Translucent Optical Networks
Nicola Sambo, Student Member, IEEE, Nicola Andriolli, Alessio Giorgetti, Filippo Cugini,
Luca Valcarenghi, Member, IEEE, Isabella Cerutti, Member, IEEE, Piero Castoldi, Member, IEEE
The evolution of optical technologies has paved the way to the migration from opaque optical networks (i.e., networks in which
optical signal is electronically regenerated at each node) to transparent (i.e., all-optical) networks. Translucent optical networks (i.e.,
optical networks with sparse opto-electronic regeneration) allow to exploit the benefits of both opaque and transparent networks,
while providing a suitable solution for dynamic connections.
Translucent optical networks with dynamic connections can be controlled by the GMPLS protocol suite. This paper discusses
the enhancements that the GMPLS suite should be provided with, for the control of dynamic translucent optical networks with
quality of transmission (QoT) guarantees. Such enhancements concern QoT-awareness and regenerator-awareness and can be
achieved by collecting and disseminating the information on QoT and regenerator availability, respectively, and by efficiently
leveraging such information for traffic engineering purposes.
More specifically, the paper proposes two distributed approaches, based on routing protocol and signaling protocol, for
disseminating regenerator information in the GMPLS control plane. Moreover, three strategies are introduced to efficiently and
dynamically designate the regeneration node(s) along the connection route. Routing and signaling approaches are compared in
terms of blocking probability, setup time, and control plane load, during provisioning and restoration.
Translucent optical networks, GMPLS, regenerators, quality of transmission, signaling protocol, routing protocol.
Translucent optical networks exploit the advantages of both transparent optical networks (where connections are switched in
the optical domain) and opaque networks (where connections are optically terminated in each intermediate node and switched
in the electrical domain) . On the one hand, optical transparency offers large bandwidth at low cost. On the other hand, by
performing opto-electronic signal regeneration at some intermediate nodes, it is possible to recover the signal degradation due
to physical impairments. Both, linear physical impairments (e.g., amplified spontaneous emission noise, chromatic dispersion,
and polarization mode dispersion) and non-linear physical impairments due to intra and inter-channel effects (e.g., self phase
modulation, four-wave mixing, cross-phase modulation, and cross-talk) contribute to the degradation of the optical signal
quality. Such effects are especially critical for high data rates and limited wavelength spacing . Opto-electronic regenerators
are used to re-amplify, re-shape, and re-time the optical signal (i.e., 3R regeneration) with the aim of guaranteeing the quality
of transmission (QoT) required by the end-to-end connections.
In translucent optical networks, requested connections can be supported either by a transparent lightpath, i.e., a single all-
optical segment, or by a translucent lightpath, i.e., a sequence of all-optical segments connected by nodes that opto-electronically
regenerate the signal. Thus, a careful regenerator placement and an intelligent regenerator utilization are fundamental to design
and manage cost-effective translucent optical networks with QoT guarantees. Several studies focused on centralized schemes
for regenerator placement and routing and wavelength assignment in translucent optical networks , , , when connection
requests are known in advance, i.e., static traffic scenario.
Translucent optical networks with dynamic connection requests present additional cross-layer challenges. In  a framework
is proposed to address these challenges assuming that updated information is available at each node. A first challenge is the
regenerator placement which should be tailored to the dynamic scenario. Indeed, specific algorithms are required to account
not only for the present and estimated future network traffic, but also to account for the dynamic provisioning and rerouting of
network resources . Other challenges are the QoT evaluation and the dissemination of QoT-related information. The work
in  proposes routing solutions when QoT information is inaccurate or outdated, for example due to coarse measurements
of QoT parameters and reduced availability of monitoring equipments. Moreover, another main challenge is the study of
strategies for regenerator discovery and selection. Such strategies need to be designed while keeping in mind that network
state information may be only locally available and may change frequently due to dynamic nature of the connection requests.
All these issues influence the path computation and wavelength assignment. Preliminary studies , ,  address some of
these challenges in control planes based on GMPLS.
N. Sambo, N. Andriolli, A. Giorgetti∗, L. Valcarenghi, I. Cerutti, and P. Castoldi are with Scuola Superiore Sant’Anna, Pisa, Italy. F. Cugini is with CNIT,
Pisa, Italy.∗Corresponding author email: firstname.lastname@example.org.
CROSS-LAYER CHALLENGES IN THE CONTROL PLANE OF DYNAMIC DISTRIBUTED TRANSLUCENT OPTICAL NETWORKS.
This paper surveys several solutions for addressing such challenges in dynamic translucent optical networks, with a distributed
control plane based on the GMPLS protocol suite, during both provisioning and restoration. While the regenerator placement
is assumed to be already decided during network design, the paper is focused on the description of a QoT-aware control
plane able to distribute regenerator availability information and to dynamically designate their utilization (see Tab. I). First,
solutions for accounting and disseminating QoT information (e.g., physical impairment parameters) among control plane nodes
are discussed. Second, several solutions based on GMPLS protocols are proposed for discovering available regenerators and
disseminating the corresponding information among control plane nodes. Third, approaches for the selection and the reservation
of the intermediate nodes, that are designated for regeneration in a translucent lightpath, are presented.
After a brief overview of the solutions to account for QoT in GMPLS-controlled networks, the paper discusses advantages
and drawbacks of the two main approaches for the collection and distribution of regenerator information. The two proposed
approaches are based on GMPLS routing and signaling protocols (i.e., OSPF-TE and RSVP-TE). A routing-based approach
advertises updated information about regenerator availability and capability and may, thus, optimize the connection routing and
the selection of the nodes designated for regeneration. A signaling-based approach collects regenerator information along the
pre-computed connection route and avoids the advertisement of a large amount of information in the control plane. Depending
on the information stored at the nodes, three strategies are defined for designating the node(s) that should perform regeneration
along the translucent lightpath, i.e., regenerator designation performed at source, intermediate, or destination nodes).
The performance of the proposed approaches and strategies are quantified by means of simulations. The comparison is
carried out in terms of blocking probability, set up time, and control plane load. To evaluate the impact of slowly-changing
and rapidly-changing information on regenerator availability, the performance is evaluated and compared in provisioning and
restoration scenarios, respectively.
II. QOT-AWARE GMPLS CONTROL PLANE
In order to guarantee the required QoT to a connection request, the GMPLS control plane needs to acquire physical layer
information for evaluating the QoT on each one of the all-optical segments. This section presents and discusses two techniques
for QoT evaluation and two approaches for disseminating QoT-related information among network nodes.
A. QoT Evaluation
QoT can be evaluated through estimation or measurement. QoT estimation requires the collection of physical layer information
and modeling of physical layer performance. In literature, a number of studies consider a single relevant parameter (e.g.,
equivalent length ) that accounts for the most detrimental physical impairment or for several physical impairments, on
each link. Then, physical layer models combine the parameters of the links and nodes forming the all-optical segment. If the
resulting QoT-estimate is within an acceptable range, the QoT of the all-optical segment is met.
Other studies propose more complex models based on multiple parameters for each physical impairment on a link. Since the
various physical impairments influence each other, the difficulty of such models consists in defining dependable and flexible
relationships, that are able to effectively relate the several parameters. In , a number of physical impairments (i.e., amplified
spontaneous emission, polarization mode dispersion, chromatic dispersion, and self phase modulation) are accounted by using
a single parameter per impairment for each link. The complex model is based on estimation of the optical signal-to-noise ratio
(OSNR) penalty caused by each physical impairment, as follows. Each physical impairment is estimated by combining the
parameter of the links and nodes along the all-optical segment and then it is converted into OSNR penalty. From the OSNR,
the bit error rate (BER) is estimated. When BER is within an acceptable range, QoT requirements of the considered all-optical
segment are met. The utilization of these models allows an a-priori (i.e., before connection set up) estimation of QoT. The
main drawback of QoT estimation is that it could be significantly complex and not sufficiently accurate.
QoT measurement can overcome the QoT estimation complexity and inaccuracy, by measuring QoT on probe traffic or
previously established transparent lightpaths. In , BER is measured on probe traffic, before transmitting data. If the measured
BER is acceptable, then, data transmission can start. Such approach has the drawback of delaying the connection set up, due
to QoT measurement that needs to be carried out.
B. Dissemination of QoT-related Information
QoT-related information (e.g., physical layer parameters or QoT measurements) can be collected and disseminated in the
network by using routing protocols or signaling protocols of GMPLS control plane.
When using a routing protocol, QoT-related information is disseminated among network nodes by means of the OSPF-TE
routing protocol. Each node is required to maintain a new database, referred to as QoT parameter database (QPD), that stores
physical parameters of the whole network.
When using a signaling protocol, QoT-related information are collected by RSVP-TE protocol messages. No QoT parameters
are required to be stored at each node. Therefore, the signaling protocol is used for performing an on-line estimation of the
physical impairments introduced by links and nodes traversed by RSVP-TE messages. Optionally, the signaling approach can
be enhanced by introducing the QPD database at each or some nodes. When a node with QPD is traversed by a signaling
message, the node fills QPD with QoT parameters carried by this message.
By adopting the described approaches, GMPLS control plane can assesses QoT for each connection to be established. When
QoT requirements cannot be met by using a single all-optical segment, regeneration is required. Next, GMPLS extensions for
handling regenerator information are discussed.
III. REGENERATOR-AWARE GMPLS CONTROL PLANE
GMPLS control plane for translucent optical network requires extensions for the dissemination of the information concerning
the regenerator availability and for the reservation of selected regenerators.
A. Dissemination of Regenerator-related Information
GMPLS control plane can be extended to support regenerator availability information (i.e., the number of available regenera-
tors at each node), by using routing protocols or signaling protocols. Regenerator-related information concerns both regenerator
availability (i.e., regenerator state) at each node and regenerator capabilities (e.g., supported bit rate, modulation format, utilized
encoding, and forward error correction type) .
• When using the routing protocol (i.e., OSPF-TE), each node disseminates the information about the state and the type
of locally installed regenerators. Each node is required to store the received information in a local database, referred
to as regenerator database (RD). OSPF-TE is triggered upon each change of regenerator state (i.e., when a regenerator
is reserved or released) . An important benefit of this solution is that routing algorithms can jointly consider link
(e.g., bandwidth) and node (e.g., regenerators) resources. However, this approach introduces some limitations stemming
from the essence of the routing protocol (such as scalability, control plane overload, and stability problems) when the
regenerator information changes frequently (e.g., with dynamic traffic or during restoration). Therefore, the distributed
routing approach seems to be more plausible when the variability of regenerator availability is moderate, i.e., when the
inter-arrival time between connection requests is long.
• When using the signaling protocol (i.e., RSVP-TE), regenerator availability information is gathered during connection set
up. Each node includes information about its available regenerators in the forwarded RSVP-TE messages. If the set up
attempt fails, the regenerator availability information is reported to the source, which can temporary store and exploit it
for successive set up attempts of the current connection request.
The signaling approach can be enhanced by introducing RD at each node. In this case, each node traversed by the RSVP-
TE messages stores the regeneration information carried by the RSVP-TE messages in RD. Since regenerator information
is not disseminated among network nodes, this approach does not suffer from problems of control plane overload, stability,
and scalability. However, information stored in RD may be outdated, as it is updated only by RSVP-TE instances passing
through the nodes.
B. Regenerator Selection and Reservation
The RSVP-TE is used in GMPLS-controlled networks for reserving link and node resources and can be used, also, for
reserving regenerators. The regeneration node(s) can be designated by the source node (i.e., source designation), by any
intermediate node (i.e., self designation), or by the destination node (i.e., destination designation).
• Source designation is performed at the source node, when regenerator information is available in RD. Thus, the source
node routes the connection requests while taking into account the nodes with available regenerators. The source node
is required to include the identifiers of the nodes designated for regeneration in the RSVP-TE Path message. Upon
receiving the message, each node is informed about the source designation.
Regenerator reservation takes place during the backward phase of RSVP-TE. For this reason, the destination node is
required to make a copy of the list of designated regenerating nodes and to include it in the Resv message to be sent to
the source node.
• Self designation is decided locally at each intermediate node. In order to meet QoT requirements, a node can designate itself
for regeneration as follows. Upon receiving the RSVP-TE Path message, a node with available regenerators estimates
the QoT of the all-optical segment terminating at itself. If the all-optical segment has an acceptable QoT, but it could not
be extended of one more hop without exceeding QoT requirements, then the intermediate node designates itself for opto-
electronic regeneration. Designation is advertised by including its own node-identifier in the RSVP-TE Path message to
In the backward phase, the destination node sends a Resv message to the source node containing the list of self-designated
regeneration nodes. By receiving this message, each node knows whether it should reserve one of its own regenerators
for the requested connection.
• Destination designation is performed at the destination node, to meet QoT requirements. Upon receiving the RSVP-TE
Path message, the destination node designates regeneration nodes, based on regenerator availability information stored
in RD and carried by the Path message.
Then, the destination node sends a Resv message to the source node containing the list of designated regeneration nodes.
By receiving this message, each node knows whether it should reserve one of its own regenerators for the requested
When an established connection is released, an RSVP-TE PathTear message is sent from source to destination. Each
intermediate node receiving this message releases the regenerator (if any) reserved for such connection.
IV. PERFORMANCE EVALUATION
Performance evaluation of the GMPLS-controlled translucent optical network is carried out on a Pan-European topology
consisting of 27 nodes and 55 links. Each link is bi-directional and carries 40 wavelengths per direction. Seven nodes,
placed according to , are equipped with a regenerator module composed of 4 regenerators. Connection requests are
dynamically generated following a Poisson process and uniformly distributed among node pairs. Inter-arrival and holding
times are exponentially distributed with an average of 1/λ and 1/µ seconds, respectively. The load offered to the network in
working conditions is, therefore, expressed in Erlang as the ratio λ/µ.
GMPLS control plane is assumed to be QoT-aware with QoT parameter database (QPD) available at each node. QoT is
estimated using OSNR approach, as described in Sec. II-A, according to the model in . Wavelength availability and QoT
for the selected path are checked by consulting, respectively, traffic engineering database (TED) and QoT parameter database
For each connection request, the source node s selects the shortest path (in number of nodes) toward destination node d,
among a set of candidate paths Ps,d. For each source-destination pair, Ps,d consists of paths whose length, in number of
nodes, is within the shortest path length plus one. If a transparent lightpath can be established with acceptable QoT, RSVP-TE
signaling is triggered, without GMPLS extension for regenerators. Otherwise, opto-electronic regeneration is required.
The following regenerator-aware control planes are compared:
• control plane with Regenerator Availability Advertisement (RAA), where regenerator availability information is dissem-
inated by the extended OSPF-TE and stored in RD. The path in Ps,dwith highest number of available regenerators is
selected for the connection request (ties are randomly broken). Source designation of the regenerators is performed with
the aim of minimizing the number of regenerators to be reserved for guaranteeing QoT.
• control plane with Regenerator Availability collected by Signaling (RAS), where regenerator availability information is
carried by the extended RSVP-TE. RD is filled with information received in signaling messages. The path in Ps,dwith
highest number of available regenerators is selected for the requested connection (ties are randomly broken). Source
designation of the regenerators is performed at source with the aim of minimizing the number of regenerators to be
reserved for guaranteeing QoT.
• control plane with Temporary Regenerator Availability collected by Signaling with self designation (TRAS-self), where
regenerator availability information is carried by the extended RSVP-TE. Information received at the source node in
signaling messages is temporarily stored until the signaling for the current connection request is terminated. In particular,
at the first set up attempt, a path in Ps,dis randomly selected and self designation is performed. In case of setup failure,
at successive set up attempts, routing and regeneration selection are performed according to the information reported
to source: the path in Ps,dthat maximizes the number of available regenerators is selected and regeneration nodes are
designated for minimizing the number of regenerators to be reserved.
• control plane with Temporary Regenerator Availability collected by Signaling with destination designation (TRAS-dest),
where regenerator availability information is carried by the extended RSVP-TE. Information received at the source node in
signaling messages is temporarily stored until the signaling for the current connection request is terminated. In particular,
at the first set up attempt, a path in Ps,dis randomly selected and destination designation is performed. In case of set up
100 150 200
Figure 1.Provisioning blocking probability vs. network load.
failure, at subsequent set up attempts, routing is performed according to the information reported to source: the path in
Ps,dthat maximizes the number of available regenerators is selected. At destination, regeneration nodes are designated
for minimizing the number of regenerators to be reserved.
Performance is evaluated in terms of blocking probability, set up time, and control plane load. Results are presented in
provisioning scenario and in restoration scenario.
A. Provisioning Scenario
In provisioning scenario, the mean connection holding time is set to 1/µ = 104s. Offered network load is obtained by
varying 1/λ. RAS, TRAS-self and TRAS-dest perform up to three set up attempts per connection request. RAA performs only
one set up attempt. The provisioning blocking probability is defined as the ratio between the number of blocked connections
and the number of the requested connections. Blocking can be caused by unacceptable QoT (i.e., QoT blocking) due to lack
of regenerators along the selected path, or by the lack of available bandwidth (i.e., wavelength continuity constraint cannot be
satisfied). The set up time is defined as the time elapsed between the connection request and the time when the connection is
successfully established (i.e., the Resv message reaches the source node) and includes the propagation delay, as well as the
transmission and queuing delay experienced at each traversed node. The control plane load is measured in number of control
messages (i.e., RSVP-TE and OSPF-TE messages) delivered per second. The transmission rate of the control plane channel is
100 Mb/s. Both wavelength and regenerator availability information are advertised upon each change of the reservation state.
100 150 200
100 150 200
100 150 200
Figure 2.TRAS-self, TRAS-dest, RAS: provisioning blocking probability after n set up attempts vs. network load.
Fig. 1 shows the blocking probability of TRAS-self, TRAS-dest, RAS, and RAA as a function of the network load. RAA
achieves the lowest blocking probability thanks to the prompt flooding of regenerator availability information performed by
the OSPF-TE routing protocol. After three set up attempts, RAS achieves a blocking probability almost as low as RAA,
while TRAS-self and TRAS-dest perform slightly worse than RAS, without requiring RD database. Blocking probabilities of
TRAS-self and TRAS-dest are comparable.
Fig. 2 shows the provisioning blocking probability of TRAS-self, TRAS-dest, and RAS, after n = 1,2,3 setup attempts as a
function of the network load. RAS outperforms TRAS-self and TRAS-dest for all values of n, thanks to the information stored
AVG. SET UP TIME [MS] AND CONTROL PLANE LOAD [PCKS/S] DURING PROVISIONING.
Avg. Set up Time
Avg. Control Plane Load
100 Erlang200 Erlang
in RD. However, TRAS-self and TRAS-dest performance significantly improves for increasing n, thanks to the utilization of
regenerator availability information reported to the source node. For n = 1 and 2, TRAS-dest performs slightly better than
TRAS-self. Indeed, the destination has the knowledge of regenerator availability information in all the nodes along the path,
while intermediate nodes are oblivious of regenerator availability in downstream nodes. For n = 3, TRAS-self and TRAS-dest
achieve similar blocking probability, as the regenerator availability information reported at source is almost the same, after two
set up attempts. RAS achieves similar performance for n = 2 and n = 3 because it exploits RD since the first set up attempt.
Tab. II shows the average connection set up time and the average control plane load experienced by routing and signaling
approaches when λ/µ = 100,200,300 Erlang.
The average set up time slightly changes in function of the considered scheme and network load. In particular, since most
of the connection requests are established at the first set up attempt, the set up time is similar for all the investigated schemes.
However, higher blocking at the first set up attempt (see Fig. 2) implies a higher set up time. Thus, RAA scheme, that always
uses only one set up attempt, achieves the lowest set up time, while RAS, TRAS-dest and TRAS-self experience increasing
set up time.
When load arises from 100 to 200 Erlang, the set up time slightly increases, since longer paths are selected in Ps,dto satisfy
wavelength continuity constraint. However, when load further increases to 300 Erlang, the average set up time decreases
or stabilizes. At such loads, the absence of a fairness control method causes longer connections to be blocked with higher
probability. Thus, the reduction of the average set up time is stemming from the reduction in the average length of the
The various approaches also experience a similar control plane load because the advertisement of regenerator availability
information represents a minor contribution with respect to the advertisement of wavelength availability information.
B. Restoration scenario
In the restoration scenario, to ensure a fair comparison, the network is always provisioned by using RAS. In this scenario,
different network loads are obtained by varying the holding time 1/µ, while keeping the inter-arrival time 1/λ fixed to 104s.
Single-link failures are randomly and uniformly generated.
RAA and RAS (i.e., the best performing signaling-based scheme) are utilized to restore disrupted connections, using path
restoration scheme. Upon link failure, the node detecting the failure (e.g., the downstream node of the failed link) sends a
notification to the source node which sends an RSVP-TE PathTear message to release resources. Then, the source node
selects a backup path from source to destination within Ps,d(considering the network topology without failed link) as previously
described and RAA or RAS is performed. During restoration, RAS and RAA perform one set up attempt for restoring each
For comparison purposes, segment restoration scheme (SRS)  based on signaling protocol is presented. SRS performs
restoration of the failed all-optical segment. Upon link failure, the node detecting the failure sends a notification to the source
node of the failed all-optical segment (i.e., branch node). Branch node computes an alternative route to the destination node
of the all-optical segment (i.e., merge node). If necessary, it designates the regenerator nodes (as in source designation), by
using the information stored in RD. SRS has the advantage that the restoration connection can exploit the regenerators already
reserved by the working connection.
The restoration blocking probability is defined as the ratio between the number of unsuccessfully restored connections and
the number of connections affected by the fault. A restoration attempt can be blocked due to unacceptable QoT (i.e., QoT
blocking) and due to wavelength contention/unavailability. The restoration set up time is defined as the time between the failure
and the time when the connection is successfully restored. The control plane load is measured in number of control messages
Fig. 3 shows the overall restoration blocking probability and the QoT blocking probability contribution experienced by RAA,
RAS, and SRS. Restoration blocking increases with the offered network load due to the increase of concurrent reservation
instances. RAS and RAA experience similar QoT and overall blocking probability. Even though the number of regenerators in
the network is limited, the regenerator availability information disseminated with RAA does not converge in due time, thus it
does not improve the blocking obtained by RAS. On the contrary, this dissemination significantly increases the control plane
load during restoration (see RAA in Tab. III). The increased control plane load, due to regenerator availability advertisement,
is evident during restoration phase because wavelength availability information are not advertised.
50 100 150 200 250 300
Restoration blocking probability
Figure 3. Restoration blocking probability vs. network load.
AVG. SET UP TIME [MS] AND CONTROL PLANE LOAD [PCKS/FAILURE] DURING RESTORATION, AT 200 ERLANG.
Set up Time
Control Plane Load
SRS strongly reduces QoT blocking and, consequently, the overall restoration blocking with respect to RAA and RAS. In
SRS, by utilizing regenerators reserved by the failed connections, fewer regenerators need to be specifically reserved during
restoration and thus inaccurate regenerator availability information in RD has a lower impact on the blocking. RAA and RAS
achieve similar restoration set up times, while SRS achieves the lowest restoration set up time, since signaling messages are
only exchanged between branch and merge nodes. Finally, the number of control plane messages sent during restoration is
lower in RAS and SRS than in RAA, since the former ones do not disseminate any regenerator availability information (see
This paper presented the most relevant open issues concerning a GMPLS control plane supporting translucent optical network
with QoT guarantees: first, how to evaluate the physical impairment impact on connection QoT; second, how to collect and
disseminate regenerator availability information among network nodes; third, how to reserve and release a regenerator resource.
Schemes based on both routing protocols and signaling protocols were presented to enforce GMPLS-controlled translucent
The approaches were tested in both provisioning and restoration scenarios. Comparison indicates that the routing approach
achieves better performance during provisioning, but suffers from scalability issues, due to the large amount of disseminated
information, especially under dynamic traffic conditions, e.g., during restoration. Signaling approaches avoids the scalability
issues, but may require multiple attempts (leading to a longer connection set up time) to achieve similar performance in terms
Finally, segment restoration between opto-electronic regenerators overcomes the problem of inaccurate regenerator informa-
tion at the node databases and guarantees a faster restoration time. This finding may call for novel provisioning and restoration
approaches for dynamic translucent optical networks.
The work described in this paper was carried out with the support of the BONE-project (“Building the Future Optical
Network in Europe”), a Network of Excellence funded by the European Commission through the 7thICT-Framework Pro-
gramme and by MIUR under FIRB project “Software and Communication Platforms for High-Performance Collaborative Grid”
 S. Gangxiang and R. Tucker, “Translucent optical networks: the way forward,” IEEE Commun. Mag., vol. 45, no. 2, pp. 48–54, Feb. 2007.
8 Download full-text
 S. Sygletos, I. Tomkos, and J. Leuthold, “Technological challenges on the road toward transparent networking,” OSA Journal of Optical Networking,
vol. 7, no. 4, pp. 321–350, Apr. 2008.
 I. Cerutti and A. Fumagalli, “Traffic grooming in static wavelength division multiplexing networks,” IEEE Commun. Mag., vol. 43, no. 1, pp. 101–107,
 G. Shen and W. D. Grover, “Segment-based approaches to survivable translucent network design under various ultra-long-haul system reach capabilities,”
OSA Journal of Optical Networking, vol. 3, no. 1, pp. 1–24, Jan. 2004.
 X. Yang and B. Ramamurthy, “Dynamic routing in translucent WDM optical networks: The intradomain case,” J. Lightw. Technol., vol. 23, no. 3, pp.
1713–1723, Mar. 2005.
 ——, “Sparse regeneration in translucent wavelength-routed optical networks: architecture, network design and wavelength routing,” Photonic Network
 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. 76–85, Dec. 2006.
 N. Sambo, N. Andriolli, A. Giorgetti, F. Cugini, L. Valcarenghi, and P. Castoldi, “Distributing shared regenerator information in GMPLS-controlled
translucent networks,” IEEE Commun. Lett., June 2008.
 N. Sambo, F. Cugini, I. Cerutti, L. Valcarenghi, P. Castoldi, J. Poirrier, E. Le Rouzic, and C. Pinart, “Probe-based schemes to guarantee lightpath quality
of transmission (QoT) in transparent optical networks,” in Proc. of European Conference on Optical Communication, ECOC 2008, Sep. 2008.
 F. Cugini, N. Sambo, N. Andriolli, A. Giorgetti, L. Valcarenghi, P. Castoldi, E. Le Rouzic, and J. Poirrier, “Enhancing GMPLS signaling protocol for
encompassing quality of transmission (QoT) in all-optical networks,” J. Lightw. Technol., vol. 26, no. 19, pp. 3318 – 3328, Oct. 2008.
 G. Bernstein, Y. Lee, and W. Imajuku, “Framework for GMPLS and PCE Control of Wavelength Switched Optical Networks,” draft-bernstein-ccamp-
wavelength-switched-02.txt, Oct. 2007, Internet Draft.
 N. Sambo, I. Cerutti, F. Cugini, A. Giorgetti, L. Valcarenghi, and P. Castoldi, “Segment restoration scheme with QoT-guarantees in GMPLS-controlled
translucent networks,” in Proc. of European Conference on Optical Communication, ECOC 2008, Sep. 2008.