JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO.

20, OCTOBER 15, 2009 4415

Optical Core Networks Research in the

e-Photon-ONe+ Project
Franco Callegati, Member, IEEE, Filippo Cugini, Paul Ghobril, Sebastian Gunreben,
Víctor López, Student Member, IEEE, Barbara Martini, Member, IEEE, Pablo Pavón-Mariño, Member, IEEE,
Marcell Perényi, Namik Sengezer, Student Member, IEEE, Dimitri Staessens, János Szigeti, and
Massimo Tornatore, Member, IEEE

Abstract—This paper reports a summary of the joint research years (2004–2005) and as it proved successful, for two
activities on Optical Core Networks within the e-Photon-ONe+ more years (2006–2007) under the e-Photon-ONe+ name. The
project. It provides a reasonable overview of the topics considered e-Photon-ONe community currently supports the BONE project
of interest by the European research community and supports the
idea of building joint research activities that can leverage on the (http://www.ict-bone.eu) that stemmed from the previous ex-
expertise of different research groups. perience. The e-Photon-ONe consortium gained worldwide
visibility and reputation. An example is the co-sponsorship
Index Terms—Congestion resolution, GMPLS, optical net-
works, optical packet switching, physical impairment, protection, (with COST and NSF) of the “US/EU Workshop on Key Issues
restoration, service oriented networks, traffic engineering, wave- and Grand Challenges in Optical Networking” [3].
length routing. The size of the project raised significant problems of manage-
ment. The concept of Virtual Departments (VDs) was defined
as the container and promoter of activities aimed at achieving
I. INTRODUCTION
durable integration, i.e., to promote joint research activities
(JAs), identify new research topics, etc..

E photon/one was a Network of Excellence (NoE) funded by

the European Commission (EC) in the context of the 6th
Framework Programme (FP6) with the primary goal of fostering
This work reports the main results of the activities developed
by the Virtual Department on optical core networks and tech-
nologies (VD-C) in the last two years of the project (results of
the integration of European research institutions active in optical the previous period are summarized in [4]). The complete list of
networking research [1], [2]. JAs in VD-C is presented in Table I with list of participants.
e-Photon-ONe was a large project, involving about 40 The rest of the paper is organized as follows. Section II pro-
institutions and 500 researchers. It was funded for two vides an overview of the reference network scenario. Then a
summary of selected JAs is reported organized per topic: traffic
engineering in Section III, network resilience in Section IV, op-
Manuscript received August 03, 2008; revised November 15, 2008, March tical packet switching in Section V, and finally service oriented
20, 2009, and May 14, 2009. First published May 29, 2009; current version
published August 28, 2009. This work was supported by the e-Photon/ONe+ and optical core networks in Section VI.
BONE (“Building the Future Optical Network in Europe”) projects funded by
the European Commission through the 6th and 7th ICT-Framework Programme. II. REFERENCE SCENARIO
F. Callegati is with the Department of Electronics, Computer Sciences and
Systems, University of Bologna, Italy (e-mail: [email protected]). The typical Optical Core Network (OCN) architecture com-
F. Cugini and B. Martini are with CNIT, Pisa, Italy (e-mail: filippo. prise a data plane (DP) and a control plane (CP). The former is
[email protected]; [email protected]). responsible for user data flow forwarding and here is assumed
P. Ghobril was with Orange Labs, Lannion, France. He is now with Envergus,
Lannion, France (e-mail: [email protected]). that it mainly exploits all-optical switching. The latter is re-
S. Gunreben is with the Institute of Communication Networks and sponsible for the logical networking functions, e.g., routing, re-
Computer Engineering, University of Stuttgart, Germany (e-mail: silience and management.
[email protected]).
V. López is with the Departamento de Ingeniería Informática, E.P.S., Uni- Alternatives for multiplexing and switching in the DP range
versidad Autónoma de Madrid, Spain (e-mail: [email protected]). over a wide set of alternatives, providing different trade-offs
P. Pavón-Mariño is with the Department of Information Technologies and between flexibility and complexity; from fibre and wavelength
Communications, Polytechnic University of Cartagena (UPCT), Cartagena,
Spain (e-mail: [email protected]). switching to sub-wavelength switching (optical time division
M. Perényi and J. Szigeti are with the HSNLab, Department of Telecom- multiplexing, OTDM, optical burst switching, OBS, optical
munications and Media Informatics (TMIT), Budapest University of Tech-
nology and Economics (BME), Hungary (e-mail: [email protected];
packet switching, OPS).
[email protected]). Because of the large traffic flows carried by the OCNs, critical
N. Sengezer is with the Department of Electrical and Electronics Engineering, issues for the CP are reliability and network survivability as well
Bilkent University, Ankara, Turkey (e-mail: [email protected]).
D. Staessens is with Department of Information Technology (INTEC), Ghent
as traffic engineering and contention resolution. GMPLS offers
University-IBBT, Belgium (e-mail: [email protected]). capabilities able to address many of these issues and is a major
M. Tornatore is with the Dipartimento di Elettronica e Informazione (DEI), candidate for OCNs’ CP.
Politecnico di Milano, Italy (e-mail: [email protected]). New topics are also emerging when considering future mass
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. market, bandwidth-greedy applications, such as Grid Com-
Digital Object Identifier 10.1109/JLT.2009.2024090 puting and Service Delivery Platform, requiring “on demand”
0733-8724/$26.00 © 2009 IEEE
4416 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 20, OCTOBER 15, 2009

TABLE I
JOINT ACTIVITIES IN e-Photon-ONe+ VD-C

network services with configurable bandwidth, availability, different focus. It assumed a network with different switching
end-to-end delay, etc. Service architectures have been defined granularities and proposed a comparison framework to as-
by the principal standardization bodies, such as the IP Multi- sess the advantages/drawbacks of using a dynamic switching
media Subsystem (IMS) by 3GPP [5] and the Next Generation technology (for instance OBS) with respect to a more coarse
Network (NGN) by ITU-T [6]. Unfortunately none of them wavelength switching [12], [13].
foresee any exploitation of the Generalized Multi Protocol
Label Switching (GMPLS) CP capabilities. This opens a whole A. Regular Reconfiguration of Light-Trees in Multilayer
set of new problems that are particularly important for OCNs. Optical Networks
All these issues were addressed at some extent by VD-C, with Multicast (MC) applications will likely increase in the future
the sole exception of Optical Burst Switching. A separate work- Internet [14] and network engineering suggests implementing
package, working in collaboration with VD-C, was devoted to multicast delivery in the lowest layers of the network to avoid
this topic but formally reported results separately. A summary waste of bandwidth due to unicast-based distribution of MC
of the activities on OBS can be found for instance in [7]. flows [8], [15], [16].
The problem considered here that of dynamic multicast (MC)
trees, where the members are continually changing, causing a
III. TRAFFIC ENGINEERING
degradation of the tree as it diverges from the optimum. Regular
Four joint activities addressed traffic engineering mainly reconfiguration of the MC tree can solve this degradation, but it
focusing on multilayer (ML) networks. The first activity con- also has drawbacks: computation effort, short disruption in the
cerned multicast in ML networks and showed the benefit of data transmission flow, additional signalling overhead. There-
reconfiguring the multicast trees periodically to cope with fore, it is important to understand the cost/benefit trade-off of
traffic variations [8], [9]. The second activity compared traffic reconfiguring the MC tree.
engineering (TE) strategies in ML networks. TE may be imple- The reference scenario is a two-layer network, where the
mented in the upper electronic layer only, or in both electronic upper electronic layer is packet switching capable, while the
and optical layer with periodic topology reconfigurations of the lower, optical layer is wavelength switching capable.1 The
latter. Results showed the benefit of traffic flow reconfiguration traffic consists of dynamic, multicast delivery demands, for
in comparison to a statically configured network [10]. The third instance a digital media distribution service, where the audience
activity focused on the realization issues of a traffic engineering is varying in time.
algorithm for ML optical networks which is implemented 1The electronic layer can perform traffic grooming. The control plane has
in an optical test-bed [11]. The last activity had a slightly information about both layers and both layers are involved in routing.
CALLEGATI et al.: OPTICAL CORE NETWORKS RESEARCH IN THE e-Photon-ONe+ PROJECT 4417

Fig. 1. Average additional cost of routing after reconfiguration.

Fig. 2. Total, network, and reconfiguration costs as a function of the reconfig-
uration period. The minimum of the total cost curve suggests the optimal length
of the reconfiguration period.
The optimal design of the MC tree can be solved by using
Integer Linear Programming (ILP) [8]. Unfortunately the exact
solution is NP-complete, and heuristic algorithms can be defined B. Implementation and Experimental Verification of a
to make the computation more efficient [16]. Obviously these Multilayer Integrated Routing Scheme for Traffic Engineering
solutions depend on the distribution of the members of the tree.
When members leave or enter the multicast tree an good solution This research activity studied a multilayer routing algorithm
is applicable only with a reconfiguration of existing paths. and its real life implementation in a test-bed including an elec-
The alternative to reconfiguration is to modify the tree tronic layer (SDH) on top of an optical transport layer (WDM)
without reconfiguring it completely. Heuristics were defined with a common CP based on GMPLS.
to this end, presented in [9]. The results presented here refers The CARISMA test-bed [11] is essentially a wave-
to the Accumulative shortest path (ASP) heuristic, that simply length-routed optical network. It sets-up end-to-end con-
connects newly arriving endpoints to the MC tree applying nections as lightpaths and also allows the set-up of finer
Dijsktra’s algorithm, and clears branches leading to departed grain connections thanks to the Forwarding Adjacency (FA)
endpoints. concept. It permits the aggregation of higher-order LSPs into
The results refer to the COST 266 European reference these lower-order ones. The routing protocol advertises these
network [17]. The ILP problem was solved using the CPLEX lower-order LSPs as FA-LSPs. The nodes may use FA-LSP
optimizer. The number of wavelengths per link was 8. Light- for path computation, nesting lower-order LSPs into FA-LSPs
paths can be routed up to the electronic layer to perform (tree-) [18]. The FA functionality was implemented with a proprietary
branching in any node. However, O/E (optical to electronic) extension to the routing protocol.
and E/O conversion was assumed to be twice more expensive The TE algorithm implemented is called Weighted Integrated
than switching in the optical layer. Routing (WIR) and was proposed in [19]. It assumes a co-loca-
It is assumed that the tree is periodically re-configured. The tion of the electrical and optical nodes and operates in a two-step
arrival of a new demand or the departure of an existing one is approach: the first step includes a shortest path search from
called an event. Fig. 1 shows the average additional cost increase source to destination; the second step chooses a subset of inter-
of the multicast tree as a function of the number of elapsed mediate nodes and checks the feasibility of a path using these
events after reconfiguration, split in routing cost, number of nodes. The optimal path is chosen according to the cost, in-
O/E, E/O conversion ports and number of wavelengths used. As cluding number of wavelength conversions, number of electrical
expected the more the events after reconfiguration the more the or optical hops or link occupancy.
MC tree diverges from the optimal and the greater the additional The implementation of the WIR faced some restrictions im-
cost. posed by the test-bed.
The interesting result is that, if reconfiguration is imple- • Wavelength continuity: GMPLS does not support it. The
mented, it is possible to identify an optimal length of the test-bed implementation includes this information using a
reconfiguration period, if we take into account the negative nonstandard OSPF-TE extension.
aspects of reconfiguration as a penalty (see Fig. 2). Our further • Number of optical hops: The FA does not provide any in-
results show that the reconfiguration seems to be especially formation of the underlying optical links and hides the
useful if grooming is not possible. Still, a number of technical number of optical nodes in the path to the electronic layer.
challenges must be addressed to make reconfiguration practical, • Cost metrics: The cost metrics available are limited; here
like the seamless switchover of traffic from the old to the new were used the number of electrical and of optical hops (if
tree. available).
4418 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 20, OCTOBER 15, 2009

tend to neglect the effect of control plane on routing perfor-

mance. The JA summarized in Section IV-A tackle this problem
[23].
Similarly, while propagating through the optical network, a
signal may degrade in quality as it encounters physical impair-
ments, and this may in turn make the Bit Error Rate (BER) at
the destination unacceptably high. Thus, Quality of Transmis-
Fig. 3. Network topology used for simulation. The links are uni-directional, sion (QoT) parameters have to be introduced in the GMPLS pro-
carrying two wavelengths with STM-16 capacity. tocol suite to provide reliable lightpath provisioning. The last JA
investigates these issues [24], [25].

A. Effects of Outdated Information on Protected Routing in

WDM Networks

This activity concentrated on the general effects of outdated

information in a control-plane enabled optical network, a
problem not investigated to date. In distributed-GMPLS net-
works, each node builds a network image to identify the best
path to route a connection (source routing).
As an effect of control delays, this image may not be up-
dated and routing not optimized. The information to build this
image depends on various factors, e.g., which protection is ap-
plied and whether wavelength conversion is enabled. Moreover,
in the case of shared protection, the state of shareable backup
resources has to be disseminated and ad-hoc routing protocol
Fig. 4. Comparison of the connection blocking probability. (such as OSPF-TE) extensions may be needed [26].
The contributions to the control delay can be summarized as:
information propagation delay, set-up delay, switching delay,
The WIR algorithm was tested in the optical test-bed. Its per- processing delay, periodical database update. It was proposed
formance was also evaluated by simulations in a simple network to evaluate them using a simple delay model based on the fol-
(Fig. 3). In the simulation, the CP advertises the link capacity lowing assumptions.
with STM-4 granularity. The traffic demands request STM-4 1) Constant Control Delay: The control delay is fixed and
bandwidth and are uniformly distributed. equal to .
2) Negligible Set-Up Time: Once routing has been identified,
Fig. 4 shows the connection blocking probability comparing
provisioning occurs without no delays.
test-bed and simulation results. Over the considered load range,
3) Identical network vision: All the nodes share the same net-
measurement and simulation fit well.
work vision, referred to the instant .
The outcome of this activity showed that the ML routing al-
Our simplified approach enables the effect of a wide range of
gorithm is in principle realizable in a real network, although
control delay values to be quantified, independent of the specific
some limitations were either solved by proprietary protocol ex- routing/control algorithms and parameters (update frequency,
tensions or skipped in the implementation. In particular the im- amount of information).
plementation required the extension of the GMPLS CP. In summary, in the JA the control-delay effects on routing
performance were analyzed [23], using realistic case-study net-
IV. NETWORK RESILIENCE work topologies in a dynamic network environment. The study
considered a network without protection, with dedicated path
Network resilience in OCNs has been widely investigated in protection (DPP) and with shared path protection (SPP) with
the past, nonetheless, some of the recent progresses pose new (VWP) and without (WP) wavelength conversion. The consid-
challenges. It was already mentioned that OCNs will likely in- ered metric is the blocking probability (BP).
terconnect several optical domains. It is reasonable to assume Fig. 5 shows the BP for the DPP and SPP case (under WP
that the end-users’ expectation is that they get the same or near and VWP assumption). Curves are plotted as a function of ,
the same reliability for interdomain as for intradomain connec- considering 100 arrivals per second (which leads to a network
tions. Therefore, protection and restoration in multidomain sce- load of around 0.55). The delay on the -axis is a relative mea-
narios is a key issue that was investigated by two JAs [20]–[22]. sure, expressed as the ratio between the absolute delay and
Furthermore, the emerging applications call for an intelligent the average holding time . These curves show three distinct
optical network control plane, such as that provided by GMPLS. phases.
Unfortunately the control plane may impact on the network per- • Phase 1—Not influential delay: In this first phase, the BP
formance, and most of the studies on dynamic traffic routing is constant and it is not influenced by the delay.
CALLEGATI et al.: OPTICAL CORE NETWORKS RESEARCH IN THE e-Photon-ONe+ PROJECT 4419

Fig. 5. Total blocking probability for DPP and SPP routing in VWP and WP Fig. 6. Blocking probability within k (transparent) set up attempts and after
case. the successive set up attempt exploitingN regenerators per node. (a)k=1 ;
(b)k=2 ; (c)k=3 .

• Phase 2—Linear increase: Outdated information starts af- is equipped with N shared-per-node regenerators. Connection
fecting the quality of source-routing, causing a significant requests are dynamically generated with uniform distribution
and linear increase of the BP. among all node pairs. Network load is kept limited in order to
• Phase 3—saturation: The BP is not affected by increases experience connection blocking due mainly to unacceptable
in the control delay, since the network image at the source QoT or lack of regenerators.
node is now uncorrelated to the actual network state. Fig. 6 shows the blocking probability (BP) of T connections
As a matter of fact, the provisioning of a connection over a within set up attempts ( equal to 1, 2 and 3). In addition it
given path in the VWP case fails only in the case where all the shows the BP of the first NoT set up attempt performed upon
channels on a link are saturated, while in the WP case only the the unsuccessful -th transparent set up attempt. In this example
chosen channel has to be free to allow a successful provisioning the nodes are equipped with shared-per-node regenera-
of the connection over the chosen path. tors. Results show that, by exploiting successive set up attempts,
the overall BP of transparent connections decreases. In addition,
B. QoT-Aware Control Plane also the NoT connection BP, due to the lack of regenerators, de-
The set up of transparent connections (T, i.e., lightpaths) creases with the increase of the number of transparent con-
or nonfully transparent connections (NoT, i.e., lightpaths with nection set up attempts. Indeed, a higher number of explored
some intermediate nodes performing opto-electronic regenera- routes and nodes allows a saving in regenerators and improves
tion), requires the enhancement of the GMPLS protocol suite the likelihood of establishing NoT connections. However, while
to include information related to both Quality of Transmission the increase from to leads to significant BP reduc-
(QoT) and to the presence of shared-per-node regenerators [27]. tions, increasing from 2 to 3 provides negligible reductions.
In this study, the Signalling Approach (SA)-based GMPLS Thus, just two set up attempts before resorting to regenerators
enhancement proposed in [24] is considered. guarantee the best performance.
In SA, no extensions are introduced in the routing protocol
which calculates routes ignoring QoT. Then, SA performs the V. CONTENTION RESOLUTION STRATEGIES IN OPTICAL
dynamic estimation of the QoT during the signalling phase by PACKET SWITCHING
collecting QoT parameters from intermediate nodes. At the des- The focus of the research on OPS was on solving contention
tination node, if the accumulated information is within an ac- by the use of scheduling algorithms exploiting wavelength con-
ceptable range, the lightpath set-up request is accepted. Other- version and delay lines in a combined way.
wise the lightpath request is rejected and further set up attempts In the former activity a new comparison metric is defined
following possibly link-disjoint routes are triggered. that allows to look at the complexity/performance trade-off of
The main advantage of SA is that it avoids the flooding of different scheduling alternatives under a new perspective [28],
QoT parameters and regenerator availability and preserves [29]. The latter activity explores the issue of designing sched-
control plane scalability. However, it may increase the amount uling algorithms that are able to maintain the packet sequence
of control plane packets and delay the lightpath establishment and, therefore, may be suitable for QoS sensitive traffic [31],
process. Expanding upon [24], this activity evaluated the per- [32].
formance of the SA when both QoT and shared regenerator Both activities referred to an OPS switching system able to
information are considered. The performance is evaluated by emulate output queuing with delay lines and converters shared
means of a custom built C++ event-driven network simulator. per output port. The input/output ports are equipped with
A Pan-European topology with 17 nodes and 32 links is consid- fibres each, carrying wavelengths and with delay lines.
ered [25]. Each link carries 40 wavelengths. Each network node Consequently the number of input/output channels per node is
4420 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 20, OCTOBER 15, 2009

. As usually in these studies, the focus is on the packet

loss probability (PLP).
The JA here reported focused on contention resolution algo-
rithms that exploit the wavelength and time domains in a co-
ordinated manner. This is called Channel and Delay Selection
(CDS) scheduling. Several algorithms were proposed to solve
this problem but a general framework to compare different al-
ternatives was missing. We defined the concept of scheduling
space and proposed to use it as the basis to assess complexity/
performance trade-offs.
Given an arriving packet that will be forwarded to a given
output port, the CDS algorithms must choose on which of the
channels potentially available on that port the packet
will be transmitted. The scheduling space is the set of choices

Fig. 7. PLP as a function of the delay unit D comparing FWC and LWC, for a
switch with jS j = 64 .
where gives the delay out of the available per channel,
gives the wavelength and the fiber. The number of elements
(i.e., the cardinality) in is . To solve the first problem a service architecture, namely Ser-
In an ideal switching matrix with full range wavelength con- vice Oriented Optical Network (SOON) architecture, based on
version (FWC), . However, in real sys- distributed signalling among designated service nodes, has been
tems as a result of hardware or software limita- designed to fulfil service requests issued by applications while
tions. For instance, if the switching matrix is equipped with lim- masking the transport related implementation details from the
ited range wavelength converters (LWC), the wavelengths abstract request of the service. To solve the second problem, a
per fiber are divided in wavebands of wavelengths and con- techno-economic algorithm is proposed to help MPLS routers
version may happen only within the same waveband, therefore take the decision whether to switch traffic flows (Label Switched
. Paths or LSPs) optically or electronically.
We believe is a measure of the cost of the CDS algorithm,
since it is correlated to the amount and kind of devices needed to A. Advanced Connectivity Service Provisioning in GMPLS
implement the switching matrix, and a fair comparison between Networks
scheduling algorithms must be done with the same values of . The SOON architecture consists of a GMPLS-enabled trans-
The engineer has to dimension the , , parameters. port network on top of which is added a new functional layer,
For instance it is known that, in general, it is more profitable to called Service Plane (SP) [33]. The SP translates a network ser-
invest in channels rather than in delays [30]. But when we com- vice request issued by an Application Entity (AE) into a set of
pare keeping fixed we discover that the best performance technology-dependent directives to the network devices. The SP
is obtained with 32 wavelengths and 1 delay per interface, and is composed by one Centralized Service Element (CSE) and
not with 64 wavelengths and no delays as known results would a number of Distributed Service Elements (DSEs). The CSE
suggest. performs AE identification and authorizes the relevant service
Another nonintuitive result is that LWC is not necessarily requests using the information stored in its Service Level Agree-
worse than FWC. Comparing FWC and LWC with the same ment (SLA) database. The DSEs process network service re-
scheduling space it happens that a small increase in delays may quests via a User to Service Interface (USI), and interact with the
well compensate the limited range conversion, whilst also im- other DSEs to perform the necessary technology-specific net-
proving the overall performance as shown by the example in work setting into the controlled edge network nodes via a User
Fig. 7 [28]. to Network Interface (UNI).
An implementation of DSE and CSE was realized in Java to
validate the SOON architecture Fig. 8. The case study presented
VI. SERVICE ORIENTED OCNs
here refers to the on-demand set-up of L2/L3 VPN with QoS
To provide advanced and QoS-enabled connectivity services, assurance across a MPLS network.
to new IT application such as Global Grid Computing, the OCNs First of all the signalling delay was evaluated using two PCs
must be enhanced with the capability to interact with the appli- connected to the Customer Edge (CE) routers. Each PC runs an
cations and consistently perform the network resource alloca- instance of VLC media player; the former configured as a Video
tion. The problems in this field were: Server transmitting DVD video, the latter as Video Client. The
• the support for direct invocation and fulfilment of QoS- time needed by the SP to fulfil the service request (i.e., overall
enabled connectivity services [33], [35]; service provisioning time) is about 13 s. This time is not signif-
• the implementation of decision algorithms to share the re- icantly affected by the number of routers involved, since the SP
sources among the incoming service and map the traffic configures them in parallel and approximately at the same time.
flows on the network resources [36]. In particular, the processing time of the SP is about 1.8 s and
CALLEGATI et al.: OPTICAL CORE NETWORKS RESEARCH IN THE e-Photon-ONe+ PROJECT 4421

Fig. 8. SOON testbed for on-demand VPN set-up and QoS validation.

Fig. 10. Optimal decisions for several T values assuming hard real-time
utility function (Dashed line = Utility).

The goal of the algorithm is to obtain the optimal decision for

the routing of the LSPs such that the Bayes risk is minimum.
This is combined with the definition of a set of utility func-
tions that measure the QoS experienced (in terms of queuing
delay) by the electronically-switched packets. Three utility
functions are proposed.
Fig. 9. Throughput versus elapsed time for HD streaming video and best effort • Mean utility: Computes the mean delay of the LSPs in the
traffic.
electronic domain.
• Hard-real time utility: Evaluates the probability that
the delay in the router queue is lower than a given
the time needed by the router to elaborate the UNI commands threshold.
is about 2.5 s [34]. • Elastic utility: Assesses the gradual degradation of elastic
To validate also the QoS QoS capabilities of proposed solu- services.
tion best effort traffic, mapped to the DiffServ BE class, and Finally a metric is introduced to quantify the relative cost of
gold traffic, mapped to the Diffserv Expedited Forward (EF) optical switching with respect to electronic switching .
class, were mixed. Specifically, the link under test was loaded We have considered a linear cost approach, that evaluates the
with a High Definition (HD) Video Stream tagged in Gold ratio at which the optical cost increases with respect to the elec-
Class (about 20 Mb/s). Then traffic congestion was produced tronic cost.
by using a traffic generator and applying a load equal to 980 Fig. 10 shows the risk function and the utility function
Mbps (leading to an overall traffic load equal to 100%) marked (dashed line) when QoS constraints changes assuming the
in BE Class. The objective throughput and the user perceived hard-real time utility. The optimal decision is given by the mark
quality were evaluated. As desired the Gold class steadily in all curves. For instance the optimal decision for ms
maintained the throughput (Fig. 9), permitting and excellent is 43; i.e., 43 LSPs out of 60 are switched using the electronic
video quality, thus proving that the QoS requirements of the layer.
video traffic were well satisfied [35]. Fig. 11 shows the impact of using mean utility function,
when the number of incoming LSPs in the system increases.
B. Multilayer Switching Algorithm for an All-Optical Router refers to the relative cost of optical switching with respect
to electronic switching. When optical switching becomes ex-
When the switching nodes have multiple switching alterna- pensive (large values of ), less LSPs are switched optically.
tives (electronic, optical fibre based, optical wavelength based, Regarding the characteristics of the traffic sources, the results
optical packet based etc.) an important question to answer is obtained show that the mean or variance of the incoming flows
how to map the traffic flows on the switching layers. This ac- influence the decisor behaviour and helps to change the deci-
tivity proposed a solution based upon a set of suitably chosen sion based on the traffic features [36]. However, when optical
metrics [36]. switching becomes too expensive, the is critical in the op-
First of all is computed a loss function that is used to quantify timal decision, thus cancelling any influence of the other param-
the effect of congestion. Based on the loss function is computed eters. In this light, the network operator has to decide where the
the Bayes risk [37] which is the expectation of the loss function optimal decision lies, trading off the parameter and the
as a function of the cost of queuing for a certain amount of time. incoming traffic parameters.
4422 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 20, OCTOBER 15, 2009

[12] P. Ghobril, “Analytical model for dynamic waveband switching,” pre-

sented at the ONDM, Catalonia, Spain, Mar. 2008.
[13] P. Ghobril, “A novel analytical model for dynamic waveband
switching,” Optical Switching and Networking, Mar. 2009, accepted
for publication.
[14] B. Quinn and K. Almeroth, IP Multicast Applications: Challenges and
Solutions, IETF RFC 3170, Sep. 2001.
[15] L. H. Sahasrabuddhe and B. Mukherjee, “Light trees: Optical multicas-
ting for improved performance in wavelength routed networks,” IEEE
Commun. Mag., vol. 37, no. 2, pp. 67–73, Feb. 1999.
[16] N. K. Singhal, L. H. Sahasrabuddhe, and B. Mukherjee, “Optimal mul-
ticasting of multiple light-trees of different bandwidth granularities in
a WDM mesh network with sparse splitting capabilities,” IEEE/ACM
Trans. Networking, vol. 14, no. 5, pp. 1104–1117, 2006.
[17] R. Inkret et al., Advanced Infrastructure for Photonic Networks: Ex-
tended Final Report of COST Action 266 Faculty Elect. Eng. Comput.,
Univ. Zagreb, 2003.
[18] J. Comellas et al., “Integrated IP/WDM routing in GMPLS-based op-
Fig. 11. Optimal decisions when R varies using the mean utility function. tical networks,” IEEE Network, pp. 22–27, 2003.
[19] M. C. Necker, C. M. Gauger, and S. Bodamer, “A new efficient inte-
grated routing scheme for sdh/sonet-wdm multilayer networks,” pre-
sented at the OFC, Atlanta, GA, 2003.
VII. CONCLUSION [20] J. Szigeti et al., “p-Cycle protection in multi-domain environment,”
Photonic Netw. Commun., Sep. 2008.
This paper reported a summary of the joint research activi- [21] D. Staessens et al., “Computation of high availability connections in
ties on core optical networks within the e-Photon-ONe+ project. multidomain IP-over-WDM networks,” presented at the Infocom, .
This experience showed that it is possible to leverage on the inte- [22] D. Staessens et al., “Enabling high availability over multiple optical
networks,” IEEE Commun. Mag., vol. 46, pp. 120–126, Jun. 2008.
gration of different expertises to tackle new problems that would [23] M. Tornatore et al., “Effects of outdated control information in con-
prove difficult to be addressed by a single research group. trol-plane-enabled optical networks with path protection,” IEEE/OSA
In the various JAs, the research approach was efficient and J. Opt. Commun. Netw., Jul. 2009, to appear.
focused on selected aspects, aiming at providing guidelines and [24] P. Castoldi et al., “Centralized vs. distributed approaches for encom-
passing physical impairments in transparent optical networks,” pre-
solutions that may be of help to the network engineer in the sented at the ONDM, Athens, Greece, May 2007.
medium/long term. [25] F. Cugini et al., “GMPLS extensions to encompass shared regenerators
We believe this paper shows that optical core networks are in transparent optical networks,” in Proc. ECOC, Berlin, Germany, Sep.
still a lively research topic and, most of all, that solutions and 2007.
[26] R. Martinez et al., “Experimental shared path protection in distributed
evolutionary paths towards their full implementation exist. all-optical GMPLS-based networks,” presented at the DRCN, La
Rochelle, France, Oct. 2007.
[27] B. Ramamurthy et al., “Impact of transmission impairments on the tele-
REFERENCES
traffic performance of wavelength-routed optical networks,” IEEE J.
[1] FP6 Instruments Task Force, Provisions for Implementing Networks Lightw. Technol., vol. 17, no. 10, Oct. 1999.
of Excellence European Commission, Edition, May 12, 2003 [Online]. [28] F. Callegati et al., “Congestion resolution in optical burst/packet
Available: http://cordis.europa.eu/fp6/instr_noe.htm switching with limited wavelength conversion,” presented at the IEEE
[2] F. Callegati, F. Neri, and P. Poggiolini, “The network of excellence Globecom, San Francisco, CA, Nov. 2006.
e-Photon/ONe on optical networks,” presented at the OFC, Anaheim, [29] F. Callegati, W. Cerroni, and G. S. Pavani, “Key parameters for con-
CA, Mar. 2006. tention resolution in multi-fiber optical burst/packet switching node,”
[3] [Online]. Available: http://networks.cs.ucdavis.edu/~mukherje/US- presented at the Broadnets, Raleigh, NC, Sep. 2007.
EU-wksp-June05.html [30] S. Yao, B. Mukherjee, S. J. B. Yoo, and S. Dixit, “A unified study of
[4] F. Callegati et al., “Research on optical core networks in the contention-resolution schemes in optical packet-switched networks,”
e-Photon/ONe network of excellence,” presented at the IEEE Infocom, IEEE J. Lightw. Technol., vol. 21, no. 3, pp. 672–683, Mar. 2003.
Barcelona, Spain, Apr. 2006. [31] P. Pavon-Marino, F. J. Gonzalez-Castano, and J. Garcia-Haro, “Round-
[5] M. Poikselka et al., The IMS: IP Multimedia Concepts and Services, Robin wavelength assignment: A new packet sequence criterion in op-
2nd ed. New York: Wiley, 2006. tical packet switching SCWP networks,” Eur. Trans. Telecommun., vol.
[6] K. Knightson et al., “NGN architecture: Generic principles functional 17, no. 4, pp. 451–459, 2006.
architecture, and implementation,” IEEE Commun. Mag., vol. 43, no. [32] J. Veiga-Gontan, P. Pavon-Marino, and J. Garcia-Haro, “Reduction of
10, 2005. packet order performance penalty in IBWR switches,” presented at the
[7] J. Aracil et al., “Research in optical burst switching within the 9th Int. Conf. Transparent Optical Networks, Sousse, Tunisia, Dec.
e-Photon/ONe network of excellence,” Opt. Switching Netw., vol. 4, 6–9, 2007.
no. 1, 2007. [33] B. Martini, F. Baroncelli, and P. Castoldi, “A novel service oriented
[8] P. Soproni, M. Perenyi, and T. Cinkler, “Grooming-enhanced multicast framework for automatically switched transport network,” presented at
in multilayer networks,” presented at the ONDM, Athens, May 2007. the Integrated Network Management (IM), Nice, France, May 2005.
[9] M. Perenyi et al., “Regular reconfiguration of light-trees in multilayer [34] F. Baroncelli et al., “A distributed signaling for the provisioning of
optical networks,” presented at the ONDM, Athens, Greece, March ondemand VPN services in transport networks,” presented at the Inte-
2008. grated Network Management (IM), Munich, Germany, May 2007.
[10] N. Sengezer et al., “A comparative study of single-layer and multi- [35] B. Martini et al., “Dynamic QoS control based on VPLS in service
layer traffic engineering approaches on transparent optical networks,” oriented transport networks,” in Proc. ICTON, Jun. 22–26, 2008, vol.
presented at the 9th Int. Conf. Transparent Optical Networks, Rome, 1, pp. 29–32.
Italy, Jul. 2007. [36] V. López et al., “A bayesian decision theory approach for the techno-
[11] S. Spadaro et al., “The CARISMA ASON/GMPLS network: Overview economic analysis of an all-optical router (extended version),” Comput.
and open issues,” presented at the ICTON, Rome, Italy, Jul. 2007. Netw., vol. 52, no. 1, 2008.
CALLEGATI et al.: OPTICAL CORE NETWORKS RESEARCH IN THE e-Photon-ONe+ PROJECT 4423

[37] S. French and D. Ríos Insúa, Statistical Decision Theory. Oxford, Pablo Pavon-Mariño (M’03) received the telecommunication engineering de-
U.K.: Oxford Univ. Press, 2000. gree in telecommunications in 1999 from the University of Vigo (UVIGO),
Spain, and the Ph.D. degree from the Technical University of Cartagena (UPCT)
Franco Callegati (M’98) received the M.S. and Ph.D. degrees in electrical en- in 2004.
gineering in 1989 and 1992 from the University of Bologna, Italy. In 2000, he joined the UPCT, where he is an Associate Professor in the De-
He currently serves as an Associate Professor at the University of Bologna. partment of Information Technologies and Communications. His research inter-
He was a research scientist at the Teletraffic Research Centre of the Univer- ests include performance evaluation, planning, and optimization of communi-
sity of Adelaide, Australia; Fondazione U. Bordoni, Italy; and the University of cation networks.
Texas at Dallas. His research interests are in the field of teletraffic modeling and
performance evaluation of telecommunication networks. He has been working
in the field of all optical networking since 1994 with particular reference to
network architectures and performance evaluation for optical burst and packet Marcell Perényi received the M.Sc. degree in computer science from the Bu-
switching. He has participated in several research project on optical networking dapest University of Technology and Economics (BUTE), Hungary, in 2005. He
at the national and international level, such as ACTS KEOPS, IST DAVID, and is currently pursuing the Ph.D. degree in the Department of Telecommunication
IST Ephoton/ONe, often coordinating work packages and research activities. and Media Informatics.
He has participated in several research projects supported by the EU and the
Hungarian government. His research activities include simulation, algorithmic
optimization, and planning of optical networks, as well as identification and
Filippo Cugini received the Laurea degree in telecommunication engineering analysis of traffic in IP networks, especially P2P, VoIP, and other multimedia
from the University of Parma, Italy. applications. He has experience in planning, optimization, and maintenance of
Since 2001, he has been a research engineer at the SSSUP/CNIT National database systems, web services, and Microsoft infrastructures.
Laboratory of Photonics Networks, Pisa, Italy. His main research interests in-
clude MPLS and GMPLS protocols and architectures, survivability in IP over
WDM networks, and traffic engineering in grid networking.
Namik Sengezer (S’04) received the B.S. and M.S. degrees in electrical engi-
neering from Bilkent University, Turkey, in 2002 and 2004, where he is currently
working toward the Ph.D. degree.
Paul Ghobril received the electrical and electronics engineering degree in 1994 His research interests include design and planning of optical networks and
from the Lebanese University and the Ph.D. degree in computer science and traffic engineering.
networking in 2005 from ENST-Paris (now Télécom ParisTech), France.
He worked from 1994 to 2001 on designing electronic boards while teaching
in major Lebanese universities. He joined France Telecom R&D in 2006 as
a postdoctorate researcher. In 2008, he created Envergus Sarl, a software and Dimitri Staessens received the M.S. degree in numerical computer science in
hardware engineering company located in Lannion, France. His main research 2004 from Ghent University, Belgium.
interest is in modeling, simulation, and optimization of optical network tech- He is now a member of the Department of Information technology, Ghent
nologies. University, and the Interdisciplinary Institute for BroadBand Technology
(IBBT). His research focuses on the design and evaluation of the next genera-
tion of communication networks, and he is currently involved in the European
projects BONE and DICONET.
Sebastian Gunreben received the Dipl.-Ing. degree in mechatronics in 2004
from the University of Stuttgart, Germany.
Since then, he has been with the Institute of Communication Networks and
Computer Engineering (IKR) at the University of Stuttgart where he works on János Szigeti received the M.Sc. degree from the Budapest University of Tech-
traffic engineering for IP-over-WDM networks in several national and European nology and Economics (BME), Hungary, in 2002, where he is currently working
projects. He focuses on control plane aspects of multi-layer networks as well as toward the Ph.D. degree in the Department of Telecommunications and Media
on the formal description of out-of-sequence packet arrivals. Informatics (TMIT).
His research interests focus on routing, design, configuration, dimensioning,
and resilience of IP, MPLS, ATM, ngSDH, and particularly of WR-DWDM-
based multilayer multidomain networks.
Víctor López (S’08) received the M.Sc. degree in telecommunications engi-
neering with Honors from the Universidad de Alcalá in 2005 and the PhD.
degree in computer science and telecommunications engineering with Honors
from the Universidad Autonoma de Madrid in 2009. Massimo Tornatore (S’03–M’07) received the degree (Laurea) in telecommu-
In 2004, he joined Telefonica I+D (R&D) where he was a researcher in next nications engineering in 2001 and the Ph.D. degree in information engineering
generation networks for metro, core, and access. During this period, he partici- in May 2006 from the Politecnico di Milano, Milan, Italy.
pated in several European Union projects (NOBEL, MUSE, MUPBED) focused During his Ph.D. course, he worked in collaboration with Pirelli Telecom
on those topics. In 2006, he joined the Networking Research Group of Univer- Systems and Telecom Italia Labs, and he was a visiting Ph.D. student in the
sidad Autónoma de Madrid as a researcher in the ePhoton/One+ Network of Networks Lab of the University of California, Davis, and in CTTC (Technolog-
Excellence. His research interests are on the analysis and characterization of ical Telecommunication Center of Catalunia), Barcelona, Spain. He is currently
services, design, and performance evaluation of traffic monitoring equipment, a postdoctorate researcher in the Department of Computer Science, University
and the integration of Internet services over WDM networks, mainly OBS so- of California, Davis. He is the author of about 50 conference and journal papers
lutions and IP over WDM architectures. and his research interests include design, protection strategies, traffic grooming
in optical WDM networks, and group communication security.
Dr. Tornatore was a corecipient of Best Paper Award from IEEE ANTS 2008
and the Optical Networks Symposium in IEEE GlobeCom 2008.
Barbara Martini (M’06) received the M.S. degree in electronic engineering in
1999 from the University of Florence, Italy.
She joined Italtel as a Hardware Engineer working on network device drivers
design and TCP/IP stack protocols and Marconi Communications in the summer
of 2000 as Software Engineer involved in network management software design
in DWDM equipments. Since 2003, she has been a Research Engineer at the
CNIT National Laboratory of Photonics Networks located in Pisa, Italy. Her
main research interests include network management system design, GMPLS
optical control planes, and service platform architectures in next generation
networks.