2010 International Congress on Ultra Modern Telecommunications and Control Systems and Workshops (ICUMT)

MCP-RWA: A Novel Algorithm for QoT-guaranteed Online Provisioning in Photonic Networks

Davide Adami
CNIT Research Unit - Dept. of Information Engineering University of Pisa Pisa, Italy d.adami@iet.unipi.it

Stefano Giordano, Michele Pagano, and Luiz Gustavo Zuliani

Dept. of Information Engineering University of Pisa Pisa, Italy {s.giordano, m.pagano, gustavo.zuliani}@iet.unipi.it without the need of OEO conversions. Data about to be carried by a WRPN are converted to the optical domain at the ingress node, and remain in the form of an optical signal while traversing the network. The main advantages and drawbacks of WRPNs come from the same reason: the absence of OEO conversion. Data are carried at high throughput transparently by lightpaths, thus they can have any format or rate. On the other hand, optical signals are not regenerated in WRPNs. Therefore, optical transmission impairments [2] are accumulated while wavelengths traverse the network. These physical layer impairments (PLIs) degrade the optical signal quality and indirectly affect the Bit Error Rate (BER) of traffic being carried. While opaque networks usually have point-topoint or ring topologies, WRPNs tend to be meshed, with longer links and larger number of nodes. In this scenario, a given lightpath may not be eligible to haul traffic due to poor Quality of Transmission (QoT), even if there are abundant available resources. Lightpath provisioning consists in finding a path between end nodes and one or more suitable wavelengths that must be available on all the links that compose the calculated path. This problem is called Routing and Wavelength Assignment (RWA). In order to assure lightpath QoT, PLI must be taken into account during the RWA phase. There is a wide gamma of impairment-aware RWA (IA-RWA) algorithms in the literature, with different grades of complexity and performance [3]. Regarding lightpath blocking probability and network resource utilization, it is known that IA-RWA algorithms that achieve better results are those that solve the routing and the wavelength assignment problems in a combined process, considering PLI during this procedure and also verifying the QoT of the lightpath subjected to setup [3][4]. IA-RWAs of this type require a deep, highly detailed physical topology knowledge, usually are processing intensive, and very few of them are available. IA-RWA with combined routing, wavelength assignment and impairment validation is the stateof-the-art solution for lightpath provisioning in WRPN, and research in this area is a hot topic [1]. This work introduces a new online IA-RWA of this new category that guarantees the QoT necessary to satisfy the requested end-to-end bandwidth and BER. It performs on-thefly multipath RWA calculation, and lightpath is selected by a

AbstractWavelength Routed Photonic Networks (WRPNs) with a fully distributed control plane promise to satisfy the increasing infrastructure needs of upper layer clients regarding automatic, on-demand guaranteed services. In this highly dynamic scenario, lightpath provisioning based on classic Routing and Wavelength Assignment (RWA) approaches is inappropriate, mainly due to poor network resource utilization and the lack of Quality of Transmission (QoT) guarantees. In order to fulfill the challenging requirements for WRPNs, online RWAs that consider impairments of the physical layer are needed. These are usually complex algorithms that require intense processing. This paper introduces a new algorithm that assures absolute bandwidth and Bit Error Rate (BER) for lightpaths, and combines the RWA and the impairment validation processes to improve resource utilization. Simple yet efficient novel heuristics based on residual link capacities for a high parallelizable RWA engine are also presented. Keywords-photonic networks; physical impairments; RWA; residual link capacity; QoT

I.

INTRODUCTION

Traditionally, optical networks are composed of Wavelength Division Multiplexing (WDM) transmission lines and Optical Network Elements (ONEs) that perform 3R (Retiming, Re-amplifying, and Reshaping) regeneration by means of OEO (Optical-Electronic-Optical) conversions. These optical networks are known as opaque networks. All the complexities of the optical layer are dealt during the design and deployment stages, and they are not taken into account during provisioning. However, such networks have issues with cost and power efficiency, and cannot leverage modern applications due to slow provisioning time (days, even weeks). Moreover, OEO conversion is a bottleneck for routing at the optical layer. Indeed, ONEs do not have enough processing power to electronically route data hauled by modern WDM systems, that can carry dozens of channels in a single fiber at speeds up to 40 Gb/s per channel. Wavelength Routed Photonic (or All-Optical) Networks (WRPNs) are slowly been adopted to circumvent the limitations of opaque networks [1]. WRPNs are composed of Photonic Cross-Connects (PXCs) that are able to perform Optical Circuit Switching (OCS) entirely in the optical domain, 978-1-4244-7286-4/10/$26.00 2010 IEEE

141

multi-criteria rule set. As main novelties, it introduces the use of optical power and wavelength residual capacities in the combined RWA procedure, as well as the Critical Link Avoidance (CLA) feature. These novelties enhance traffic balance and network resources utilization. The rest of this paper is organized as follows. Section II briefly describes the most relevant optical transmission impairments and how they are treated by our algorithm. Next, Section III extensively details the proposed IA-RWA algorithm. Section IV discusses the utilization of the proposed algorithm in a fully distributed control plane. Section V presents some results concerning the fine tuning of some of the algorithm components and also a performance comparison with respect to other RWA solutions. Finally, conclusions are drawn in Section VI. II. PHYSICAL IMPAIRMENTS IN WRPNS

to the total optical power on links. This is known as the maximum power constraint. On the other hand, the minimum power constraint, which is best known as sensitivity level, assures that optical signals can be properly detected by all optical devices. Our algorithm uses the maximum power constraint to effectively handle ASE and fiber nonlinearities, the most important impairments in current photonic networks. Moreover, the minimum power constraint is used to guarantee QoT in a simple yet powerful way. The details are discussed in the next section. III. PROPOSED IA-RWA ALGORITHM

Transmission in optical fibers is affected by a number of physical impairments. Among those, the most relevant factors are: chromatic dispersion, polarization mode dispersion (PMD), amplified spontaneous emission (ASE) and the socalled nonlinear effects. Chromatic dispersion [5] is not an issue in cutting edge WDM systems anymore. It can be compensated on a per-link basis, and therefore does not need to be directly taken into account by an IA-RWA algorithm. PMD compensation [6] on the contrary is very difficult to perform, and is the principal challenge in the deployment of cutting edge 40 Gb/s systems. In todays production networks, PMD effects over QoT are usually avoided by setting up an upper bound for lightpath length, although it is far from optimal. However, the maximum allowed lightpath length constraint tends to be relaxed, due to quality enhancements in optical fibers (smaller PMD parameters) and late advances in all-optical PMD compensation [7]. ASE degrades the optical signal to noise ratio (OSNR), and is one of the dominant impairments at any bit rate. ASE noise accumulates as the optical signal traverses a path, and saturation effects may disturb the effectiveness of signal amplification, influencing the BER at the receiver. Demultiplexing filters can strengthen the ASE noise leading to linear crosstalking as well. ASE noise can be either measured or approximately calculated, using ONE information like optical input power and losses. In the former case, the WRPN must be equipped with Optical Performance Monitoring (OPM) devices [8]. Nonlinear impairments [9] strictly depend on optical power and are the most difficult ones to be treated. Examples are stimulated Raman and Brillouin scattering, four-wave mixing, self phase and cross phase modulation. They can substantially affect the performance of very dense DWDM (Dense WDM) systems due to nonlinear crosstalking. As stated in [5] and [8], to consider nonlinear impairments in IA-RWA algorithms requires an extensive knowledge of the physical topology, and currently analytical models for such IA-RWA have not been envisaged yet. As fiber nonlinearities are completely powerdependent, they can be indirectly managed by imposing a limit

The IA-RWA algorithm introduced in this work combines routing, wavelength assignment and impairment validation processes. Therefore, it is able to satisfy circuit setup requests with strict bandwidth and BER requirements, to assure lightpath QoT and, at the same time, to minimize the blocking probability of future requests by maintaining acceptable levels of optical power as well as adequate OSNR throughout the WRPN domain. Three parameters must be specified in lightpath setup requests: the source-destination pair, the requested bandwidth (usually 2.5, 10 or 40 Gb/s) and the BER, numerically expressed in function of the Q-factor [10]. In case of a successful operation, the proposed IA-RWA engine returns not only the path and the wavelength tuple that will compose the new lightpath, but also the transmission optical power that must be used in the tunable laser at the source node to assure QoT (respecting the requested bandwidth and BER). In addition, the estimated BER for the new lightpath (which is equal or less than the required BER) is made available.

Figure 1. MCP-RWA macro-level flowchart.

At a macro-level, the proposed algorithm operation can be divided in three parts (see Fig. 1): on-the-fly lightpath candidates computation (multipath IA-RWA with CLA, taking into account optical power and wavelength residual capacities); power-aware impairment validation (to discard candidates with inadequate QoT); and finally lightpath selection (to pickup the best candidate based on a multi-criteria rule set). All these steps that together compose the proposed algorithm, from now on

142

referred as MCP-RWA (Multipath CLA power aware-RWA), are detailed in the following subsections. A. Multipath RWA The first part of the algorithm performs on-the-fly multipath RWA computation. Multipath techniques are commonly used to solve the routing subproblem offline, while just the wavelength assignment subproblem is solved online (usually with heuristics). Within the MCP-RWA context, multipath RWA means to calculate the best possible route for each usable wavelength. Considering the issues related to the actual use of all-optical converters (partial converters still have high costs, and full conversion is yet an immature technology), the wavelength continuity constraint (WCC) is applied. In order to promote multipath RWA, the physical topology is described by a series of isolated wavelength planes, called wavelength graphs (WGs). Each WG describes the current topology view for a single wavelength. Given two adjacent nodes in a WG, for instance A and B, A is connected to B only if the wavelength for that particular WG is available in the fiber link from A to B. When a lightpath setup request arrives, the first step is to prune from the physical topology all WGs whose wavelengths at source node are unusable, saving processing time. Two conditions may render unusable a wavelength in a source node: the wavelength is being used on all fibers (just outgoing fibers for unidirectional lightpaths), or there is no available transponder for the requested bandwidth that can tune in that particular wavelength. After pruning, for each WG an instance of the Dijkstra's algorithm is executed. After this calculation, the minimum cost path of each WG is obtained. As the calculations of the paths in all WGs are completely independent, they are highly parallelizable, which leads to an optimization of the algorithm execution time. PLIs are considered during the RWA process to weight links in WGs. Unlike the majority of RWA algorithms which use simple hop count (distance metric) to weight links [3], MCP-RWA uses an empirically defined PLI-aware link cost formulation. Due to the maximum power constraint, a request can be blocked even when there are continuously available wavelengths along a path for the requested endpoints. In some particular configurations, a single pre-established lightpath can use most of, or even all, the allowed optical power in a given fiber. Therefore, the residual power capacity, i.e., the optical power that still can be injected in a fiber, is an important metric to evaluate a link cost, together with the number of available wavelengths (or residual wavelength capacity). To find the best generic expression that calculates the link cost of a fiber in function of its wavelength and optical power residual capacities, simulations were carried out by taking into account a number of empirical formulas. Since the link cost must get higher and higher as its residual capacities decrease, the best expression is one where the link cost grows exponentially, i.e.:
P res1 res i, j a i , j *res f Pires , round 10 * = ,j i, j

where

res Pi ,res j and i , j are the power and wavelength residual

capacities for the link (i, j). The performance of (1) are strictly related to the value of the exponential base, the a parameter. Therefore, a set of simulations were also performed to find which value of a would provide the top performance variant of the above expression (see Section V for details). Another strategy introduced by MCP-RWA to minimize the blocking probability is CLA. The altruist idea of avoiding using particular links to save them for future requests was introduced by the Asynchronous Criticality Avoidance (ACA) protocol [11]. Except for sharing this concept, CLA technique is completely different from ACA by any perspective. All links that are labeled as critical by the CLA are initially pruned from the physical topology. If, after the first attempt to find the minimum cost path in all WG that describe the topology, not even a single lightpath candidate is found, the process is repeated again, but this time considering all critical links that were not visible in the first pass. The key aspect of CLA is the rule that defines the criticality of a link. For that purpose, the wavelength and power residual capacities were initially considered as candidate metrics. Different combinations of thresholds for these two metrics were used in simulations in order to find the best configuration for most WRPNs. The best results were found taking into account only the power residual capacity to define a link as critical, when 20% or less of the original capacity remains useable. The details are discussed in Section V. B. Impairment validation When the multipath RWA phase of the algorithm ends, the impairment validation process of the lightpath candidates (described in the list of minimum cost paths from each WG) takes place. It is divided in two steps: optical power estimation and pre-setup evaluation. In the first step, for each single lightpath candidate the minimum transmission power necessary to guarantee the required Q-factor is calculated. Also, the fraction of the original transmission power for each link of the path is computed. The second step consists in verifying the feasibility of the lightpath, which means to check that all links can accommodate their share of optical power. The maximum power constraint must be respected. If a link does not comply with this constraint, the lightpath is discarded. The impairment validation process of MCP-RWA calculates the sensitivity level based on the ASE noise and the desired Q-factor using the analytical model defined in [12], and also discussed in [13]. C. Lightpath selection The last phase of the algorithm is the final lightpath selection, that consists in selecting the best lightpath candidate among those whose QoT is already assured in the previous phase, respecting the bandwidth and Q-Factor specified in the setup request. At this point, all candidates (now described by path wavelength transmission power tuples) satisfy the setup request. Hence, the best candidate is such that, after its

(1)

143

successful establishment, the WRPN is in a state where the blocking probability of future requests is minimized. To effectively choose the best candidate, a number of simple heuristics were considered. Through simulations, the relevance of these (isolated and combined) heuristics was analyzed. The best results were obtained using a multi-criteria rule set, evaluated in the following order: 1. 2. 3. 4. 5. lowest number of critical links; lightpath whose wavelength is the Most Used (MU) [14]; lowest transmission power; lowest number of hops; lightpath whose wavelength is the First Fit (FF) [14].

that the selection process will always end up with precisely one lightpath, and is seldom matched. It is worth to mention that rule number 1 is ignored if all lightpath candidates were calculated when CLA was active, i.e. without using critical links. The complete MCP-RWA algorithm is fully detailed in the flowchart presented in Fig. 2. IV. MCP-RWA AND CONTROL PLANE INTEGRATION

When more than one candidate matches a criterion, the matching ones are compared on the basis of the next criterion.
START Source Destination BER Bandwidth

To achieve fast lightpath provisioning, which is necessary to deploy advanced and profitable services in reconfigurable WRPNs, an intelligent distributed control plane is mandatory. Generalized Multi-Protocol Label Switching (GMPLS) [15], that can be seen as an instantiation of the protocol independent Automatically Switched Optical Networks (ASON) [16] framework, is being refined to allow fully control of WRPNs. Within the GMPLS context, RWA algorithms are implemented in Path Computation Elements (PCEs) [17]. The topology information used by these entities is stored in a Traffic Engineering Database (TED), and is collected by the Link Management Protocol (LMP) and the routing protocol (usually OSPF-TE). TEDs are also handled by the signaling protocol, commonly RSVP-TE. For a proper integration of the MCP-RWA algorithm within a GMPLS-controlled WRPN some extensions to the routing and signaling protocols must be performed (no matter if a centralized or distributed PCE approach is used). The routing protocol must be able to disseminate the properties and the operational status of tunable transponders and fiber links. Extensions to enable the flooding of this information by the OSPF-TE protocol in the form of new TLVs (type-lengthvalues) are extensive detailed in [18]. A fraction of these extensions are already being standardized [19]. Regarding the signaling protocol, extensions must be performed to allow the correct establishment of bidirectional lightpaths. After the successful completion of a setup request for bidirectional lightpath, the source node must notify the destination node of not only the wavelength to be used in the upstream direction, but also the input optical power to configure the tunable transponder. One possible way to achieve this is to extend the RSVP-TE Suggested Label Object to carry the input optical power information.

Wavelength Graphs Pruning

Lightpath Candidates Computation

Lightpath Candidates List

Impairments Validation

QoT-assured Lightpath Candidates List

Does any lightpath exist?

YES

Final Lightpath Selection

V.

MCP-RWA VALIDATION

NO Route Wavelength Power Estimation

Deactivate Critical Link Mode

YES

CLA Active?

NO

No lightpath found: Fail

Success

A simulated network scenario was built to design and refine the PLI-aware link cost function of the MCP-RWA algorithm. It was also used to evaluate the CLA optimization in function of diverse definitions of link criticality, based on different metrics combinations and thresholds. At last, the simulation environment was used to compare the performance of MCPRWA with two other IA-RWA algorithms, using the mean blocking probability and the processing time as evaluation metrics. Simulations involving WRPNs are usually performed using classic, real-world topologies like the NSFNET [19] and the Italian High Speed Network [21], with 14 and 21 nodes respectively. In order to avoid polarization of results due to

Figure 2. MCP-RWA detailed flowchart.

If a criterion is matched by only one candidate, that one is chosen to be established. The last criterion in the list guarantees

144

singularities of real-world and random topologies, it was chosen for the simulations an uniform 7 x 7 Manhattan topology. This simulated WRPN is therefore composed by 49 nodes, interconnected by pairs of unidirectional fibers. The chosen topology has more than the double of the number of nodes of classic topologies, as expected for future WRPNs. All links have a fixed length of 150 km, with inline amplifiers at each 50 km that have a constant gain of 11 dB. Optical transmitters have an operational power ranging from -20 to +15 dBm, and they are capable of tuning in 16 different wavelengths. For the simulations, the maximum optical power allowed per channel was set to 9 dBm, while the maximum total power per link was set to 12 dBm. Connection requests are generated with randomly chosen source-destination pairs, with a bandwidth of 10 Gb/s and a Q-factor equal to 7 (BER 10-12). In the next subsections, design decisions and the MCPRWA evaluation are drawn. For the sake of clarity, graphs only show the most relevant curves obtained from simulations. A. PLI-aware link cost function The effectiveness of eq. (1) as the PLI-aware link cost function for MCP-RWA was evaluated taking as reference function the simple hop count link cost. A series of simulations was performed using as cost functions both the reference one and eq. 1, considering different values for the a parameter. It was found that the proposed link cost function based on residual optical power and wavelength capacity performs better than simple hop count, when 1 < a < 2. The maximum efficiency was obtained when a = 1.1. Fig. 3 shows the blocking probability obtained for the most significant values of a.
0.25

wavelength residual capacity is equal or less than 20%; optical power residual capacity is equal or less than 20%; either power or wavelength capacities are equal or less than 20%.

The best performance is attained when CLA is active and only the power residual capacity, at a rate of 20% or less, is considered to define a resource as critical.
0,14

CLA Deactivated
0,12

Wavelength Residual Capacity 20% Power Residual Capacity 20% Wavelength or Power Residual Capacity 20%

0,10 Blocking Probability

0,08

0,06

0,04

0,02

0,00 110 120 130 140 150 160 170 180 190 200 Connection Requests 210

Figure 4. Criticality Threshold Comparison.

It is worth to be noted that an inappropriate configuration of CLA parameters can led to poor algorithm performance. C. Overall MCP-RWA evaluation To validate the overall performance of MCP-RWA, it was compared with two other IA-RWA algorithms in the above described scenario. For fairness of comparison, all the IARWA approaches were implemented using the same impairment validation process of MCP-RWA, as described in Section III. Furthermore, all IA-RWA algorithms (MCP-RWA included) require the same physical topology information knowledge to operate properly. In the first approach, the routing, wavelength assignment and impairment validation processes are completely decoupled from each other. The well known Yen's Fixed-Alternated Routing (FAR) algorithm [22] is used to offline calculate 3 shortest paths for all sourcedestination pairs. When a connection request arrives, the MU heuristic is employed to assign a wavelength to the candidate paths. The connection request can not be satisfied if there is no continuous wavelength available in any previously calculated path, or if the selected lightpath can not offer the required level of QoT. This strategy (FAR+MU) is the same as the one proposed in [13], except for the wavelength assignment heuristics (MU instead FF). The second approach (MU+WG) also employs the MU heuristic to perform wavelength assignment, and uses WGs to online calculate lightpaths as MCP-RWA does, but with a few limitations. The simple hop count link cost function is used to weight links, and CLA is not present. When a connection request arrives, the shortest path is calculated in the WG whose wavelength is the first one found by the MU heuristic. If it is not possible to find a path, the next

Hop Count
0.20 Blocking Probability

a=1.1 a=2 a=10

0.15

0.10

0.05

0.00 110 120 130 140 150 160 170 180 190 200 Connection Requests 210

Figure 3. Cost function comparison.

B. Criticality thresholds In order to find the most appropriate rule to define a link as critical during the CLA process, several simulations were carried out using power and wavelength residual capacities as metrics, and also different thresholds as lower bounds for these metrics. Fig. 4 shows how MCP-RWA performs when CLA is deactivated and when CLA is operating using the following conditions (criticality thresholds) to set a link as critical:

145

wavelength plane is used, always as defined by the MU heuristic. Setup fails if no path can be calculated in any WG, or if any calculated lightpath can not offer QoT based on the required BER and bandwidth. Simulation results are shown in Fig. 5, 6 and 7. The mean blocking probability (Fig. 5) and the width of the 95% confidence intervals (Fig. 6) are presented in separate figures for the sake of clarity. As expected, FAR+MU presents the worst performance regarding the blocking probability, caused by the high wavelength fragmentation (an undesirable consequence of path computation techniques that do not take into account wavelength availability). The other two approaches that rely on WGs by far outperform FAR+MU. Fig. 5 also shows that MCP-RWA presents better performance than WG+MU due to the multipath RWA, the PLI-aware link cost function, the CLA technique and the multicriteria rule set. Moreover, as reported in Fig. 6, the introduction of the PLI-aware link cost function in MCP-RWA narrows and makes less variable the confidence intervals of the blocking probability, which is a desired feature. Fig. 7 depicts the average processing time needed by the algorithms in order to comply with the set of requests. The performance of FAR+MU degrades proportionally to number of connection requests due to reduction of the available resources for lightpath setup, and eventually becomes the worst. The processing time used by MCP-RWA is slightly superior of the WG+MU one. Thus, the benefits of MCP-RWA over WG+MU have almost no impact in the processing time needed to satisfy setup requests.
0.25

FAR+MU
0.03

MU+WG MCP-RWA

Confidence Intervals Size

0.02

0.01

0.00

60

70

80

90

100

110

120

130

140

150

160

170

180

190

200

210

Connection Requests

Figure 6. Confidence Intervals for the Blocking Probability.

4.0

3.5 Processing Time (s)

3.0

2.5
MCP-RWA FAR+MU WG+MU

2.0 5 30 55 80 105 130 155 180 205 230 255 Connection Requests

FAR+MU MU+WG
0.20

MCP-RWA

Figure 7. Processing Time.

Blocking Probability

0.15

VI.

CONCLUSIONS

0.10

0.05

0.00

Connection Requests

Figure 5. Mean Blocking Probability.

WRPNs are expected to offer highly dynamic on-demand services with QoT guarantees, and classic RWA approaches can not fulfill the high requirements set by this scenario. Effective online IA-RWA approaches must be considered. This paper describes an online IA-RWA that assures absolute BER and bandwidth, combines the routing, wavelength assignment and impairment validation process, and introduces a number of features. Path-wavelength candidates are discovered using a high parallelizable procedure in order to minimize the computation time. Impairment validation and QoT assurance are performed with optical power considerations, which are efficient to deal with ASE and nonlinear effects, the most relevant impairments when transmitting at 40 Gb/s. The newly introduced PLI-aware link cost function and CLA are simple but efficient heuristic techniques that optimize the blocking probability and network resource utilization. Further research can yet be conducted to deeply investigate how less connected topologies of different sizes can influence the performance of the PLI-aware link cost function and CLA.

60

70

80

90

100

110

120

130

140

150

160

170

180

190

200

210

220

146

220

ACKNOWLEDGMENT The authors would like to thank Dr. Claudio Bastianelli and Dr. Marcello Gollinucci for their collaboration to this work. REFERENCES
R. Muoz, R. Martnez, and R. Casellas, Challenges for GMPLS Lightpath Provisioning in Transparent Optical Networks: Wavelength Constraints in Routing and Signaling, IEEE Commun. Mag., vol. 47, no. 8, pp. 2634, Aug. 2009. [2] J. Strand, A. Chiu, and R. Tkach, Issues for routing in the optical layer, IEEE Commun. Mag., vol. 39, no. 2, pp. 8187, Feb. 2001. [3] S. Azodolmolky et at., A survey on physical layer impairments aware routing and wavelength assignment algorithms in optical networks, Computer Networks, vol. 53, no 7, pp. 926944, May 2009. [4] Y. Lee, G. Bernstein, D. Li, and G. Martinelli, A Framework for the Control of Wavelength Switched Optical Networks (WSON) with Impairments, IETF draft-ietf-ccamp-wson-impairments-03.txt, Jul. 2010, work in progress. [5] J. Strand and A. Chiu, Impairments and other constraints on optical layer routing, IETF RFC 4054, May 2005. [6] A. Vannucci and A. Bononi, A change of perspective on single- and double-stage optical PMD compensation, J. Lightw. Technol., vol. 26, no. 14, pp. 20872097, Jul. 2008. [7] A. T. Erdogan, A. Demir, and T. M. Oktem, Automatic PMD compensation by unsupervised polarization diversity combining coherent receivers, J. Lightw. Technol., vol. 26, no. 13, pp. 18231834, Jul. 2008. [8] R. Martinez, C. Pinart, F. Cugini, N. Andriolli, L. Valcarenghi, P. Castoldi, L. Wosinska, J. Comellas, and G. Junyent, Challenges and requirements for introducing impairment-awareness into the management and control planes of ASON/GMPLS WDM networks, IEEE Commun. Mag., vol. 44, no. 12, pp. 7685, Dec. 2006. [9] G. P. Agrawal, Nonlinear Fiber Optics. New York: Academic, 1995. [10] R. Ramaswami and K. Sivarajan, Optical Networks: a Practical Perspective, 2nd ed. Morgan Kaufmann Publishers, 2002. [1]

[11] T. Feng and H. T. Mouftah, Implementation issues for asynchronous criticality avoidance protocol in multifiber WDM networks," in IEEE CCECE03, vol. 2, pp. 871874, May 2003. [12] G. S. Pavani and H. Waldman, Using genetic algorithms in constrained routing and wavelength assignment, in 8th IFIP Working Conference on Optical Network Design and Modelling ONDM04, vol. 1, pp. 565 584, Feb. 2004. [13] G. S. Pavani, L. G. Zuliani, H. Waldman, and M. Magalhes, Distributed approaches for impairment-aware routing and wavelength assignment algorithms in GMPLS networks, Computer Networks, vol. 52, no 10, pp. 19051915, Jul. 2008. [14] H. Zang, J. P. Jue, and B. Mukherjee, A review of routing and wavelength assignment approaches for wavelength-routed optical WDM networks, Optical Network Mag., vol. 1, no. 1, pp. 47-60, Jan. 2000. [15] E. Mannie et at., Generalized Multi-Protocol Label Switching (GMPLS) Architecture, IETF RFC 3945, Oct. 2004. [16] ITU-T Rec. G.8080/Y.1304, Architecture for the Automatically Switched Optical Network (ASON), Nov. 2001. [17] A. Farrel, J.-P. Vasseur, and J. Ash, A Path Computation Element (PCE)-Based Architecture, IETF RFC 4655, Aug. 2006. [18] L. G. Zuliani, M. Savasini, G. S. Pavani, R. Pasquini, F. L. Verdi, and M. Magalhes, An implementation of an OSPF-TE to support GMPLScontrolled all-optical WDM networks, in Proc. International Telecommunications Symposium ITS06, pp. 300305, Sep. 2006. [19] T. Otani and D. Li, Generalized Labels for Lambda-Switching Capable Label Switching Routers, IETF draft-ietf-ccamp-gmpls-g-694-lambdalabels-07.txt, Apr. 2010, work in progress. [20] D. Banerjee and B. Mukherjee, Wavelength-routed optical networks: linear formulation, resource budgeting tradeoffs, and a reconfiguration study, IEEE/ACMTrans. Networking, vol. 8, pp. 598607, Oct. 2000. [21] M. Ali and J. S. Deogun, Power-efficient design of multicast wavelength-routed networks, IEEE J. Sel. Areas Commun., vol. 10, pp. 18521862, Oct. 2000. [22] J. Y. Yen, Finding the K shortest loopless paths in a network, Manage. Sci., vol. 17, no. 11, pp. 712716, Jul. 1971.

147