Scalability Analysis of WSS-based ROADMs

Transcription

1 Scalability Analysis of WSS-based ROADMs (nvited Paper) Gianluigi Notamicola, Giuseppe Rizzelli, Guido Maier, Achille Pattavina Politecnico di Milano, Piazza Leonardo Da Vinci 32, Milan, taly Tel: , Fax: , Abstract-Reconfigurable Optical AddlDrop Multiplexers (ROADMs) are becoming the industry standard for optical transparent switching in long-haul networks. ROADMs introduce new flexibility features towards full dynamic networks. n this paper, we investigate the scalability limits of the ROADM. We introduce a method to compute the number of ROADM internal subsystems (namely the wavelength selective switches) for a given traffic request. We evaluate the working and the blocking states of the ROADM with relation to its features of colorless, directionless and contentionless. We propose a method to compute the node requirements for a given set of network parameters. ndex Terms-ROADM, scalability, WSS, optical switching. NTRODUCTON f the continously increasing demand for optical bandwith services is the "fuel" for a constant evolution of transport networks, Reconfigurable Optical Add/Drop Multiplexers (ROADMs) have become the "engines" capable of crunching data flows at the required rate. ROADMs are the key elements for the development of backbone transport networks. They add flexibility by supporting dynamic optical channel switching and add/drop operations. Reconfigurability is achieved by using Wavelength Selective Switches (WSSs) instead of passive (de)multiplexers (MUX) that were used in the past generation of OADMs. n fact, WSSs are the core switching elements for today's ROADM. At current state of the art, a WSS is a lxn bidirectional device (i.e. having a single inlet and n outlets or a single outlet and n inlets, depending on the direction they are used). This means that a selected wavelength can be switched from the WSS inlet to one of the outlets, or viceversa, under software control. Typical WSS sizes are lx5, lx9 and lx24. mxn WSSs (for instance 8x8, 16x16) will be introduced in the next future to support large multi-degree nodes and massive add/drop operations. WSSs are implemented by using various technologies like Micro Electro-Mechanical System (MEMS) mirror arrays, Liquid Crystal on silicon (LCoS) phased array beam steering and Liquid Crystal (LC) based polarization/phase beam deflection [1]. WSSs can be remotely and dinamically configured in order to provide an efficient use of each tunable access resource (Transponders-TXPs) for on-demand switching and high bandwidth networking. WSS-based ROADMs have special features named colorless, directionless, contentionless and gridless. A ROADM is colorless if a wavelength can be set under software control and it is not fixedly associated to a physical add/drop port of the ROADM. We can assign a certain wavelength to an optical channel independently on its source in the add/drop section. This can be done by using a tunable transmitter (TX) connected to any of the WSS inlets. A ROADM is directionless if all the add (drop) operations can be accomplished to (from) any outgoing (incoming) fiber. This can be obtained by adding two extra WSSs to the add/drop section. A ROADM is contention less if multiple add/drop operations are possible for a given wavelength. This can be achieved by increasing the number of the add/drop modules. A ROADM is gridless if the channel bandwidth can be modified to optimize the frequency usage. n particular, it allows to dynamically operate with different channel types (10, 100 Gbps and higher) depending on the service requests. This is an important topic around which research colmnunity has been giving a lot of efforts [2], [3]. The property of the WSS-based ROADM of being remotely configurable, brings many advantages. Routing can be dynamically controlled to accomodate connection requests, thus enabling load balancing. The same internal device can be used for different operations. For instance, a TX is nonnally used to locally generate a new optical signal (add operation) as well as an RX is used to locally receive a channel (drop operation). However, the same pair of TX and RX can be connected backto-back forming a TXP in order to perform 3R regeneration or the wavelength conversion of a transit channel. The flexibility allows for a substantial reduction of the number of devices in the node, thus decreasing the cost. Network failure recovery is also improved by enabling remotely-controlled restoration, thus increasing network reliability. Physical impainnents that mostly affect the signal in ROADM networks are WSS filtering, concatenated filtering and crosstalk. They can have a significant impact on the signal Bit Error Rate (BER). n fact, the first two produce a cut on the high frequency portion of the signal spectrum thus leading to signal distortion [4], [5] while the cross-talk generates noise and distortion [6]. This work is focused on the colorless, directionless and contentionless ROADM architecture. Gridless enhancements will be investigated in future works. The rest of this paper is organized as follows: Section describes colorless, directionless and contentionless ROADM architectures, showing how the internal devices guarantee the three important features previously underlined. n Section we present the scalability model. Section V reports some illustrative numerical results. Conclusions are drawn in Section V /12/$ EEE

2 -- W --- Add/Drop A 5C L - -1 ;r-,,:;: ::: => r--- 1/0 E --=::rv all u :e fill _J -- W 5 =.1! W55 + 5C...,... W55 RX. l- f.1.l " 1 '.... l. <5 TX L- - J - ' n iillil 5C l wss S Add/Drop All 5-1/0 c) RX TX d) Fig. 1. WSS-based ROADM Architecture: (a) colored, directioned and contentioned (b) colorless directioned and contentioned (c) colorless directionless and -contentionless (contentionless degree ) (d) colorless directionless and 2-contentionless (contentionless degree 2). WSS-BASED ROADM ARCHTECTURE The ROADM is an evolution of the OADM used in the past mainly for WDM ring networks. n the OADM, wavelength channels could be added and dropped to/from transit fibers, but no inter-fiber switching was possible. This limitation was overcome in the ROADM by the adoption of the WSS and the broadcast-and-select architecture_ Fig. 1 represents the evolution of the ROADM architecture since the introduction of WSSs_ A generic ROADM has two types of ports: fiber ports, connecting the ROADM to other peer ROADMs via long range DWDM transmission systems; local ports, connecting the ROADM to local TXs and RXs. For both types we assume the number of input ports to be equal to the number of output ports (realistic assumption). We define a link port as the pair of input/output fiber ports connected to the same transmission system_ The number of link ports is the ROADM nodal degree_ Link ports are named after the cardinal points. The hardware associated to each link port can be partitioned into two sections: /O and add/drop. The /O section is composed by one star coupler (SC) that connects the ingress fiber to all the other fiber ports and to the drop section and by one WSS that collects signals from every direction and from the add section and it multiplexes them into the outgoing fiber port. Within the same link port, one SC outlet is directly connected to one WSS inlet (oopback connection) to implement link-based protection switching. Figure a shows the basic configuration called colored, directioned and contentioned [7], [8]. The add/drop section is composed by a WDM MUX and a DEMUX connected to fixed-wavelength TX and RX respectively_ Since the MUX has a fixed transfer function (for instance, an Array Waveguide Grating (AWG)), each TX is connected to a single inlet of the MUX: this makes the ROADM colored. n Fig. b each DEMUX is replaced by a WSS and each MUX by a SC in every add/drop section. By that, the ROADM becomes colorless since tunable TXs are used and RXs can be tuned to any wavelength independently of the SC inlet which it is connected to.

3 Figure c shows how the directionless feature is achieved. The add/drop section becomes common and shared by all the fiber ports. The add side is composed of a series of WSSs connected to a cascade of two SCs, the first one collecting all the flows from the TXs and the second one used as a splitter to broadcast them to all the output WSS. The drop side is composed by a WSS that collects all the channels from the input SCs and forwards them to a SC, to the outlet of wich a set of WSSs is connected, leading to the RXs. Any TX can be connected to any fiber port, reducing the structure complexity and routing constraints. The total number of TXs is less than that in the previous cases and any wavelength can reach all directions. This structure is also l-contentionless (i.e. contentionless degree 1), that means that each wavelength channel can be added by the ROADM and connected to at most one outgoing fiber. This is due to the fact that the add/drop section can support at most one transit signal at a given wavelength. Figure d shows a ROADM architecture which enables colorless, directionless, and 2-contentionless add/drop operations (i.e. contention less degree 2). n fact, to overcome the contention limitation, the contention less degree can be increased by increasing the number of add/drop sections that composes the ROADM. The input (output) fiber ports are represented on the left (right) side. n this case any TX (RX) can transmit (receive) to (from) any direction (directionless) at the same wavelength (colorless-contentionless) as long as the number of channels with the same wavelength is lower than the number of add/drop sections in the node. Network planning is simplified, since any add/drop port can support all colors and it can be connected to any outgoing fiber port with the possibility that the node can add the same wavelength more than once. We assume that, in all the three scenarios, all WSSs and all SCs in the ROADM are homogeneous in size (i.e. all WSSs are lxn; all SCs are lxq; with n and q constant). The rational of this assumption is that having all building-blocks of each type with the same form factor implies flexibility in the system and, ultimately, cost reduction.. SCALABLTY MODEL FOR A ROADM-BASED NETWORK Let us present the scalability model we have studied for ROADM-based networks. This simple theoretical tool is topology independent and it can be used to benchmark the results of a detailed network design. The purpose is to describe how to configure all the ROADMs of the network to support a certain amount of connection requests. n this first version of the model we do not consider physical impairments (left for a future release) and we assume unlimited availability of resources on the network links. Therefore, blocking of a path may occur only as consequence of unavailability of resources inside the ROADMs for either pass-through or add/drop operations. We consider three scenarios differentiated on the basis of the ROADM architecture (for simplicity in each scenario all the Fig. 2. WSS SC WSS /O Usage direction Split Split -all WSS splitting section for /O-Split and Split-all scenario ROADMs of the network have the same architecture type): No Split, llo-split and Split-all. The No-Split scenario assumes all ROADM colorless, direction less and contentionless (as in Fig. d). n the O-Split scenario the multiplexing capability of the ROADM fiber ports can be increased. When the number of interconnections to be set up between input and output fibers or add section and output fiber exceeds the /O WSS size, we can replace the WSS with the splitting section represented in Fig. 2 (operated in the upward direction) (the 110 SC size will be correspondingly incremented). n the Split-all scenario a similar expansion operation can be performed also in the add/drop section by replacing one of the WSSs with the add/drop splitting section of Fig. 2 (operated in the downward direction). A. The node perspective Our model requires a number of input parameters to characterize each ROADM. Let us define these parameters, reminding that the model is based on the assumption that all the input/output ports are symmetric and connections are bidirectional: nodal degree G; contentionless degree C; total number of wavelength channels in use on the input (output) fiber ports f; total number of connections to be added (dropped) or regenerated A; total number of TX (RX) in the node T; maximum number of wavelengths supported by the DWDM transmission system W; number of SC inlets (outlets) Ps p lit for all SCs; number of WSS inlets (outlets) Pwss for all WSSs. As we have explained before, in the No-Split scenario we do not consider the possibility of expanding the ROADM base architecture to increase the total number of available WSS outlets. Therefore, the maximum number wavelengths that can be added/dropped is limited by the number of WSS outlets in the add/drop section. With reference to Fig. led), this number

4 can be written as: T = Pwss. Psplit. C (1) The maximum number of used wavelengths is obviously bounded: rw G. (2) The number of connections that are either added/dropped or regenerated is bounded by the availability of TXs and RXs. Then: AT. (3) A final constraint regards the /O section and its architecture: the required number of WSS ports depends on the nodal and contentionless degrees: added can be any, so that all the connection setup/regeneration requests can be satisfied. Therefore, the constraint of Eq. 4 is no longer applicable, while all the other constraints (Eqs. 2 and 6) continue to hold. Considering the splitting, the number of required WSSs becomes: Q= G+C l + ll C+2.C.l (8) Pwss Pwss Pwss n the Split-all architecture, also the add/drop section is expandend by adding extra lxpsplit SCs to satisfy as many as requests arrive to the ROADM. Thus, constraint of Eq. 2 no longer holds. n both VO-Split and Split-all scenarios the contentionless degree is unconstrained so that ex can increase up to ex = 1. G+C Pwss. (4) B. The network perspective We shall highlight that Eq. 4 is very important, as it shows the direct constraint on ROADM scalability coming from the WSS size. Note also that this constraint is not alleviated by adding expansion modules to the add/drop section. The WSSs required to build a ROADM, which has the parameters specified above and which supports A requests for connection add/drop and regeneration, are counted by the parameter Q as: Q = G + C + 2. C l. Pwss t should be noted that Q is independent of r. Therefore, without losing generality, we can assume from this point on that all the wavelengths in the fiber links are active, i.e. r = W G. For instance, if we consider W = 120 and G = 4, then r = 480. By substituting Eq. 1, 3 and 4 into Eq. 5, we obtain an upper bound on the ROADM complexity: Q Pwss + 2C2. Pspl it. The previous equation regards the total number of connection as a whole. Let us now focus on the behaviour per link port. Be Ai the number of connections added (dropped) or renerated and directed to (coming from) fiber port i (A = Li=l Ai). Obviously, 0 Ai W. Due to the limited availability of TXs and RXs and the limited wavelength reuse, not all the potential requests for add/drop or regeneration may be satisfied. For simplicity, we assume a uniform distribution of requests between the link ports. Therefore: Ai = A = A/G Vi. Let us introduce the parameter ex =,\fw (0 ex 1). Each wavelength can be reused up to C times, and for uniformity, each link has access to l/g of the overall TX/RX resources. Thus, C. Due to Eq. 4: A G. mm(w;pwss. Psplit) (5) (6) ex ( p :; s -1). min (1, PWSSPlit). (7) n the VO-Split scenario, the VO section of the ROADM can be expanded by introducing further splitters and WSSs. n our model we assume the number of extra 1XPsplit SCs that can be n this section, we extend our scalability model from the single ROADM to a ROADM-based network. The goal is estimating how many WSSs need to be installed in each ROADM to support all the connections requested. We have chosen to develop a type of model called Network Global Expectation Model (NGEM) [9]. The NGEM approach provides an estimation of the main network parameters in terms of their expected values; the advantage is being totally networktopology independent, allowing for a fast and simple comparison of alternative node architectures. According to the NGEM method, a network can be characterized by a number of coverage parameters. f the network topology and the set of static connections established through the network are given, these parameters can be computed. Otherwise, they can be estimated for a generic network of N nodes and L links, under suitable simplifying assumptions. Let us consider this second situation. We assume uniform traffic (i.e., one connection per node pair) and all connections are routed on the shortest path (according to the minimum-hop metric). We need the following expected values: average nodal degree < <5 >= 2L/N; average number of connection requests per node pair < d >= N -1; average connection length (in hops) < h >= J(N -2)/«<5 > -1 ); average number of channels used for each link < Ch >= «d>. < h» / < <5 >; average ratio between the number of channels incoming to (outgoing from) a node and the number of connections added (dropped) in the node < ex >= 2/(1+ < h». Starting from this set of parameters computed by the NGEM, we can compute all the variables that are needed in our model presented in Section -A, thus relating the ROADM to an extensive network environment. n this way the ROADM can play the role of an average representative of the nodes of the considered network. n particular, we can derive the ROADM nodal degree, the number of active WDM channels on the adjacent links and the number of

5 TABLE MAXMUM'\' AND WSS COUNT FOR DFFERENT G, C AND PWSS N THE A) NO-SPLT, B) /O-SPLT AND C) SPLT-ALL SCENAROS (a) (b) (c) G=4 WSS 1x5 WSS 1x9 WSS 1x20 WSS 1x5 G=4 G=4 WSS 1x9 WSS 1x5 WSS 1x9 C=l 10! 21 18! 21 30! 17 A Q A Q A Q -----=C--= =2=-----1L[L? j ----'C...; =3= L_i L P}- C=4 /: / 72 : : 200 G=5 WSS 1x5 WSS 1x9 WSS lx20 A Q A Q A Q G=6 WSS 1x5 WSS 1x9 WSS 1x20 A Q A Q A Q WSS 1x5 WSS 1x5 G=6 WSS 1x9 WSS 1x5 WSS 1x9 6, 6 Q -50, !.? , ,09-40, !! ,63 31,84 WSS 1x9 WSS 1x5 WSS 1x9 added/dropped connections as: G=<6">l r = G 1< Ch > l 'Y =< 0: >. V. LLUSTRATVE NUMERCAL RESULTS A. The node analysis (9) (10) (11) Table a represents the total number of WSS (Q) in a ROADM, corresponding to the maximum achievable values of,\ for different G, C, Pwss in the No-Split scenario. We have considered Psplit = 8 in all cases (i.e. all SCs are lx8) and pwss = {5, 9, 20} (the most common WSS sizes available nowadays). Furthermore, we set W = 120 so to take into account future DWDM systems exploiting Raman amplification (most current DWDM systems have up to 80 wavelength-channels). Cells reporting a "/" indicate combinations of C, G and pwss that are not feasible, as they violate Eqs. 2, 4 or 6. {G; C} combinations wich are feasible in the No-Split architecture only with Pwss = 20, can be achieved in the /O-Split scenario with smaller WSSs (Pwss = {5; 9}). This is shown in Table lb. This comes however at the cost of reducing the max number of added/dropped/regenerated connections and of incrementing the number of WSSs.,\ reductions and Q increments are reported by A and Q as percent variations with respect to the values of,\ and Q of the NoSplit, pwss = 20 case. n a similar fashion, the Split-all architecture can again be exploited to achieve all the considered {G; C} combinations as with Pwss < 20. This time, as shown in Table e, there is no reduction of '\, but the number of WSSs per ROADM remarkably increases. 35, , o i 20 E " c rn1 10, -e-a=o.1 -+-a=o.3 «=0.5 -e-«=0.7...,..«= Nodal Degree Fig. 3. No-Split scenario: WSS lx20, C = 1, Psplit = 8, W = 120 Figure 3 refers to the NoSplit scenario again and it shows the scalability of the ROADM architecture based on lx20 WSSs. Q is reported as a function of G (growing beyond the range considered in Table ), for different values of 0:. Q = 0 (resulting for G = 20) indicates unfeasibility. Note that the ripples of the curves are due to the ceiling operation in Eq. 5. B. Network case-study We have applied the NGEM-based model to four casestudy networks: a planar mesh networks, a spidergon-like topology, and the European and US backbone networks. We have assumed an uniform traffic demand model, with one request per node-pair, contentionless degree C = 2 and Psplit. We have considered all the three No-Split, /O-Split and Splitall scenarios, and Pwss = {5; 9; 20}. Results are reported in Fig. 4 in terms of mean number of WSS QT that must be installed in a node to satisfy all the requests. As in Fig.

6 ,.140 CJ a; 120.Q 100 <: V V 80 '" { NoSplitl/OSplit Split NoSplitl/OSplit Split NoSplitl/OSplit Split NoSplitl/OSplit Split \ 1\ / PLANA MESH SPDJRGON EU Bacbone US BaJkbone.lx5 W5S.lx9WSS.lx20WSS Fig. 4. Application of the ROADM-network model (based on NGEM [9]) to four case-study networks 3, QT = 0 conventionally indicates that there is no feasible solution (violating Eqs. 1, 4 or 6). The simple planar mesh network has N = 6 and L = 9. The average parameters estimated by the NGEM approach according to the procedure reported in Section -B, are: G = 3; r = 9; = The SPDERGON network is an ideal topology composed of a ring and a set of diameter links connecting all pairs of opposite nodes. This networks has N = 20 and L = 40. The NGEM parameters are: G = 4; r = 48; = The third network is the well known Pan-European backbone, with N = 28, L = 171 and NGEM parameters: G = 3; r = 108; = Finally, we have included in the comparison the US backbone network which has N = 100, L = 171 and NGEM parameters: G = 4; r = 736; = Results show that in case of high values of (SPDERGON) or r (US Backbone) the No-Split architecture is feasible only with large WSSs (Pwss 20 for the US Backbone). With = smaller WSSs, the /O-Split provides a viable alternative for the SPDERGON, while for the US Backbone we have to rely on the Split-All architecture. Thus alternatives, however, imply a substantial increase of the number of WSSs (huge increment for US Backbone). V. CONCLUSONS n this paper we have carried out a scalability analysis of WSS-based ROAD Ms. Our novel analytical model can be applied both to a single node analysis and to a network casestudy. Simulation results show that ROADMs composed of small size WSS present scalability limits due to the need of high port count in 110 and add/drop sections. n order to overcome such limitations, we have proposed two strategies based on a cascade of splitting stages (i.e., /O Split and Splitall) which allow us to use even lx5 and lx9 WSSs to achieve large nodal degrees (G) and high add/drop ratio values (ex). lthe topologies of the networks are not reported in this paper due to space limitations. n fact, the same performances attainable by WSS with high port availability (i.e., lx20 WSS) can be achieved at a cost of higher number of devices which can be prohibitive in certain cases. By using the NGEM, the proposed analytical model can be also thought as an helfpful tool for a network designer as it gives an instantaneous hint on the number and size of WSSs needed to satisfy a certain set of traffic demands given few topology-related input parameters. ACKNOWLEDGMENTS The research leading to these results has received funding from the European Community's Seventh Framework Programme FP7/ under grant agreement n (STRONGEST project). REFERENCES [l] T. Strasser and J. Taylor, "Roadms unlock the edge of the network," EEE Communications Magazine, vol. 46, no. 7, pp , July [2] c. M. c. Politi, Y. Anagnostopoulos and A. Stavdas, "Routing in dynamic future tlexi-grid optical networks," in Proc. of Optical Network Design and Modelling (ONDM), April [3] B. Kozicki, H. Takara, and M. Jinno, "Enabling technologies for adaptive resource auocation in elastic optical path network (slice)," in Proc. of Asia Communications and Photonics Conference and Exhibition (ACP), Dec [4] F. Heismann, "System requirements for wss filter shape in cascaded roadm networks," in Proc. of Optical Fiber Communication Conference (OFCNFOEC), March [5] B. Pal, L. Zong, E. Burmeister, and H. Sardesai, "Statistical method for roadm cascade penalty," in Proc. of Optical Fiber Communication Conference (OFCNFOEC), March [6] S. Tibuleac and M. Filer, "Transmission impairments in dwdm networks with reconfigurable optical add-drop multiplexers," Lightwave Teclmology, Journal of, vol. 28, no. 4, pp , feb. 15, [7] S. Gringeri, B. Basch, Y. Shukla, R. Egorov, and T. Xi a, "Flexible architectures for optical transport nodes and networks," EEE Communications Magazine, vol. 48, no. 7, pp , July [8] G. Prasanna, B. Kishore, G. Omprasad, K. Raju, R. Gowrishankar, K. Venkataramaniah, R. Johnson, and P. Voruganti, "Versatility of a colorless and directionless wss based roadm architecture," in Proc. of Communication Systems and Networks (COMSNETS), Jan [9] S. Korotky, "Network global expectation model: a statistical formalism for quickly quantifying network needs and costs," Journal of Lightwave Technology, vol. 22, no. 3, pp , March 2004.