision beween he by-

16 16 Ken-ichi Kiayama and Masayuki Muraa passing packe and he adding one, he second opical delay line buffers he bypassing packe for one packe duraion. I is noed ha he leaky-bucke conrol is very simple; only regiser fech and add operaions are necessary o implemen he leaky-bucke mechanism. When we consider 100 Gbps ransmission line and 1,000 bi-long packe as described in he numerical examples in 4.1, a packe ransmission ime becomes 10ns. Thus, i is sufficien o execue several insrucions for performing he leaky-bucke algorihm even by he curren processor echnology. Node j Phoonic label selecor j Inpu Leaky-bucke conroller k Bypass & add τ τ 1 sw1 sw3 sw2 E/O Buffer Discarded Drop Add Fig.9 Opical implemenaion P-ADM for asynchronous rae-conrolled packe ransmission 4. DISCUSSIONS 4.1 Numerical examples of rae-conrolled asynchronous packe ransmissions By using he analysis mehod in Subsecion 3.1, we now provide several numerical examples. We consider he balanced raffic, i.e., λ=λ ij for all (i, j) hroughou his subsecion. The line capaciy C and he packe lengh are fixed a 100 Gbps and 1,000 bis, respecively. The oken generaion rae a

17 Packe-selecive phoonic and is applicaion node i is deermined by aking ino accoun wo kinds of raffic flow s; (N-I- 1) connecions are originaed a node i and desined for he downsream nodes. Oher i (N-i-1) connecions are ransi flows which are originaed a he upsream nodes and desined for he downsream nodes. In Fig.10, we firs show he packe loss probabiliies dependen on he node posiion of he line for N = 12. In he case of he rae conrolled access (abbreviaed as RCA in he figure), he arge raffic load for each connecion is se o be C/36 where 36 is he oal number of ransi and originaing connecions a node 5. Three cases for he buffer size are shown; K = 10, 15 and 20. The oken pool size (or bucke deph) σ is se o be 10. The raffic load is 2.08 Gbps for each connecion, which corresponds o 75% of he bandwidh allocaed o each connecion in he pseudo TDMA sysem. The packe loss probabiliies are slighly increased a downsream nodes in he case of he rae-conrolled access sysem. I is because in he downsream node, he raffic from he upsream node becomes increased, and i suppresses he packe ransmission from he local node. For example, in he case of he buffer size K = 20, he maximum waiing ime a he buffer ranges from µs (a node 5) o µs (a node 0 and 10) in he case of he rae-conrolled access sysem. By considering he ime duraion of 1,000 bilong packe is 0.01 µs for he bi rae of 100 Gbi/s, he waiing ime would be reasonable. We nex show he packe loss probabiliies dependen on he raffic load in Fig.11. The horizonal axis shows he raffic load a each node. We plo he wors case, i.e., he packe loss probabiliies a (N-2)nd node. The buffer size K is changed as 10, 15 and 20 in he figure. Las we show he packe loss probabiliies dependen on he number of nodes on he line in Fig.12. For ha purpose, we again fix he raffic load a 1.0 Gbps; idenically se for all connecions. Thus, as he number of oal nodes is increased, he raffic load a each link ges large and he performance is degraded in boh wo cases. The packe buffer size is se o be 10, 15 and 20. The oken pool size of all nodes is idenically se o be 10. All resuls are for (N-2)nd node.

18 18 Ken-ichi Kiayama and Masayuki Muraa 1 1E-02 Packe Loss Probabiliy 1E-04 1E-06 K=0 K=15 1E-08 K=20 1E Node Number Fig.10 Packe loss probabiliies a nodes

19 Packe-selecive phoonic and is applicaion E-02 Packe loss probabiliy 1E-04 1E-06 1E-08 K=10 K=15 1E-10 K= Traffic load [pk/ms] Fig.11 Packe loss probabiliies vs. raffic load

20 20 Ken-ichi Kiayama and Masayuki Muraa 1 1E-02 Packe loss probabiliy 1E-04 1E-06 1E-08 RCA (K=10) RCA (K=15) 1E-10 RCA (K=20) Toal number of nodes Fig.12 Packe loss probabiliies vs. he number of nodes 4.2 Perspecive for ulrahigh-speed opical daa neworking The rae-conrolled access has a disinc advanage over convenional mul i- ple access proocols because i adops a sor of so-called open-loop conrol [23]. Anoher is a closed loop conrol where he conrol line (or conrol slo) is used o reserve he nex ransmission on he line. Such a conrol mehod can be found in, e.g., Disribued Queue Bus (DQDB). The minimum packe size in Eherne has o be a leas wice he propagaion delay in bis so ha he node will hear a collision before i sops sending and misakenly assume he ransmission was successful. As increasing he bi rae, more and more bis can be packed ino he fiber. For example, he minimum packe size for 100Gbps is 10 5 bis for he propagaion delay of only 100m-long opical fiber. Such a large packe size is impracical and herefore, he access echniques such as GbE (E: Eherne), 10GbE, and DQDB will have a difficuly in achieving high performance for a long disance because heir performance can be easily influenced by he ransmission lengh. On he conrary, he performance of he rae-conrolled access is by no means liable o he ransmission lengh, and his characerisic can be exploied for ulra high-speed access.

21 Packe-selecive phoonic and is applicaion There is also an advanage in he rae-conrolled access compared wih opical TDMA (OTDMA), which is a muliple access proocol in opical layer [24]. A ulrahigh-speed OTDM has been realized up o a hundred Gbps [25]. All-opical regeneraive OTDM add/drop muliplexing a 40Gbps has been maerialized [26]. OTDMA is no as efficien for bursy daa ransmissions as he rae-conrolled asynchronous access because i uses a imesloed access proocol. Ulimae performance of he rae-conrolled asynchronous access will be prediced from a viewpoin of is opical implemenaion. The overall processing speed of P-ADM is limied by he capabiliy of he phoonic label selecor, which is governed by he propagaion delay required for he opical correlaion. I has been experimenally demonsraed ha opical packe consising of phoonic label using 8-chip-long opical code is processed, and opical packe of 64bi-long payload daa a 10Gbi/s is roued via opical swich [14]. The processing capabiliy reaches 29Gpps wih he ime duraion of phoonic label of 35ps. Assume ha he packe lengh is 1,000 bi, P- ADM can be inerfaced wih he link of 29Tbps. The accessible number of nodes depends upon he lengh of opical code. For he chip pulsewidh of 1ps he maximum code lengh will become 5ps for he bi rae of 100Gbps. For 5-chip long opical code, he number of available opical codes are =16. Noe ha he facor of -1 is due o he eliminaion of symmery code sequences. 5. CONCLUSION Packe-selecive phoonic add/drop muliplexing (P-ADM) using phoonic label selecor based upon opical code correlaion and is novel applicaion o ulrahigh-speed opical daa neworking have been proposed. P-ADM provides a finer granulariy of raffic han convenional λ-adm, and i can handle he raffic on a packe-by-packe basis. The opical implemenaion P- ADM has been presened by using off-he-shelf opical devices such as FBG and opical gae swich, and he ulrahigh-speed processing capabiliy as high as 30Gpps has been prediced. The applicaion o opical daa neworking in LAN and MAN, he rae-conrolled asynchronous access has been sudied, and he numerical examples have confirmed performances of low packe loss probabiliy and low laency. Finally, anoher promising applicaion of he packe-selecive P-ADM is o he nework proecion and resoraion. P-ADM will be able o provide he proecion o individual raffic on a wavelengh pah. Hence, his will provide wih differeniaed class of proecions; each raffic has differen prio r- iy of proecion and resoraion.

22 22 Ken-ichi Kiayama and Masayuki Muraa References: [1] C. A. Bracke, Dense wavelengh division muliplexing neworks: Principles and applicaions, IEEE J. Selec. Areas in Commun., vol. 8, pp , [2] K. Sao, S. Okamoo, and H. Hamada, Nework performance and inegriy enhancemen wih opical pah layer echnologies, IEEE J. Selec. Areas in Commun., vol.12, pp , [3] P. Bonenfan and A. R. -Moral, Opical daa neworking, IEEE Commun. Magazine, pp.63-70, March [4] M. Muraa, Challenges for he nex-generaion Inerne and he role of IP over phoonic neworks, o appear in IEICE Trans. Commun. vol.e83-b, No.8, [5] E. Modiano and A. Narula-Tam, "Mechanism for providing opical bypass in WDM-based neworks," Opical Neworks, vol.1, pp.11-18, [6] R. Ramaswami and K. N. Sivarajan, Opical Neworks -A Pracical Perspecive-, S. F.: Morgan Kaufmann Publishers, Inc, 1998, ch.8. [7] Y. Takushima andk. Kikuchi, "Phoonic swiching using spread specrum echnique," Elecron. Le., vol.30, pp , [8] M.D. Vaughn and D.J. Blumenhal, All-opical updaing of subcarrier encoded packe headers wih simulaneous wavelengh conversion of baseba nd payload in semiconducor amplifiers, IEEE Phoonic Technol. Le., vol.9, pp , [9] A. Carena, M.D. Vaughn, R. Gaudino, M. Shell, and D.J. Blumenhal, OPERA: An opical packe experimenal rouing archiecure wih label swiching capaciy, J. Lighwave Technol., vol.16, pp , [10] X. Jiang, X. P. Chen, and A.E. Willner, All opical wavelengh independen packe header replacemen using a long CW region generaed direcly from he packe flag, IEEE Phoonic Technol. Le., vol.10, pp , [11] M. C. Cardakli, D. Gurkan, S. A. Havsad, A. E. Willner, Variable-bi-rae header recogniion for reconfigurable neworks using unable fiber-bragg-graings as opical correlaors, 2000 Opical Fiber Conference (OFC2000 ), TuN2 (Balimore, 2000). [12] W. I. Way, Y.-M. Lin, and G. -K. Chang, A novel opical label swapping echnique using erasable opical single-sideband subcarrier label, 2000 Opical Fiber Conference (OFC 99), WD6 (Balimore, 2000). [13] K. Kiayama andn. Wada, Phoonic IP rouing, IEEE Phoonic Technol. Le., vol.11, pp , [14] N. Wada and K. Kiayama, Phoonic IP rouing using opical codes: 10Gbi/s opical packe ransfer experimen, 2000 Opical Fiber Conference (OFC 99), WM51 (Balimore, 2000). [15] H. Soobayashi and K. Kiayama, Opical code based label swapping for phoonic rouing, o appear in IEICE Trans. Commun. vol.e83-b, No.8, [16] J. Salehi, "Code division muliple access echniques in opical fiber neworks-par I: Fundamenal principles," IEEE Trans. Commun., vol.37, pp , [17] P. Prucnal, M. Sanro, and T. Fan, "Spread specrum fiber opic local area nework using opical processing," IEEE/OSA J. Lighwave Technol., vol.4, pp , [18] K. Kiayama, Code division muliplexing lighwave neworks based upon opical code conversion, IEEE Seleced Areas in Commun., vol.16, pp , [19] K. Kiayama, H. Soobayashi, and N. Wada, Opical code division muliplexing (OCDM) and is applicaions o phoonic neworks (INVITED), IEICE A., vol.e82-a, pp , 1999.

23 Packe-selecive phoonic and is applicaion [20] P.C. The, P. Peropoulos, M. Ibsen, and D. J. Richardson, "A 10Gb/s, 160Gchips/s OCDMA coding: decoding sysem based on supersrucured fiber graings," 2000 Opical Fiber Communicaion Conference (OFC2000), PD9 (Balimore, March 2000). [21] M. Sidi, W.-Z. Liu, I. Cidon, and I. Gopal, Congesion conrol hrough inpu rae regulaion, IEEE Transacions on Communicaions, vol. 41, March [22] M. Muraa, Y. Ohba, and H. Miyahara, Analysis of flow enforcemen algorihm for bursy raffic in ATM neworks, Proceedings of IEEE INFOCOM 92 (Florence, Ialy), pp , May [23] H. Ohsaki, M. Muraa, H. Suzuki, C. Ikeda, and H. Miyahara, Rae-based con gesion conrol for ATM neworks, ACM SIGCOMM Compuer Communicaion Review, vol. 25, pp , April [24] S-. W. Seo, K. Bergman, and P. R. Prucnal, Transparen opical neworks wih imedivision muliplexing, IEEE J. Seleced Areas in Commun./J. Lighwave Technol., vol.14, pp , [25] S. Kawanishi, 120Gbi/s OTDM sysem prooype, 24h European Conference on Opical Communicaions (ECOC 98) ), PD, p.41 (Madrid 1998) [26] A. Buxens, A. T. Clausen, H. N. Poulsen, P. Jeppesen, S. Fischer, M. Dulk, E. Gamper, W. Vog, W. Hunziker, E. Gini, and H. Melchior, "All-opical regeneraive OTDM add/drop muliplexing a 40Gbi/s using monolihic InP Mach-Zehnder inerferomeer," CLEO2000, CWD3, 2000.