Off-line Survivable Impairment-aware Routing and Wavelength Assignment

Transcription

1 Off-line Survivable Impairment-aare Routing and Wavelength Assignment Anteneh Beshir, Roeland Nuijts, Richa Malhotra, and Fernando Kuipers Introduction Physical impairments caused by noise and signal distortions affect the quality of an optical signal. The effect of physical impairments becomes more significant ith an increase in distance and bit rates. In order to minimize the bit error rate (BER), an optical signal may need to be regenerated after a certain distance. This is mainly achieved through re-amplification, re-shaping, and retiming, hich are collectively knon as R regeneration. In order to guarantee a certain BER, system vendors offer a certain level of optical signal-to-noise ratio (OSNR) at the output of a system. In a system here signal poer levels are lo enough that nonlinearities can be neglected, OSNR is an important parameter to measure the quality of an optical signal []. It is customary to place amplifiers at several points along a fiber link in order to overcome fiber losses. The segment of a link beteen to consecutive amplifiers isknonasafiber span. Hoever, optical noise is added by each amplifier along a fiber. This noise is referred to as Amplifier Spontaneous Emission (ASE). ASE degrades the OSNR and is reflectedinthatmeasure. In practice, vendors generally provide bounds on the length of a transparent (non-regenerated) path and number of spans in order to ensure an acceptable level of OSNR [].. Figure of Merit (FoM) The folloing formula can be used to compute OSNR. H OSNR = j= hvrn sp.j P in,j () here h is Planck s constant, v is the frequency of the input signal, R is the optical bandidth, P in,j is the input poer (W) at amplifier j, N sp.j is the noise figure of amplifier j, andh is the number of spans. If the net gain of a fiber link is unity, i.e., each amplifier is placed/tuned in such a ay that it exactly cancels out the loss of its preceding span, then the

2 noise figure of a link is the sum of the noise figures of its spans []. The noise figure of a system is usually given in (db). Hoever, in order to add the noise figures of each span to obtain that of a link, the values should be changed to linear units using the folloing formula [], hich e refer to as the Figure of Merit (FoM). FoM = H j= Lj, () here L j is the fiber loss of span j in db (it is the same as the noise figure of amplifier j hen the gain of the amplifier is one). The FoM value of a given fiber link is not directly proportional to its length, instead it depends mainly on the number of spans and the distance covered by each span. For example in Figure, let the fiber loss per km be. db/km. Using Eq., the FoM value of the scenario in Figure (a) is, hile it is for that of Figure (b). km km km km km km km (a) km km km km km km km (b) Figure : An example that shos ho the placement of amplifiers along a given fiber link affects its FoM value.. Regenerators Back-to-back optical transponders can be used not only to add/drop traffic but also to regenerate optical signals. Thus, regeneration can also be considered as an add/drop since the signal is added/dropped from the optical layer []. The impairment threshold of a lightpath is the maximum length (measured in FoM) that can be traversed by the lightpath ithout regeneration. After this length, the quality of the signal drops belo the acceptable level. The impairment threshold of a lightpath depends on the type of transponder (interface) used for regeneration. Figure shos some transponders and their specifications for a Gb/s data rate. A given avelength can be regenerated under to scenarios: () hen regeneration is required so that the impairment threshold is not exceeded, and

3 Gb/s Dispersion OSNR FoM Relative Interfaces tolerance (db,.nm) Threshold Cost (Transponders) (ps/nm) estimate DWDM FP NRZ NRZ ith EDC Figure : Different types of interfaces (transponders) and their specifications. () hen traffic carried by the avelength is added/dropped. We refer to the former as true regeneration, hile to the latter as add/drop regeneration.. Types of Nodes Up to no, e have only considered physical impairments associated ith links. Hoever, equipment at the nodes also add noise and contribute to signal distortions. The FoM value associated ith a node depends on the type of node, hich in turn depends on ho avelengths are added/dropped at the node. In early avelength division multiplexing (WDM) netorks, optical-to-electricalto-optical (O-E-O) conversion of avelengths as frequently required, regardless of hether these avelengths ere dropped or ere passing through a node. These O-E-O conversions had to be minimized in the netork since the O-E-O devices are quite costly and require significant poer. This led to the development of Optical Add/Drop Multiplexer (OADM) technology to locally add/drop individual or groups of avelengths, hile the rest are optically passed through ithout O-E-O conversions. Hoever, most early OADM systems are fixed in that the avelengths that are added/dropped at a given node are not reconfigurable, thus restricting netork reconfigurability in response to ne service demands and traffic patterns []. In order to provide flexibility, dynamically Reconfigurable Optical Add/Drop Multiplexers (ROADMs) are introduced. ROADMs simplify the planning process for DWDM-based netorks by alloing the addition, removal, or modification of one or more avelength channels ithin a netork automatically, ith minimal user intervention. Whereas in fixed OADM systems, the netork tuning process is performed manually and requires considerable equipment, trafficmanagement, and personnel []. The donside of ROADMs is that they are costly... Fixed OADM nodes Fixed OADM nodes usually consist of the folloing equipment.

4 Insertion Loss.dB.dB GMD GMD.dB scmd.db scmd GMD GMD = Transponder Total loss (incl. patch):.db FOM =. = scmd scmd Figure : A GMD-based fixed OADM node. Channel Mux/Demux (CMD): The CMD is a -port or -port mux/demux filter that feeds into one of the nine ports on the Group Mux/Demux (GMD). Both types of CMDs can coexist on the same line to offer anyhere beteen and avelengths on the DWDM system. Group Mux/Demux (GMD): The GMD provides a second stage mux/demux capability and supports nine CMD filters capable of offering a total of up to avelengths. The GMD provides a communications infrastructure in order for a node to interface ith other nodes and elements ithin the node. Figure shos a typical GMD-based fixed OADM node... Nodes ith Reconfigurable OADMs (ROADMs) In general, there are to types of ROADMs []: to-degree and multi-degree, here the degree refers to the numbers of DWDM fibers entering and exiting the ROADM node. This refers to traffic moving in one direction only. In practice, pairs of fibers are generally used ith each set carrying traffic in an alternate direction, so there ould be tice as many fibers entering and exiting the ROADM as its degree. The key enabling technology in ROADM configuration is the Wavelength Selective Sitch (WSS). This is an advanced fiber-optic module that can be used under softare control to dynamically select individual avelengths from multiple DWDM input fibers and sitch these to a common output fiber, or individual avelengths on a common input fiber can be selectively sitched to any of multiple output fibers. Figure shos a typical ROADM node ith a WSS.

5 Insertion Loss.dB.dB Demux block passive poer splitters Mux Block WSS scmd sc MD = Transponder Total loss (incl. patch):.db FOM =. = scmd scmd Mux Block WSS Demux block passive poer splitters Figure : A ROADM node ith a Wavelength Selective Sitch (WSS).. Survivability Since WDM netorks transport large amounts of data, failure of lightpaths can be costly. Hence, survivability, hich is the ability to restore communication after failure is indispensable in WDM netorks. In this paper, e assume a single-link failure model, since single link failures are the most common types of failure []. In this model, only a single link is assumed to fail at a time. For single-link failures, it is sufficient to have link-disjoint primary and backup lightpaths so that the backup lightpath takes over hen a link fails in the primary lightpath. Related Work Most of the related ork in the literature focuses on the placement of regenerators for unprotected lightpaths. For these studies, the objective can be of to types: minimizing the total number of regenerators [][] and minimizing the total number of nodes here regenerators are placed (i.e., regenerator nodes) [][][][][]. Since regenerators are active elements and require maintenance, it may be desirable to minimize the number of places here they are placed. Thus, minimizing the number of regenerator nodes is aimed at reducing the operational expenditure (OPE). On the other hand, minimizing the total number of regenerators reduces the capital expenditure (CAPE) since regenerators are costly. In addition, it ill also reduce the OPE since regenerators, hich generally use O-E-O conversions, have a high poer consumption. Depending on ho requests arrive, the traffic can be modeled as on-line (dynamic) [][][] or off-line (static) [][]. In an on-line traffic model, the

6 requests are unpredictable and arrive over time; hereas, in an off-line traffic model, the requests are knon beforehand and do not vary over a large time scale. A lightpath is a simple path beteen to nodes on a given avelength. Most previous studies assume that each lightpath requires a single avelength [][][][][]. Hoever, traffic to and from different nodes can be aggregated in a single avelength using technologies such as synchronous optical netork (SONET)/synchronous digital hierarchy (SDH) over WDM. In this paper, a lightpath refers to a connection (hich may require less than the capacity of a single avelength) beteen to pair of nodes, and a avelength can carry multiple lightpaths. Unlike most previous studies, e not only consider impairments associated ith links, but also nodal impairments. In addition, e take into account the types of nodes, i.e., the type of a node determines the FoM value associated ith it. We also note that a avelength is regenerated henever a traffic is added/dropped from it at the source and destination nodes of its lightpaths. In Section, e sho that the survivable impairment-aare routing problem, hich minimizes the total number of regenerators placed in the netork is NPhard. Subsequently, e provide an exact integer linear programming (ILP) formulation. Since the exact ILP does not scale ell, e provide a simple but heuristic approach. We study the performance of this approach in Section using the SURFnet netork. Finally, e conclude in Section. Survivable Impairment-aare Routing and Wavelength Assignment In this section, e give a formal definition of the survivable impairment-aare routing and avelength assignment problem and provide algorithms for solving it. We assume that there is no avelength conversion. Thus, each lightpath should use the same avelength in all of its links. We also assume that the same type of interface is used everyhere in the netork. The transponders are assumed to be bidirectional, thus one transponder suffices for both directions of signal flo beteen the source and destination nodes of a request. For each request, a pair of transponders are needed at its source and destination nodes, one for the primary and another for the backup lightpaths. If a lightpath is regenerated at an intermediate node, to transponders are needed, one on either side. Problem Given are an optical netork G(N, L,W), heren is the set of nodes, L is the set of links and W is the number of avelengths per link, and a set F of F requests. Associated ith each link (u, v) L is an FoM value r(u, v). Also associated ith each node u N is an FoM value r(u). The FoM threshold is. Thesurvivable impairment-aare routing and avelength assignment (SIRWA) problem is to minimize the total number of regenerators (transponders) needed in the netork so that () each request is assigned a pair of disjoint paths and corresponding avelengths; () the same avelength is used

7 on all the links of a lightpath; () the capacity of each avelength in a link is not exceeded; and () for any lightpath, the impairment values beteen regenerator nodes should not exceed the FoM threshold. Theorem The SIRWA problem is NP-hard. We provide a proof based on the impairment-aare path selection problem, here given a sparse regeneration netork (i.e., a netork herein only a fe nodes have regeneration capacity), an impairment threshold, and a request beteen to pairs of nodes, the problem is to find a feasible simple path for the given request. The impairment-aare path selection problem is proved to be NP-hard []. As explained in Section., e have to scenarios that lead to the regeneration of a given avelength: add/drop regeneration (i.e., hen traffic is added/dropped from the avelength), and true regeneration (i.e., hen the FoM value since last regeneration exceeds the threshold). Proof. Instance: A given avelength and a set of requests that can all fit in this avelength. In this instance, the number of add/drop regenerators is fixed, i.e., tice the total number of distinct source and destination nodes. Hence, the objective reduces to minimizing the number of true regenerations. A decision problem related to the given instance of the SIRWA problem is described as follos: Question: Is it possible to feasibly route all requests ith at most K true regenerations? For K =, the question reduces to: is it possible to feasibly route each request using only add/drop regenerations? In this scenario, the source and destination nodes of the given requests are the only regeneration nodes. In other ords, the netork has a sparse regeneration capacity. By solving the decision problem, each request ill be assigned feasible primary and backup lightpaths using only the existing regeneration netork. Hoever, this is equivalent to solving the NP-hard impairment-aare path selection problem. We first provide an exact integer linear programming (ILP) formulation, folloed by a simpler but efficient heuristic approach. We convert node eights to link eights by adding half of the FoM values of its end points to obtain the modified FoM of the given link, i.e. r(u, v) =r(u, v)+ ³ r(u)+r(v). The source and destination nodes of a request are assumed to have an FoM value of zero for the given request.

8 . Exact ILP Indices, constants, variables: f =,...,F ID of the lightpaths in the netork. =,...,W ID of the avelengths in the netork. N The set of nodes in the netork. N (u) The set of neighboring nodes of node u. L The set of links in the netork. B The capacity of avelength. B f The bandidth requirement of request f. a f,,u,v,t is if the primary lightpath of request f uses avelength at link (u, v) after being regenerated at node t. b f,,u,v,t is if the backup lightpath of request f uses avelength at link (u, v) after being regenerated at node t. γ f,,t,u is if the primary lightpath of request f uses avelength and regenerated at node t and at node u, inthatorder. τ f,,t,u is if the backup lightpath of request f uses avelength and regenerated at node t and at node u, inthatorder. x u, is if avelength is added/droped at node u. Objective: Minimize the total number of regenerators. Minimize : (γ f,,t,u + τ f,,t,u )+ x u, f t u u Constraints: Disjointedness constraint: The primary and the backup lightpaths of a request should be link-disjoint. (a f,,u,v,t +a f,,v,u,t +b f,,u,v,t +b f,,v,u,t ) f =,..., F ; (u, v) L. t Wavelength Constraint: The bandidth requirement of lightpaths using a given avelength of a link should not exceed the capacity of the avelength. B f (a f,,u,v,t +a f,,v,u,t +b f,,u,v,t +b f,,v,u,t ) B =,..., W ; (u, v) L. f t Flo Conservation constraints: The primary and backup lightpaths of a given request should originate at its source node. v N (s f ) a f,,sf,v,s f = and v N (s f ) b f,,sf,v,s f = f =,..., F. At intermediate nodes:

9 If a given node u is not the source or the destination node, then the flo related to the primary/backup lightpath that enters u has to leave it, here γ f,,t,u =for the primary and τ f,,t,u =for the backup path if it is regenerated, and γ f,,t,u =for the primary and τ f,,t,u =for the backup path if it is not. v N (u) v N (u) (a f,,u,v,t a f,,v,u,t )=γ f,,t,u and (b f,,u,v,t b f,,v,u,t )=τ f,,t,u f =,..., F ; =,..., W ; u N\{s f,d f }; t N\{u}. If a lightpath is regenerated at node u, the last regenerator node in the ne segment should be node u, and not any other node. v N (u) v N (u) a f,,u,v,u b f,,u,v,u t N\{u} t N\{u} γ f,,t,u τ f,,t,u = = and f =,..., F ; =,..., W ; u N\{s f,d f }. Simple path constraints: The lightpaths should not contain loops. At the source node, there should not be a flo associated ith any of its incoming links. (a f,,v,sf,t + b f,,v,sf,t) = f =,..., F. v N (s f ) t In addition, any flo that exits the source node, other than the one originating at the source node, should explicitly be set to. (a f,,sf v,t + b f,,sf v,,t) = f =,...,F. v N (s f ) t N\{s f } Similarly, for any intermediate node, there can at most be one flo of the primary or backup lightpath entering the node. a f,,v,u,t and b f,,v,u,t v N (u) t v N (u) t u N\{s f }; f =,..., F. Impairment constraints:

10 The FoM of any transparent segment (i.e., that of the links and the nodes) should not exceed the threshold, r(u, v)(a f,,u,v,t + a f,,v,u,t ) u v N (u) t N ; f =,..., F ; =,..., W. Add/drop Regenerations: If traffic is added/dropped at a given node, then there is regeneration. f {f s f =u} v N (u). Heuristic Approach (a f,,u,v,u + b f,,u,v,u ) F x u, u N ; =,..., W. The exact ILP does not scale ell even for small sized netorks as the SURFnet netork considered in Section. The complexity of the problem can be reduced by limiting the number of paths that are considered. Thus, e no propose a to-phase heuristic approach that makes use of a precomputed set of paths to solve the SIRWA problem... Phase : Precomputed Paths In the first phase, K pairs of (shortest) disjoint paths are pre-computed for each request (using an algorithm given in []), and the solution ill be selected from these pairs of paths using an ILP formulation. In this phase, the objective is to minimize the number of transponders required for adding/dropping the avelengths. It is based on the assumption that putting lightpaths that originate or end at a given node in the same avelength minimizes the total number of transponders needed. This approach has also an additional advantage in that the total number of avelengths used is minimized, since it tends to aggregate traffic in a smaller number of avelengths.

11 Indices, constants, variables: P f,k = {P f,k,,p f,k, } for k =,...,K α f,k, γ f,k, a f,k,l b f,k,l Objective: Minimize the total number of regenerators. Minimize A set of precomputed pairs of disjoint paths for request f. is if the k th disjoint path pair is selected and the primary lightpath uses avelength ; otherise. is if the k th disjoint path pair is selected and the backup lightpath uses avelength ; otherise. is if the primary of the k th disjoint path pair uses link l; otherise. is if the primary of the k th disjoint path pair uses link l; otherise. u x u, Constraints: For each request, only one pair of disjoint paths is selected. α f,k, =and α f,k, = for f =,...,F. k k The primary and backup paths should be from the same pair. α f,k, = γ f,k, for f =,...,F; k =,...,K. The capacity of each avelength on each link should not be exceeded. B f (a f,k,l α f,k, + b f,k,l γ f,k, ) B f k for l L; =,...,W. There is add/drop regeneration henever traffic is added/dropped. f {f s f =u or d f =u} (α f,k, + γ f,k, ) F x u, for u N ; =,..., W. k In the given ILP formulation, the primary and backup lightpaths of a request can be on different avelengths. Hoever, assigning the same avelength to the primary and backup lightpaths of a request not only simplifies the ILP formulation by providing symmetry, but it also reduces the number of transponders needed since the primary and backup lightpaths share starting and end points.

12 .. Phase : Rerouting Lightpaths In phase, the objective is to reduce the number of transponders needed to add/drop the given set of requests at their source and destination nodes. Hoever, some of the lightpaths obtained in phase may not be feasible, thus requiring the placement of extra regenerators. Algorithm Reroute tries to minimize the additional number of regenerators by rerouting lightpaths that are infeasible. Let P be the set of requests that need extra regeneration. A request needs extra regeneration if its primary or backup lightpath has an infeasible segment (i.e., its FoM value exceeds ) in the current setup. Let N be the set of regenerator nodes for avelength in the netork. The algorithm (see Algorithm ) orks as follos. In Step, it (randomly) chooses a request f among all the requests in P. In the next steps, it tries to find a feasible pair of disjoint paths using only the existing regenerators. This is done by constructing a ne graph on each avelength. In Step, graphg represents a graph in avelength, hich is made up of links that have enough capacity to support graph G, or belong to the disjoint paths of request f. In Step b, a ne graph G is obtained from graph G as follos. Its nodes are the regenerator nodes of avelength (including the source and destination nodes of the request), and there is a link beteen to nodes if they are directly reachable (i.e., ithout a regenerator). Then in Step c, to disjoint paths are computed using Suurballe s algorithm [] in graph G. These paths are then translated to their equivalent paths in G by replacing the links in G ith their corresponding subpaths in G. If the paths are simple and feasible, they are accepted as a solution. Otherise, e add regenerators to make the original paths feasible. Adding regenerators, hoever, may make some of the requests in P feasible. These paths are removed from P before continuing to the next iteration. Simulation Results We have performed simulations on the SURFnet netork hose FoM values are giveninfigure. Thetraffic matrix is shon in Figure. The given traffic represents synchronous digital hierarchy (SDH) data over the WDM netork. Each unit of traffic represents one VC, hich is equivalent to Mb/s. Each avelength has a capacity of Gb/s ( VCs). We assume that the type of transponders used in the netork is DWDM FP (see Figure ). Thus, the FoM threshold is. We compare our heuristic approach ith an on-line sequential approach. In the sequential approach, each request is assigned the shortest link-disjoint pair of paths beteen its source and destination nodes. Then, the lightpaths are sequentially allocated avelengths in such a ay that a lightpath is assigned to the loest-indexed avelength that has sufficient capacity for its traffic. In the sequential approach, avelengths and a total of FPs are required for the given traffic matrix. Figure shos the number of avelengths on each link, and

13 Algorithm Reroute. While P is not empty, pick a request f P. Let its assigned disjoint pair of paths be {P f,,p f, }.. For each avelength, letbl, be the residual capacity of avelength on link l. Let G =(N, L ),herel = {l L Bl, B f or l P f, or l P f, }. (a) For any u, v N {s f,d f }, let r (P u v ) be the length of the shortest path (in terms of FoM) beteen nodes u and v in G. (b) Create graph G =(N, L ), heren = {N,s f,d f } and L = {(u, v) u, v N and r (P u v ) }. Assign a cost of to each link in G. (c) Find to disjoint paths P and P in graph G. (d) For P and P, find their corresponding paths P and P in G. (e) If P and P are simple and disjoint paths: i. Assign them to request f. ii. Remove f from P and update the residual capacities of all links that belong to the old and ne paths of f. iii. Go to Step. (f) Else, go to Step for the next avelength.. If all avelengths are exhausted and no feasible paths are found, (a) Place the minimum number of regenerators needed to make P f,,p f, feasible. (b) Remove f from P. (c) Remove all requests in P hose paths are no feasible. (d) GotoStep.

14 =Fixed OADM =ROADM Figure : FoM values of the SURFnet netork. Figure shos the number of transponders needed at each node using our tophase heuristic approach. It can be seen that both the number of transponders and avelengths required by our heuristic approach are significantly less than those of the sequential approach. In addition, in our result, all the regenerations in the netork are handled using add/drop regenerations, i.e., no extra FPs are needed for true regenerations. The exact ILP formulation of Section. could not finish ithin a reasonable time on this netork. Conclusions In this paper, e have studied the off-linesurvivableimpairment-aarerouting and avelength assignment (SIRWA) problem, here given a netork and a set of requests, the problem is to assign link-disjoint primary and backup lightpaths for each request such that the total number of regenerators required in the netork is minimized. We have shon that this problem is NP-hard, and provided an exact integer linear programming (ILP) formulation for it. Hoever, since the exact ILP does not scale ell for even medium sized netorks, e have provided a simpler but efficient heuristic approach. We have performed simulations using a given traffic matrix on the SURFnet netork. The simulation results have shon that the number of regenerators and avelengths required by our heuristic approach are significantly less than those of an on-line

15 Figure : The traffic matrix in terms of VCs. Figure : The number of avelengths needed on each link.

16 Total Total Nodes Wavelengths Total Total Nodes Wavelengths Figure : The number of filters needed at each node for each avelength.

17 sequential approach. Minimizing the number of regenerators ill not only lead toasignificant reduction in the CAPE, but also results in a reduced OPE because of the significant decrease in poer consumption and heat dissipation. In addition, the reduced number of avelengths decreases the operating cost (OPE) associated ith each avelength. References [] M. Allen, C. Liou, S. Melle, and V. Vusirikala, Digital optical netorks using photonic integrated circuits (PICs) address the challenges of reconfigurable optical netorks, IEEE Communications Magazine, vol., no., pp. -, January. [] M. Batayneh, D.A. Schupke, M. Hoffmann, A. Kirstdter, and B. Mukherjee, Optical netork design for a multiline-rate carrier-grade Ethernet under transmission-range constraints, IEEE/OSA Journal of Lightave Technology, vol., no., pp., Jan.. [] S. Chen, I. Ljubic, and S. Raghavan, The regenerator location problem, Netorks, vol., no., pp. -, Dec.. [] M. Flammini, A.M. Spaccamela, G. Monaco, L. Moscardelli, and S. Zaks, On the complexity of the regenerator placement problem in optical netorks, Proc. of SPAA, pp. -,. [] A. Gumaste and T. Antony, DWDM Netork Designs and Engineering Solutions, Cisco Press, Dec,. [] J. Homa, K. Bala, ROADM architectures and their enabling WSS technology, IEEE Communications Magazine, vol., no., pp. -, July. [] F.A. Kuipers, A.A. Beshir, A. Orda, and P. Van Mieghem, Impairmentaare path selection and regenerator placement in translucent optical netorks, to appear in Proc. of ICNP, Kyoto, Japan, October -,. [] K. Manousakis, K. Christodoulopoulos, E. Kamitsas, I. Tomkos, and E.A. Varvarigos, Offline impairment-aare routing and avelength assignment algorithms in translucent WDM optical netorks, J. of Lightave Technology, vol., no., pp. -, June. [] Q.V. Phung, D. Habibi, H.N. Nguyen, and K. Lo, K pairs of disjoint paths algorithm for protection in WDM optical netorks, Proc. of Asia-Pacific Conference on Communications, pp. -, Oct.. [] G. Shen, W.D. Grover, T.H. Cheng, and S.K. Bose, Sparse placement of electronic sitching nodes for lo blocking in translucent optical netorks, OSA Journal of Optical Netorking, vol., no., pp. -, Dec..

18 [] N. Shinomiya, T. Hoshida, Y. Akiyama, H. Nakashima, and T. Terahara, Hybrid link/path-based design for translucent photonic netork dimensioning, J. of Lightave Technology, vol., no., pp. -, Oct.. [] J. M. Simmons, Optical netork design and planning, Springer, [] J. Strand, A.L. Chiu, and R. Tkach, Issues for routing in the optical layer, IEEE Communications Magazine, vol., no., pp., Feb.. [] J. Suurballe, Disjoint paths in a netork, Netorks, vol., no., pp. -,. [] K. i and H.J. Chao, IP fast rerouting for single-link/node failure recovery, Proc. of BROADNETS, pp.-, Sept.. []. Yang and B. Ramamurthy, Sparse regeneration in translucent avelength-routed optical netorks: architecture, netork design and avelength routing, Photonic Netork Communications, vol., no., pp. -, July. [] E. Yetginer and E. Karasan, Regenerator placement and traffic engineering ith restoration in GMPLS netorks, Photonic Netork Communications, vol., no., pp. -, Sep.. [] ROADMs in Netork Architectures, Ciena hite paper,