Inter-doman Routng n Optcal Networks wth Wavelength Converters Anteneh Beshr, Marcelo Yannuzz, and Fernando Kupers Network Archtectures and Servces Delft Unversty of Technology Mekelweg 4, 2628 CD Delft, the Netherlands {A.A.Beshr, F.A.Kupers}@tudelft.nl Advanced Network Archtectures La (CRAAX) Techncal Unversty of Catalona Vlanova la Geltrú, Catalona, Span yannuzz@ac.upc.edu Astract Wth the ncreasng deployment of wavelengthdvson multplexng (WDM) optcal networks, the need for advanced lghtpath provsonng algorthms and protocols n a mult-doman settng s ecomng evdent. In order to ncrease effcency y relaxng the wavelength contnuty constrant n WDM optcal networks, wavelength converters are often placed at certan nodes n the network. In ths paper, we study the effcency of usng converters n a mult-doman settng. We have made smple ut mportant modfcatons to exstng optcal nter-doman routng protocols n order to utlze the power of wavelength converters and have tested ther performance. These modfcatons can e seamlessly ntegrated nto these protocols (.e., wthout changng ther algorthmc aspects) to sgnfcantly reduce ther lockng rato. We also show that there s a clear performance dfference among the consdered protocols. I. INTRODUCTION Optcal networks usng wavelength-dvson multplexng (WDM) technology are eng wdely deployed wthn domans. Future optcal networks wll requre new protocols n order to route and support on-demand provsonng of lghtpaths etween dfferent domans. Unlke tradtonal IP multdoman networks, the study of optcal mult-doman ssues s at a very early stage. One mportant ssue s what type of nformaton should e exchanged among neghorng domans n order to ncrease effcency. Prevous works [9] [10], have proposed approaches where neghorng domans are ale to exchange oth Network Reachalty Informaton (NRI), and hghly aggregated Path State Informaton (PSI). However, the presence of wavelength converters s not analyzed n these works. Our man contruton n ths paper s to seamlessly ncorporate modfcatons to the protocols proposed n [9] and [10], so that wavelength converters are utlzed. In WDM optcal networks wthout wavelength converters, a lghtpath has to use the same wavelength all along ts path. Ths mples that lghtpath requests may e locked, even though there are unused wavelengths. In order to decrease the lockng rato, wavelength converters are employed. Moreover, the optcal sgnal can e regenerated at converter nodes to extend ts reach. There are dfferent methods for sharng a pool of wavelength converters at a gven node among the wavelengths of ts dfferent fer lnks [4]. Due to ts sharng effcency, we assume a share-per-node approach, where there s a sngle ank of converters at a gven swtchng node shared y all ts lnks and only wavelengths that need to e converted are drected to ths ank. Snce wavelength converters are costly (yet usually more affordale than addng fers n already exstng networks), we assume that for nter-doman traffc n a gven doman, the wavelength converters are placed at order optcal crossconnects (OXCs). Ths assumpton s a realstc representaton of emergent mult-doman optcal networks [6]. Due to the large amount of traffc that goes through order OXCs, puttng wavelength converters at the order OXCs s expected to have a sgnfcant performance mprovement. II. RELATED WORK In the lterature, there are only few works dealng wth optcal mult-doman networks; there are even fewer works that study the effect of wavelength converters. The three relevant standardzaton odes, namely, the Internatonal Telecommuncatons Unon (ITU), the Internet Engneerng Task Force (IETF), and the Optcal Internetworkng Forum (OIF) have analyzed some of the topcs related to mult-doman optcal networks. In 2002, the OIF proposed the Doman-to-Doman Routng Protocol (DDRP). The drawacks of DDRP are that t represents a major change n the routng system and t s not sutale for path protecton. The IETF has proposed the generalzed mult-protocol lael swtchng (GMPLS) framework, whch extends the features of mult-protocol lael swtchng (MPLS) for provsonng crcut-swtched connectons va lael astractons for wavelengths, tmeslots, etc. The ITU- T has specfed a road-ased automatc swtched optcal network (ASON) framework. However, most of the research surroundng GMPLS and ASON s lmted to ntra-doman routng. OBGP (Optcal BGP) s an extenson of BGP that has een proposed to glue mult-doman optcal networks [1], [3], [8]. The strength of ths approach s that future optcal networks wll eneft from the advantages of the BGP-ased routng model, such as scalalty, clear admnstratve lmts of routng domans, etc. However, esdes nhertng the well-known dsadvantages of BGP, a mult-doman routng model manly ased on the exchange of network reachalty nformaton, whch s currently the case n BGP, may not e suffcent. Ths has ntated the proposal of dfferent path state aggregaton
2 schemes and updatng polces at the nter-doman level for WDM optcal networks [5] [9] [10]. In [9], the authors showed that y ntegratng only plan and hghly aggregated PSI n OBGP (n the form of an extended protocol called OBGP+), t s possle to drastcally mprove ts performance, wthout ncreasng the numer or the frequency of routng updates exchanged etween domans. In [10], a novel dstruted route control model s proposed, whch s ased on the deployment of nter-doman routng agents (IDRAs). We refer to the routng protocol runnng among the IDRAs as an IDRAs-ased routng protocol (IDRP). IDRP s ale to sgnfcantly reduce the lockng rato compared to that of OBGP. However, mechansms to take advantage of the presence of wavelength converters n these protocols were not developed. In ths paper, we make smple ut mportant modfcatons that wll allow OBGP+ and IDRP to eneft from the use of wavelength converters. The modfcatons are smple n that the algorthmc detals of these protocols are not affected, and they are mportant ecause a sgnfcant reducton n the lockng can e acheved due to these modfcatons. We also show the performance gan otaned y havng wavelength converters at order OXCs, and compare the performances of OBGP, OBGP+, and IDRP n the presence of wavelength converters. In Secton III, we gve a ref descrpton of OBGP+ and IDRP. In Secton IV, we show how these protocols can e modfed to take nto account the presence of wavelength converters. In Secton V, we present smulaton results comparng the performance of the three protocols and also the mprovement assocated wth havng wavelength converters at the order OXCs. Fnally, we gve conclusons n Secton VI. III. OBGP+ AND IDRP The major advantage of our approach s that our modfcatons can e seamlessly ntegrated n OBGP+ and IDRP. In other words, the algorthmc detals of these protocols can e reused snce our modfcatons concern only the wavelength aggregaton process. For completeness and n order to ntroduce the notaton used n Secton IV, we gve a ref ntroducton to OBGP+ and IDRP. For a detaled descrpton of these protocols, the reader s referred to [9] and [10]. OBGP+ s an mproved verson of OBGP n that PSI s advertsed esdes the usual NRI exchanged n OBGP; whereas IDRP s a novel optcal routng protocol that allows the exchange of useful traffc engneerng (TE) nformaton. A. Network Reachalty Informaton (NRI) NRI messages are trggered when a new destnaton ecomes avalale, or an already known one ecomes unreachale. The reachalty nformaton contaned n the NRI messages conveyed y OBGP+ conssts of: 1) The set of destnaton networks {d} and ther assocated autonomous system (AS)-path. 2) The Next-Hop (NH) to reach those destnatons,.e., the address of the ngress OXC n the neghorng doman from whch the advertsement was sent. 3) A set of pars (λ, W(λ )) avalale for each destnaton d, where λ denotes a partcular wavelength, and W(λ ) denotes the maxmum multplcty of λ. Unlke BGP/OBGP, the NRI exchanged among the IDRAs does not nclude the AS-path to reach a destnaton. In IDRP, rather than comparng canddate routes accordng to the length of the AS-path, the IDRAs use the TE nformaton contaned n the routng advertsements. B. Aggregated Path State Informaton (PSI) At a gven OXC, PSI messages aggregate () ntra-doman PSI; () PSI related to the nter-doman lnks towards ts downstream domans; and () the already aggregated PSI contaned n the nter-doman advertsements receved from downstream domans. In OBGP+, the PSI s ascally composed of aggregated wavelength avalalty nformaton. In IDRP, the PSI s not only composed of aggregated wavelength avalalty nformaton, ut t also contans aggregated load nformaton, whch s represented y assocatng a cost wth each canddate (path, wavelength) par [10]. For notaton purposes, we descre how the aggregated wavelength avalalty s computed. The aggregated wavelength avalalty nformaton s otaned y computng the Effectve Numer of Avalale Wavelengths (ENAW) for each type of wavelength, oth nsde an AS and across ASs. Insde an AS, the aggregaton process s as follows. Let u and v e a par of OXCs nsde an AS, P(u,v) e a canddate path etween u and v, and l e a lnk wthn the path P(u,v). The ENAW of wavelength type λ etween the OXCs u and v wthn a doman s computed as follows: { W u,v (λ ) = max P(u,v) mn [W l(λ )] l P(u,v) The ratonale ehnd eq. (1) s that the ENAW of a wavelength λ along a path P, whch s ascally the numer of lghtpaths that can possly e setup on P usng λ, s determned y the value of λ at the ottleneck lnk,.e., the lnk wth the mnmum numer of λ along P. Among all the paths etween u and v, the path wth the largest ENAW s chosen. The nter-doman part s composed of the unused wavelengths on the drectly-connected nter-doman lnks of the OXC, and wavelengths that are avalale downstream, whch are known through the PSI advertsements from neghorng OXCs. Let W l,l (λ ) e the ENAW of type λ etween OXC l and a local order OXC l,,r (λ ) e the numer of free wavelengths of type λ n the nter-doman lnk etween the local order OXC l and a remote order OXC r ; and Wr adv,d (λ ) e the ENAW of type λ etween the remote order OXC r and the destnaton OXC d, whch s advertsed y r or the IDRA of r. By comnng these nter-doman components and eq. (1), the OXC advertses to upstream neghors the ENAW etween the local order OXC l and the destnaton OXC d as: { } W adv l,d (λ ) = mn } W l,l (λ ),,r (λ ), W adv r,d (λ ) (1) (2)
3 IV. WAVELENGTH AGGREGATION IN THE PRESENCE OF WAVELENGTH CONVERTERS In ths secton, we present one of the man contrutons of the paper, whch s the extenson of OBGP+ and IDRP to deal wth the presence of wavelength converters. Havng wavelength converters relaxes the wavelength contnuty constrant, therey ncreasng the avalalty of wavelengths. We show that wth smple ut necessary modfcatons, ths nformaton can e ncorporated n the wavelength aggregaton process. Our approach does not ental too much overhead snce the only addtonal nformaton s the numer of wavelength converters at the remote order router. We dentfy two types of unoccuped wavelength channels at any gven order OXC: converter and non-converter channels. A converter channel conssts of dfferent types of wavelengths on ether sde of the OXC, thus requrng wavelength converson f t s to e used for lghtpath estalshment. A non-converter channel, on the other hand, s made up of the same wavelength on oth sdes of the OXC and does not requre wavelength converson. In ths secton, unless explctly specfed, wavelengths/channels refer to unoccuped wavelengths/channels. Snce wavelength converters are scarce, t s assumed that they are used only when asolutely necessary. Therefore, we frst compute the numer and type of non-converter channels the same way as n the case where there are no converters. Then, the remanng wavelengths on ether sde of the OXC are canddates of converter channels. However, snce a sngle wavelength converter can translate only one nput wavelength to another output wavelength, the numer of unused wavelength converters also affects the possle numer of converter channels. Usually, there are more canddate wavelengths than the possle numer of converter channels. Hence, there should e a mechansm to pck a specfc wavelength for each converter channel (e.g., frst-ft, random-ft, etc.) efore eng advertsed upstream. Ths approach provdes a hghly aggregated state nformaton, whle capturng the avalalty of wavelength channels. We now explan how the ENAW s computed usng Fg. 1, whch shows an example network wth two ASs, ther order OXCs and the unoccuped wavelengths at each OXC. For AS1, l andl represent ts order nodes, whereasr s the node that s drectly connected to AS1. The downstream AS (n ths case AS2) advertses a set of avalale wavelengths to the upstream Fg. 1: An example depctng order OXCs of two domans connected y a sngle nter-doman lnk. Ths example shows the numer of wavelengths and wavelength converters avalale at order OXCs of the two domans. AS (n ths case AS1). Let Wr adv,d (λ ) e the advertsed numer of wavelengths of type λ from the downstream AS. Also, let R adv = R r e the advertsed numer of avalale converters at r.,r (λ ) s the numer of wavelengths of type λ on the lnk etween l and r. Ths value s known to l snce the lnk s physcally attached to t. Thus, the numer of non-converter channels of type λ at l s: { },d(λ ) = mn,r (λ ),Wr adv,d(λ ) (3) In Fg. 1,,d(λ 1 ) = mn{3,2} = 2,,d(λ 2 ) = mn{2,4} = 2, and,d(λ 3 ) = mn{5,3} = 3. The remanng wavelengths can e part of converter channels at l. The maxmum numer of possle converter channels s determned not only y the numer of wavelengths that are not n the non-converter channels, ut also y the numer of avalale converters. Hence, t can e shown that the maxmum numer of converter channels s, {[ ( mn,r (λ ),d(λ )) ], (4) [ ( ) ] } Wr adv,d(λ ),d(λ ),R adv In Fg. 1, the numer of converter channels s: mn{{(3 2) + (2 2) + (5 3)}, {(2 2) + (4 2) + (3 3)}, 4} = mn{3,2,4}= 2. For these converter channels, wavelengths are selected from the set {,r (λ )}\{,d(λ )},.e., the set of wavelengths n,r (λ ) that are not n the non-converter channels. Then,,d(λ ) s updated so that t ncludes oth the converter and non-converter channels efore eng advertsed upstream. Let us assume that a random selecton s used and the updated,d(λ 1 ) = 3,,d(λ 2 ) = 2, and,d(λ 3 ) = 4. Smlarly, the numer of non-converter channels of type λ at l s: { } W l,d(λ ) = mn W l,l (λ ),W l,d(λ ) (5) In Fg. 1, W l,d(λ 1 ) = mn{6,3} = 3, W l,d(λ 2 ) = mn{4,2} = 2, and W l,d(λ 3 ) = mn{1,4} = 1. The total numer of converter channels at l s, {[ ( mn W l,l (λ ) W l,d(λ )) ], (6) [ (,d(λ ) W l,d(λ )) ] },R l where R l s the numer of converters at l. In Fg. 1, ths s equal to mn{{(6 3) + (4 2) + (1 1)}, {(3 3) + (2 2)+(4 1)}, 3} = mn{5,3,3}= 3. Let us assume that after randomly selectng from wavelengths that are not n the non-converter channels for the three converter channels, the updated Wl adv,d (λ 1) = 5, Wl adv,d (λ 2) = 3, and Wl adv,d (λ 3) = 1. Fnally, AS1 advertses Wl adv,d (λ ) and R adv = R l to upstream domans. However, wthout the modfed wavelength
4 aggregaton process (see eq. (2)), AS1 would have nstead advertsed Wl adv,d (λ 1) = 2, Wl adv,d (λ 2) = 2, and Wl adv,d (λ 3) = 1. In [9] [10], t s proposed to pggyack Keepalve messages that are exchanged etween neghorng OXCs wth PSI messages. In ths approach, keepalve messages are, just lke n BGP, exchanged to notfy f the neghorng node s stll operatve. However, unlke n BGP, the keepalve messages are extended to convey PSI messages. A major advantage of ths strategy s that t does not ncrease the numer of routng messages exchanged etween domans. In ths paper, we employ the same approach. V. RESULTS AND DISCUSSION In ths secton, we present smulaton results that compare the performance of OBGP, OBGP+ and IDRP. Our performance metrcs are the Blockng Rato (BR) of nterdoman lghtpath requests, and the numer of routng messages exchanged to acheve ths lockng rato. To ths end, we have conducted extensve smulatons usng OPNET. In our smulatons, we have used a PAN-European topology, whch was ntroduced n [2] as a reference topology sutale for a PAN-European fer-optc network. The network conssts of 28 domans and 41 nter-doman lnks, and the nodes were chosen n such a way that some of the man European Internet Exchange Ponts are ncluded. Insde each doman of the PAN European network, we placed a random numer of OXCs, whch s equal to or hgher than the numer of nter-doman lnks of that doman. There are 18 source and 10 destnaton OXCs randomly located coverng the entre PAN European network n such a way that each doman has one source or destnaton OXC. In other words, we smulate nter-doman traffc whch s transferred etween domans. Each lnk n the network conssts of 5 fers and each fer has 14 wavelengths. In our smulaton, traffc was modeled accordng to a Posson dstruton wth exponentally dstruted nter-arrvals. The lockng rato and routng messages are collected under dfferent traffc loads, varyng from 100 up to 300 Erlangs. In order to evaluate the mpact of the frequency of updates n the PSI messages, we have tested three scaled and normalzed Keepalve Update Interval (K T ) of the Keepalve messages: K T = 1, K T = 3, and K T = 5 unts. In terms of the avalalty of converters, we have consdered three scenaros: no converters, 5 converters and 10 converters at each order OXC of the domans n the network. For each case, the results are the averages of over 30 randomly generated PAN European network confguratons. These network confguratons are dfferent from each other n the network topology nsde each doman, and the locaton of source and destnaton OXCs over the entre network. Due to space constrants, we are ale to show only some of the results. Fgs. 2 and 3 show the effcency of usng wavelength converters n OBGP+ and IDRP for K T = 1. Smlar results have een otaned for K T = 3 and K T = 5. Tale I shows the mprovement factor (IF) n the lockng ratos of OBGP+ and IDRP over OBGP and the numer of messages generated under traffc values 200, 250 and 300 Erlangs for 5 converters. Smlar results have een otaned for 10 converters. The followng oservatons can e made from our results. Increasng the update nterval K T causes more lockng ecause a hgher value of K T means that the PSI s not accurate enough snce messages are exchanged less frequently. In fact, a major advantage of emeddng PSI messages n Keepalve messages s that when K T s decreased so as to mprove the responsveness of OXC neghors, PSI messages wll e updated more frequently. IDRP always sgnfcantly outperforms oth OBGP+ and OBGP (whereas OBGP+ outperforms OBGP). Ths s due to the fact that IDRP addtonally utlzes aggregated load nformaton. In fact, for 10 converters and K T = 1, IDRP acheves a lockng rato of less than 0.1% for all smulated traffc values. The 0.1% lockng rato s a threshold recommended y the IST FP6 NOBEL project [7] for optcal networks n order to support real-tme and streamng applcatons. The total numer of messages generated decreases as more wavelength converters are used n the network. The reason for ths s that n the presence of wavelength converters, the wavelength contnuty constrant s relaxed and there wll e more wavelengths avalale along a path. Therefore, t s less lkely for the wavelengths of a path to e exhausted fast, therey trggerng reachalty messages and path exploraton. The lockng ratos for IDRP and OBGP+ decrease as more wavelength converters are placed n the network. But ths s not the case n OBGP (results not shown here) f t always chooses the wavelength wth the lowest dentfer (Frst-Ft) along the shortest path. Such a frst-ft approach ncreases conflcts as dfferent OXCs tend to smultaneously choose lower dentfer wavelengths, whle hgher dentfer wavelengths are avalale. The stuaton s worsened as the numer of converters n the network s ncreased, snce the avalalty of these lowest ndexed wavelengths s also ncreased, therey exaceratng the posslty of conflcts. Ths stuaton can e avoded y choosng wavelengths randomly (Random-Ft) nstead of always choosng lower ndexed wavelengths. VI. CONCLUSIONS In ths paper, we have made smple ut mportant modfcatons to two nter-doman optcal protocols, namely, OBGP+ and IDRP, to handle the presence of wavelength converters. We have also performed extensve smulatons comparng the performance of OBGP (Optcal BGP) and these protocols. The results otaned n a PAN European network show that IDRP sgnfcantly outperforms OBGP+ and OBGP, and OBGP+ outperforms OBGP. The performance metrcs n the smulaton were lockng rato and the numer of messages generated (for a duraton of one week). From these results, t can e nferred that the exchange of aggregated path state nformaton (PSI), and the presence of
5 Keepalve Update Interval (K T = 1) Keepalve Update Interval (K T = 3) Keepalve Update Interval (K T = 5) 200 Erlangs 250 Erlangs 300 Erlangs 200 Erlangs 250 Erlangs 300 Erlangs 200 Erlangs 250 Erlangs 300 Erlangs IF (OBGP+) 3673.38 68.50 10.97 572.01 50.79 10.30 315.59 38.55 9.69 IF (IDRP) 13395.80 172.95 25.14 827.13 103.67 18.42 518.14 69.97 15.69 Traffc Routng Routng Routng Routng Routng Routng Routng Routng Routng (Erlangs) Messages Messages Messages Messages Messages Messages Messages Messages Messages OBGP OBGP+ IDRP OBGP OBGP+ IDRP OBGP OBGP+ IDRP 100 7,393,754 4,106,295 3,316,670 6,538,899 4,045,308 3,262,196 5,810,185 3,996,540 3,204,790 150 8,433,843 3,999,754 3,362,538 7,066,636 3,920,216 3,278,214 6,113,651 3,830,622 3,195,828 200 9,139,267 4,002,240 3,377,721 7,593,317 3,881,279 3,260,521 6,381,400 3,775,545 3,149,897 250 9,149,884 4,025,679 3,378,519 7,410,155 3,916,994 3,239,177 6,323,175 3,823,118 3,110,352 300 9,420,468 4,771,478 3,455,367 7,433,076 4,469,614 3,279,149 6,254,482 4,252,166 3,121,277 TABLE I: Improvement Factors (IF) n the lockng ratos of OBGP+ and IDRP over OBGP for 200, 250, and 300 Erlangs, and overall numer of routng messages exchanged for 5 converters. Blockng rato (%) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 No converters 5 converters 10 converters 0 100 150 200 250 300 Traffc (Erlangs) Fg. 2: Average lockng rato and standard devaton. Comparson of dfferent numer of wavelength converters for OBGP+ (K T = 1). Blockng rato (%) 0.6 0.5 0.4 0.3 0.2 0.1 No converters 5 converters 10 converters 0 100 150 200 250 300 Traffc (Erlangs) Fg. 3: Average lockng rato and standard devaton. Comparson of dfferent numer of wavelength converters for IDRP (K T = 1). wavelength converters at order OXCs mprove the lockng rato and the numer of messages generated sgnfcantly. In fact, usng IDRP wth enough wavelength converters, t s possle to acheve the 0.1% lockng rato threshold that s recommended y the IST FP6 NOBEL project [7] for optcal networks to support real-tme and streamng applcatons. The decrease n the lockng rato s otaned wthout an ncrease n the total numer of messages exchanged, ecause we have employed a strategy of pggyackng PSI updates n the Keepalve messages exchanged etween neghorng IDRAs/OBGP+ nodes. A venue for future research can e to nvestgate the performance of the nter-doman routng protocols n larger nter-doman networks, and the optmzaton of the placement of wavelength converters n nter-doman networks. ACKNOWLEDGEMENTS M. Yannuzz acknowledges the support receved from the Spansh Mnstry of Scence and Innovaton under contract TEC2009-07041, and the Catalan Government under contract 2009-SGR 1508. REFERENCES [1] M. Blanchet, F. Parent, and B. St-Arnaud, Optcal BGP (OBGP): IDRA lghtpath provsonng, IETF draft, etf-draft-parent-ogp-01, Mar. 2001. [2] S. De Maesschalc, D. Colle, I. Levens, M. Pckavet, P. Demeester, C. Mauz, M. Jaeger, R. Inkret, B. Mkac, and J. Derkacz, Pan-European optcal transport networks: an avalalty-ased comparson, Photonc Network Communcatons, vol. 5, no. 3, pp. 203-225, May 2003. [3] M. J. Francsco, L. Pezoulas, C. Huang, and I. Larnadars, End-to-end sgnalng and routng for optcal IP networks, Proc. of IEEE ICC, Apr. 2002. [4] K.-C. Lee and V.O.K. L, A wavelength-convertle optcal network, Journal of Lghtwave Technology, vol. 11, no. 5/6, May/June 1993. [5] G. Lu, C. J, and V. Chan, On the scalalty of network management nformaton for nter-doman lght-path assessment, IEEE/ACM ToN, vol. 13, no. 1, pp. 160-172, Fe. 2005. [6] Q. Lu, M.A. Kök a, N. Ghan, and A. Gumaste, Herarchcal routng n mult-doman optcal networks, Computer Communcatons, vol. 30, no. 1, pp. 122-131, Dec. 2006. [7] NOBEL: A Next Generaton Optcal Networks for Broadand European Leadershp, IST Integrated Project, http://www.st-noel.org. [8] L. Wang, H. Zhang, and L. Zheng, Reducng the OBGP protecton swtchng tme n WDM mesh networks, Proc. of OFC, Mar. 2006. [9] M. Yannuzz, X. Masp-Brun, S. Sanchez-Lopez, and E. Marn-Tordera, OBGP+: An mproved path-vector protocol for mult-doman optcal networks, Optcal Swtchng and Networkng, Elsever, vol. 6, ssue 2, pp. 111-119, Apr. 2009. [10] M. Yannuzz, X. Masp-Brun, G. Farego, S. Sanchez-Lopez, A. Sprntson, and A. Orda, Toward a new route control model for multdoman optcal networks, IEEE Communcatons Magazne, vol. 46, no. 6, pp. 104-111, June 2008.