![Pan-European Basic topology with 28 nodes and 41 links [12]. Regenerator pools are sparsely deployed (blue nodes) and dimensioned according to the RRPD algorithm employed.](https://www.researchgate.net/profile/Keren-Bergman/publication/261457661/figure/fig2/AS:296661239779336@1447740959811/Pan-European-Basic-topology-with-28-nodes-and-41-links-12-Regenerator-pools-are_Q320.jpg)
Content uploaded by Keren Bergman
Author content
All content in this area was uploaded by Keren Bergman on May 19, 2015
Content may be subject to copyright.
Cross-Layer Enabled Translucent Optical Network
with Real-time Impairment Awareness
Oscar Pedrola∗† , Balagangadhar G. Bathula∗, Michael S. Wang∗, Atiyah Ahsan∗,
Davide Careglio†, and Keren Bergman∗
∗Department of Electrical Engineering, Columbia University, New York, NY 10027 USA
†Department of Computer Architecture, UPC BarcelonaTech, 08034 Barcelona, Spain
e-mail: opedrola@ac.upc.edu
Abstract—The existing dimensioning strategy for translu-
cent, sub-wavelength switching architectures relies on over-
provisioning, and consequently, overuse of costly, power-
consuming optical-electrical-optical (O/E/O) regenerators. In ad-
dition, due to a variety of external phenomena, many physical
layer impairments are time-varying, and hence, can strongly
degrade network performance. In this work, we introduce a
Cross-Layer Optical Network Element (CLONE) used for the
dynamic management of physical layer impairments in the
network. We investigate the impact of real-time impairment
aware routing in a CLONE-enabled optical network with sub-
wavelength switching. Simulation results show that the CLONE-
enabled network architecture provides improvements in: (1)
energy efficiency by optimizing the usage of regenerators, and
(2) network performance in terms of the packet loss probability.
I. INT ROD UC TI ON
Ultra high-bandwidth, dense wavelength division
multiplexing (DWDM)-powered optical transport
networks (OTNs) are the prevailing communications
infrastructure poised to support the delivery of emerging
bandwidth-hungry applications and services such as video
streaming/conferencing, high-definition TV, and video on
demand, in a cost-effective, energy-efficient way. Moving
forward, forecasts predicting traffic scenarios characterized
by short-lived, small granularity flows, make it crucial for
next-generation OTNs to engage highly agile optical transport
technologies that include sub-wavelength switching [1]. By
leveraging recent advances in nanosecond-range photonic
devices such as fast tunable lasers and fast switching elements
([2], [3]), future OTNs supporting dynamic sub-wavelength
switching can flexibly accommodate diverse traffic conditions
with better efficiency [4].
At the same time, the rapid advances on relevant optical
functions allowing for longer transmission distances, higher
bit-rates, and more closely spaced wavelength channels,
have dramatically increased the sensitivity to physical layer
impairments (PLIs), which accumulate during the signal
end-to-end transmission [5]. For this reason, translucent
architectures have emerged as potential candidates to bridge
the gap between the opaque and transparent networks,
and therefore, to reduce costs and energy consumption in
OTNs [6]. The existing dimensioning strategy for translucent
architectures relies on the offline estimation of PLIs to
strategically deploy a limited number of signal regenerators,
which ensure that a target quality of transmission (QoT)
network performance is met (see e.g., [7], [8]). However, such
an approach results in both over-provisioning and overuse
of regenerators due to inaccuracies in these estimations [9].
Furthermore, the lack of real-time access to physical layer
performance metrics prevents the network from efficiently
adapting to dynamically changing PLIs, and consequently,
network performance can be adversely affected [5], [10]. Note
that, in this paper, the term regenerator implicitly refers to
electrical 3R regenerator, that is, the optical signal undergoes
an optical-electrical-optical (O/E/O) conversion in order to be
regenerated.
In this work, we show that the introduction of real-time
impairment aware routing in a translucent, sub-wavelength
switching network through the novel Cross-Layer Optical
Network Element (CLONE) concept [11], leads not only
to significant energy savings by optimizing the usage of
regeneration resources, but also to network performance
improvement as CLONE-enabled networks can effectively
react to time-varying PLIs.
The remainder of this paper is structured as follows: Section
II, provides a survey on recent work in PLI-aware OTNs
and highlights the main contributions of this work. Section
III, details the problem framework. Section IV, provides and
discusses the simulation results. Finally, conclusions are given
in Section V.
II. RE LATE D WOR K AN D CON TR IBUTIONS
In order to meet a target QoT network performance,
translucent architectures require a limited number of
regenerators to be sparsely deployed across the network. This
is done using a routing and regenerator placement (RRP)
algorithm [7] in wavelength-routed networks, and a RRP
and dimensioning (RRPD) algorithm [12] in sub-wavelength
switching networks. Using a QoT estimator and a pre-specified
minimum signal QoT performance (QoTth), below which
the signal is considered beyond the receiver’s sensitivity,
these algorithms can determine the impact that PLIs will
have on the optical signal, and eventually, the feasibility
of establishing a connection between two network points [13].
Such QoT models are based on two main methods, namely
the numerical calculation of the optical signal to noise ratio
(OSNR) [14], and the analytical or experimental evaluation
of the Q-factor ([7], [15]); both these figure-of-merit have
a direct relation to the signal bit-error-rate (BER) [16]. In
this work, we assume the OSNR as the main signal QoT
performance indicator, and hence, hereinafter we refer to
QoTth as OSNRth.
The adverse effect that PLIs (e.g., amplified spontaneous
emission noise (ASE), polarization dependent loss (PDL),
chromatic dispersion (CD), polarization mode dispersion
(PMD), cross-phase modulation (XPM), self-phase modulation
(SPM) and four wave mixing (FWM)) have on DWDM
systems due to fiber loss, dispersion and non-linearity,
drives research to develop analytical models to accurately
predict their impact [17]. However, these models are still not
mature enough (in terms of both accuracy and computational
efficiency), and hence, it is necessary to consider a penalty
margin (m) (e.g., 2dB extra) when determining an
adequate OSNRth [14]. Since a tight adjustment of mmay
cause a number of connections to be over-estimated (i.e.,
establishment of unfeasible connections), network operators
wishing to guarantee stringent OSNR levels have to set
higher mvalues; this in turn leads to a high number of
under-estimations, and hence, to an over-provisioning of
regenerators [9]. Under these circumstances, the lack of
real-time feedback from the physical layer results in an
overuse of costly, power-consuming regenerators. Moreover,
a fixed penalty margin may still result in strong network
performance degradation, as many PLIs are time-varying and
can be affected by a wide range of higher-order time scale
phenomena such as temperature variations, voltage drifts,
component degradations and network maintenance activities
[18], [19].
Therefore, in order to ensure efficient and robust operation
in future dynamic OTNs, it is necessary to perform reliable
and cost-effective optical performance monitoring (OPM),
and, by this means, gain real-time access to the main physical
layer parameters such as the OSNR and PMD [10]. In fact,
this area has already received great attention in the context
of wavelength-routed networks, where innovative proposals
include, for example, real-time OPM coupled with path
computation element (PCE)-based control planes (CP) to
support dynamic management of either wavelength-switched
optical networks (WSONs) [20] or the more recent elastic
optical networks (EONs) [19].
However, no work as of yet has addressed the challenges of
introducing real-time impairment awareness in a translucent,
sub-wavelength switching scenario, which, due to its statistical
multiplexing nature, requires dedicated OPM and CP solutions
to optimize the use of the available (over-dimensioned) regen-
erators as well as to adapt to time-varying PLIs. To this end, in
this paper we introduce the novel CLONE approach to support
the dynamic management of PLIs. Assuming a pre-deployed
translucent, sub-wavelength switching network, we perform a
series of simulation experiments to compare the performance
of the existing network approach (hereinafter referred to as
STATIC network), where no real-time OPM is available, with
that of a network of CLONEs using realistic time-varying
PLI models. Through real-time access to OSNR measurements
and a dynamic, distributed CP to efficiently disseminate this
data, CLONEs optimize the use of regeneration devices, thus
greatly improving energy efficiency. Moreover, we also show
that the network of CLONES can dynamically adapt to time-
varying PLIs and take decisions on-the-fly to re-route, drop
or regenerate optical packet flows, resulting in improved
network performance. Note that in this work we use the term
packet generically to refer to the optical data unit of the sub-
wavelength optical network (e.g., packets/bursts).
III. PROB LE M FR AM EW OR K
A. Notation
We use G= (V,E)to denote the graph of a sub-wavelength
switching network; the set of nodes is denoted as V, and the
set of bidirectional links is denoted as E. Let f(s→d), denote
a packet flow between source sand destination dnodes, s, d ∈
V. Adequately, let Vx−ddenote the ordered set of nodes that
define the path that f(s→d)has to follow from node xto d,
x, d ∈ V. Let also R ⊆ V denote the subset of nodes that are
equipped with a pool of regenerators. Finally, let Kx−ddenote
a set of pre-computed shortest-path routes from node xto d,
x∈ Vs−d\{d}.
B. RRPD and OSNR models
In [12], we presented several mixed integer linear pro-
gramming (MILP)-based RRPD algorithms. To tackle RRPD,
the routing and the regenerator placement and dimensioning
(RPD) subproblems are solved sequentially so as to reduce
complexity and improve network performance. Specifically, a
source-based routing approach which minimizes congestion in
bottleneck network links and a load-based RPD algorithm are
used in this paper (see [12] for more details). To model the
OSNR, we assume the method as described in [14] and [8]. To
estimate the OSNR level for an optical path traversing klinks
(Posnr), this model requires the OSNR contributions of both
links (Losnr) and nodes (Nosnr ) on such path. Then, Posnr
can be computed as:
Posnr = 1/(
k
∑
i=1
1
Li
osnr
+
k
∑
i=1
1
Ni
osnr
),(1)
Hence, once the routing problem is solved, all paths
whose estimated Posnr (in dB) is lower than OSNRth
are considered as input data for the RPD algorithm, as
regeneration is required at some intermediate node. At this
point, a translucent, sub-wavelength switching network can
be dimensioned through the use of both the RRPD and OSNR
models.
Regarding the OSNR model, it has been shown in [8] that
a static value for both Losnr and Nosnr can be estimated
offline by performing an adequate system power budget
and noise analysis. However, as mentioned in Section II,
network operators have to add a penalty margin (m) on
OSNRth. Hence, the threshold is defined as OSNRth =
OSNRmin +m, where OSNRmin represents the OSNR
tolerance of the receiver and maccounts for OSNR penalties
due to maximum tolerable PMD, residual CD, and all the
other non-linearities. In addition, due to a range of higher
order time-scale phenomena, both Losnr and Nosnr are not
static but time-varying contributions, making the penalty
margin an even more critical parameter to deal with.
In this context, the lack of real-time OPM leads to a
translucent infrastructure which is not able to dynamically
adapt to changing PLI conditions even though it relies on an
overuse of regeneration resources. As previously mentioned,
we refer to this infrastructure as STATIC network.
To address these issues, we first propose to model both
Losnr and Nosnr as Gaussian random-variable functions.
In the context of EONs, authors in [19] assume OSNR
variations in a network link on the order of 1 dB every 30
seconds. Therefore, no abrupt/substantial changes in OSNR
are expected at smaller time-scales (e.g., ms). For each node
and link in the network, we denote such random function
as N(µ[dB], σ[dB]), where µis the mean of the series and
corresponds to the OSNR level estimated offline (link or
node), and σis a certain standard deviation. Under these
conditions, we propose the CLONE concept to provide an
efficient solution for the dynamic management of PLIs in the
network.
In the next subsections, we first detail the main drawbacks
of the STATIC network, and then, provide the features of the
CLONE architecture.
C. The STATIC network
The RRPD algorithm disseminates both the routing and
regeneration information to all network nodes so that they
are able to determine, for each incoming flow of packets,
the corresponding output port and whether such flow has to
be regenerated. For instance, once RRPD determines that
f(s→d)requires regeneration at some node x∈ Vs−d\{s, d},
then f(s→d)will always be regenerated at xindependently
of the actual PLI conditions.
Therefore, due to the lack of real-time OSNR monitoring,
the STATIC network approach exhibits the following
operational issues:
•Packet flows which might not need regeneration as they
have high OSNR (well above OSNRth), are always
regenerated in accordance to the RPD algorithm decision.
Thus, they unnecessarily consume regeneration resources.
•Packet flows whose OSNR level has dropped below
OSNRth cannot be detected, and therefore, continue
their trip until the egress node consuming network
resources. Note that these flows consume unnecessarily
both regeneration (if RPD determined so) and capacity
resources.
•Finally, since Losnr and Nosnr are in fact time-varying
functions, a certain route in the network can become
unfeasible during any given time period. This issue
cannot be detected either, leading to significant increases
in packet loss.
D. The CLONE enabled network
The CLONE concept arises as a result of the ever growing
traffic demand and rising challenges of controlling it,
which dictate the need for the development of innovative
architectures able to provide dynamic, intelligent interaction
between network layers [21]. We envision the CLONE
network model as a promising, integrated platform that
leverages emerging physical layer technologies and systems
to allow for introspective access to the optical layer. Hence,
CLONE-enabled networks will facilitate the retrieval of
real-time OPM measurements which can then be used to
achieve greater energy efficiency and optimized network
performance [22].
1) CLONE architecture:Figure 1, depicts a modular,
generalized description of a CLONE. Featuring a bidirectional
cross-layer signaling scheme between the data, OPM, and
CP planes, the CLONE enables real-time physical layer
measurements affect, for example, re-routing, dropping or
regeneration decisions on a per packet flow basis. Furthermore,
thanks to a distributed, high-speed field-programmable gate
array (FPGA)-based optical CP, which allows CLONEs to
communicate with each other, local OPM metrics can be
efficiently disseminated across the network of CLONEs.
As to the OPM plane, we propose a dedicated OPM device
per input port, embedded directly within the optical layer.
OPM is performed over one of the channels carrying actual
packet-rate data. Since future integrated OSNR monitors are
expected to allow for ultra-fast measurements (e.g., hundreds
of ns [11]), we envision OPM systems able to monitor
ultra-fast, packet-rate channels. In fact, a proof-of-concept
packet-rate OSNR monitor supporting 18 ms packet lengths
has already been experimentally demonstrated [11]. Note
that although in this work only OSNR is considered, we aim
for OPM planes consisting of a set of sub-systems able to
monitor a comprehensive range of PLIs such as OSNR, PMD
and CD. In fact, PMD is also random and time-varying,
and hence, PMD monitoring is crucial to manage highly
reliable, ultra-high speed OTNs. For example, in [23], authors
Fig. 1. Packet-switched CLONE: A system level description indicating the bidirectional information flow between the control, OPM and data planes.
experimentally investigate an OPM technique that extracts
PMD-induced signal degradation from the sum including
degradations induced by others (e.g., OSNR, CD and XPM)
in high speed optical links.
Finally, at the data plane, we assume a translucent node
architecture which is based on semiconductor optical amplifier
technology and has a switching fabric configured as tune-
and-select [8]. It must be noted, however, that regenerator
pools are only available at selected CLONEs in the network,
and that their location and size is determined by the RRPD
algorithm.
2) Real-time impairment aware routing:This section
details the real-time impairment aware routing algorithm
that allows for the network of CLONEs to manage both
regenerations and time-varying PLIs. The implemented
algorithm relies on a distributed CP to efficiently disseminate
the OSNR measurements, which allows it to compute
the physical layer parameters Losnr and Nosnr. Using
the embedded OPM modules, these two attributes can be
computed as explained in [14]. The algorithm is executed at
source nodes s∈ V and intermediate nodes v∈ R. Such
nodes do not operate with full network information (i.e., each
CLONE does not need to be aware of Losnr and Nosnr for
all nodes and links in the network) but they are restricted to
the transparent region they belong to and the physical layer
parameters of nodes and links therein. Specifically, CLONEs
take forwarding/dropping decisions based on whether
f(s→d)can reach the next regeneration node (or d) in its
path. However, since packet flows may traverse a transparent
region without undergoing regeneration, communication
between nodes v∈ R, which delimit transparent regions, is
needed to flood the OSNR history of each f(s→d). To this
end, we use the term OSNRs−d,i to denote the OSNR value
of f(s→d)at the last regeneration node or source i(i.e., i
can either be sor a node v∈ R :Vs−d∋v). For example, in
Fig. 2, f(26→6)following path V26−6={26,15,25,5,6},
might not need regeneration at node 25, and hence, node
25 has to send OSNR26−6,25 to node 5 so that node 5 can
determine whether the flow can eventually reach 6.
As illustrated in Procedure 1, once the algorithm is executed
at node x∈ Vs−d, it first computes the OSNR for f(s→d)at
the next node in the path that is either a regeneration node
or d(lines 1-2). Then, computes the OSNR level at such
node and decides whether f(s→d)can be forwarded with or
without regeneration. It is worth noticing that thanks to real-
Procedure 1 Real-time impairment aware routing algorithm
INPUT: Current node x,f(s→d),OSNRs−d,i,Vx−d
OUTPUT: Forwarding or dropping decision
1: v∗←next node v∈ Vx−d:{v∈ R || v=d};
2: OSNRv∗←compute OSNR at v∗;
3: if OSNRv∗>OSNRmin without regeneration then
4: Forward f(s→d)and exit; /∗Exit the algorithm ∗/
5: else if OSNRv∗>OSNRmin with regeneration then
6: Regenerate and Forward f(s→d)and exit;
7: else
8: for all route k∈ Kx−ddo
9: Re-route f(s→d)through kand generate Vk
x−d;
10: v∗←next node v∈ Vk
x−d:{v∈ R||v=d};
11: Re-compute OSNRv∗;
12: if OSNRv∗>OSNRmin without regeneration then
13: Forward f(s→d)and exit;
14: else if OSNRv∗>OSNRmin with regeneration
then
15: Regenerate and Forward f(s→d)and exit;
16: else
17: Continue;
18: end if
19: end for
20: Drop f(s→d);
21: end if
time impairment awareness the threshold for decision becomes
OSNRmin. In case f(s→d)cannot be forwarded, a set of k
shortest-paths from node xto dare evaluated. If all attempts
fail, f(s→d)is temporarily dropped at node x. Note that
in Procedure 1, if node x=sand x∈ R, a regeneration
will never be performed as the OSNR level is already at its
maximum.
IV. RES ULT S AN D DI SCUSSION
A. Simulation scenario
Simulations are performed considering a large-scale Pan-
European backbone network topology known as Basic (shown
in Fig. 2). We assume that network links are bidirectional each
dimensioned with 32 wavelengths; the transmission bit-rate is
set to 10 Gbps; the traffic is uniformly distributed between
nodes; each node offers the same amount of traffic to the
network; this offered traffic is normalized to the transmission
bit-rate and expressed in Erlangs. Arrivals follow a Poisson
process with a fixed packet size of 1 Mb (100 µs). Since we
consider Poisson arrivals and a constant load, changing the
packet size would imply either reducing or increasing inter-
arrival times, and hence, the packet size does not have any
impact on the packet loss probability (PLP) results obtained
[24]. We consider 19 dB to be the OSNR receiver sensitivity
(OSNRmin) and a penalty m=2 dB, thus OSNRth =21 dB.
RRPD dimensions the translucent network considering a net-
work load of 10.72 Erlangs, and a target loss rate in the access
to regenerator resources equal to BQoT = 10−3. As shown in
Fig 2, such dimensioning results in a sparse placement of 456
Fig. 2. Pan-European Basic topology with 28 nodes and 41 links [12].
Regenerator pools are sparsely deployed (blue nodes) and dimensioned
according to the RRPD algorithm employed.
signal regenerators. CLONE nodes v∈ R monitor load at
regenerator pools’ access and can solve contentions at output
ports using regeneration as long as target BQoT = 10−3is
met. A set of |Ks−d|= 3 shortest-path routes ∀s, d ∈ V
is pre-computed offline and available at each CLONE for re-
routing purposes. Simulations are conducted on the JAVOBS
sub-wavelength switching simulator [25].
B. OSNR scenarios
We consider two different scenarios for σ:
•Scenario 1 (Sc1): σis set to 0.8 dB. In this case, we
have corroborated that no unexpected losses due to PLI
impact occur in the network (i.e., margin mmitigates
perfectly PLI impact). The objective of Sc1is to exhibit
the high energy-savings in terms of regenerator usage
that can be achieved with the CLONE approach.
•Scenario 2 (Sc2): σvaries over time, and randomly can
take either a very high/high value (1.8 dB, 1.4 dB), a
medium value (1.1 dB) or remain the same (0.8 dB)
with probability 0.1,0.1,0.3and 0.5respectively. The
goal of Sc2is to generate PLI situations which cause
some routes to become unfeasible for a certain amount
of time, and thus, force CLONEs to react and re-route
packet flows temporarily.
We assume that every 2seconds a new OSNR value is
generated. However, in order to avoid abrupt variations and
smooth the curve of the series, 20 points are interpolated
between two consecutive Gaussian values. Hence, we assume
that OPM modules report to the CP an OSNR measurement
every 100 ms. Figure 3, shows two randomly selected Losnr
and Nosnr from the Basic topology under the two OSNR
scenarios proposed. Although variations considering a large
050 100 150 200
25
30
35
OSNR value generated (dB)
3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4
Time (seconds)
26
27
28
29
30
31
32
33
34
Link (Sc2)
Node (Sc2)
Node (Sc1)
Link (Sc1)
Link
Node
Fig. 3. OSNR randomly-generated series for both a link and a node in the
Basic topology. Values over 200 seconds (top) and 1second (bottom) highlight
that no abrupt changes in OSNR are expected at the system sampling time-
scale (i.e., 100 ms).
time window (200 seconds on the top) exhibit substantial
variations in the OSNR contributions, a close view to the time-
scale of the CLONE system (1second on the bottom) shows
that no abrupt changes are expected at the system monitoring
time-scale (i.e., 100 ms). Besides, flow forwarding/dropping
decisions are taken between regeneration nodes, and hence,
the largest propagation delay through the CP equals that of
the largest transparent segment in the network. For the network
topology considered, the worst-case delay is on the order of 7
ms. Taking this value into account as well as the smooth OSNR
curve exhibited by both links and nodes at the ms time-scale
allows us to assume that CLONEs take their decisions based
on updated OSNR measurements.
C. Results
In Figs. 4(a) and 4(b), the total PLP represents the sum of
losses due to contention in network links and OSNR losses,
which account for packets being lost at regenerator pools’
access, packets that are dropped due to low ONSR (CLONE
network) as well as packets whose OSNR at the destination
node is beyond the receiver’s sensitivity (STATIC network).
In Fig. 4(a), it is possible to observe that under OSNR Sc1
the overall PLP (left y axis) is the same in both networks.
This is because in this network scenario OSNR losses are
negligible with respect to the contention ones, and hence,
do not have a noticeable impact on the total PLP. However,
Fig. 4(a), also exhibits the great optimization of regenerator
resources that is achieved with the CLONE approach. Around
two orders of magnitude difference in losses caused by
contention at regenerator pools are achieved, which means
that the CLONE network, using real-time impairment aware
routing, only regenerates those flows that really need it. As
a result of this efficient usage of regeneration resources, the
CLONE network is able to provide, regardless of the load, a
reduction of more than 60% in the average per packet number
of regenerations (ppr) (right y axis).
On the other hand, in Fig. 4(b) (Sc2), where unexpected
losses due to PLI can occur, we observe that in lightly
loaded scenarios PLI-induced losses become significant, and
as a result, there exists a strong degradation in terms of the
overall PLP for the STATIC approach. The CLONE network,
in contrast, is aware of such dynamic, time-varying PLIs,
and is able to take decisions on-the-fly to either re-route or
simply drop packets flows and, up to some extent, improve
network performance. Note that this occurs within the typical
operation range of sub-wavelength switching networks, that
is, for overall PLP values of 10−3and lower. Besides, the
optimized management of regenerator resources is maintained
as in Fig. 4(a).
Finally, in Figures 5(a) and 5(b), we show the amount of
regenerators that can be turned off (i.e., that are not used)
during network operation under both the STATIC and CLONE
approaches. Due to the reduction of the traffic load sent to
regenerator pools, the CLONE network can provide substantial
energy savings as allows for a notable number of regeneration
devices to be turned off during network operation.
V. CONCLUSIONS AND FUTURE WORK
We show that a network of CLONEs can achieve both
greater energy efficiency and dynamic adaptation to time-
varying PLIs. We propose real-time impairment aware routing
to minimize regenerator usage and to improve PLP due to the
impact of time-varying PLIs. The performance of the CLONE
network is compared against the STATIC approach which
relies on offline estimations of the PLI impact. Our next goal
is to extend the CLONE network model to a test-bed and
experimentally validate the simulation results.
ACK NOW LE DG ME NT
This work was supported in part by the NSF Engineering
Research Center for Integrated Access Networks (CIAN)
(sub-award Y503160), the Spanish Ministry of Science and
Innovation under both the FPU program and the ”DOMINO”
project (Ref. TEC2010-18522), and the Catalan Government
under the contract SGR-1140.
REF ER EN CE S
[1] S. J. Ben Yoo, “Optical Packet and Burst Switching Technologies for
the Future Photonic Internet, ” IEEE/OSA J. Lightwave Technol., vol.
24, no. 12, pp. 4468-92, 2006.
[2] J. Gripp, J. E. Simsarian, J. D. LeGrange, P. G. Bernasconi, and D. T.
Neilson, “Architectures, Components, and Subsystems for Future Optical
Packet Switches,” IEEE J. Sel. Top. Quant., vol. 16, no. 5, pp. 1394-404,
2010.
[3] G. S. Zervas et. al., “Multi-Granular Optical Cross-Connect: Design,
Analysis and Demonstration,” IEEE/OSA J. Opt. Commun. Netw., vol.
1, no. 1, pp. 69-84, 2009.
[4] C. P. Lai, A. Shacham, K. Bergman, “Demonstration of Asynchronous
Operation of a Multiwavelength Optical Packet-Switched Fabric,”
IEEE/OSA Photonic. Tech. L., vol. 22, no. 16, pp. 1223-25, Aug. 2010.
[5] C. V. Saradhi and S. Subramaniam, “Physical layer impairment aware
routing (PLIAR) in WDM optical networks: issues and challenges,”
IEEE Commun. Surveys and Tutorials, vol. 11, no. 4, pp. 109-30, 2009.
78 9 10 11 12 13
Offered Erlangs per Node
10-7
10-6
10-5
10-4
10-3
10-2
10-1
Packet Loss Probability (PLP)
Total (CLONE)
Total (STATIC)
OSNR (CLONE)
OSNR (STATIC)
Contention (CLONE)
Contention (STATIC)
8 10 12 50
55
60
65
70
Avg. regenerations per-packet reduction (%)
Regeneration_ppr (CLONE)
regenerator usage
Optimizing
(a)
6 8 10 12
Offered Erlangs per Node
10-6
10-5
10-4
10-3
10-2
10-1
Packet Loss Probability (PLP)
Total (CLONE)
Total (STATIC)
OSNR (CLONE)
OSNR (STATIC)
Contention (CLONE)
Contention (STATIC)
50
55
60
65
70
Avg. regenerations per-packet reduction (%)
Regeneration_ppr (CLONE)
PLP degradation
(b)
Fig. 4. Packet loss probability (left y axis) and average regenerations per-
packet reduction (ppr) (right y axis) as a function of the offered load in the
Basic topology for both the CLONE and STATIC networks under a) S c1, and
b) Sc2.
[6] G. Shen and R. S. Tucker, “Translucent optical networks: the way
forward,” IEEE Commun. Mag., vol. 45, no. 2, pp. 48-54, Feb. 2007.
[7] B. Garcia-Manrubia, P. Pavon-Mari˜
no, R. Aparicio-Pardo, M.
Klinkowski and D. Careglio, “Offline impairment-aware RWA
and regenerator placement in translucent optical networks,” IEEE/OSA
J. Lightwave Technol., vol. 29, no. 3, pp. 265-277, Feb. 2011.
[8] O. Pedrola, D. Careglio, M. Klinkowski and J. Sol´
e-Pareta, “Offline
routing and regenerator placement and dimensioning for translucent OBS
networks, ” IEEE/OSA J. Opt. Commun. Netw., vol. 3, no. 9, pp. 651-
666, Sep. 2011.
[9] F. Leplingard, T. Zami, A. Morea, N. Brogard, D. Bayart, “Determina-
tion of the impact of a quality of transmission estimator margin on the
dimensioning of an optical network,” in Proc. IEEE/OSA OFC/NFOEC
2008, Paper OWA6, Mar. 2008.
[10] D. C. Kilper et. al., “Optical performance monitoring,” IEEE/OSA J.
Lightwave Technol., vol. 22, no. 1, pp. 293-304, Jan. 2004.
[11] C. P. Lai et. al., “Experimental demonstration of packet-rate 10-Gb/s
OOK OSNR monitoring for QoS-aware cross-layer packet protection,”
OSA Opt. Express, vol. 19, no. 16, pp. 14871-82, Aug. 2011.
[12] O. Pedrola, D. Careglio, M. Klinkowski and J. Sol´
e-Pareta, “Regener-
ator placement strategies for translucent OBS networks,” IEEE/OSA J.
Lightwave Technol., vol. 29, no. 22, pp. 3408-20, Nov. 2011.
[13] B. Lavigne et. al., “Method for the determination of a quality-of-
transmission estimator along the lightpaths of partially transparent
networks,” in Proc. ECOC 2007, vol.3, pp. 287-8, 2007.
[14] R. Mart´
ınez, R. Casellas, R. Mu˜
noz, and T. Tsuritani, “Experimen-
tal translucent-oriented routing for dynamic lightpath provisioning in
GMPLS-enabled wavelength switched optical networks,” IEEE/OSA J.
Lightw. Technol., vol. 28, no. 8, pp. 1241-55, Apr. 2010.
0
10
20
30
40
8 10 12 14
CLONE
78 9 10 11 12 13 14
Offered Erlangs per node
0
10
20
30
40
Avg. number of regenerators off (%)
STATIC
(a)
6 8 10 12
0
10
20
30
40
50
10
CLONE
678 9 10 11 12
Offered Erlangs per node
0
10
20
30
40
50
Avg. number of regenerators off (%)
STATIC
(b)
Fig. 5. Percentage of regeneration devices turned off during network
operation as a function of the offered load in the Basic topology for both
the CLONE and STATIC networks under a) S c1, and b) Sc2.
[15] A. Morea et al., “QoT function and A* routing: an optimized combina-
tion for connection search in translucent networks,” IEEE/OSA J. Opt.
Netw., vol. 7, no. 1, pp. 42-61, Jan. 2008.
[16] G. P. Agrawal, “Fiber-optic communications systems,”3rd ed. New York,
John Wiley & Sons, pp.490-7, 2002.
[17] H. Song and M. Brandt-Pearce, “Range of influence of physical im-
pairments in wavelength-division multiplexed systems,” in Proc. IEEE
Globecom 2011, Dec. 2011.
[18] F. Leplingard, A. Morea, T. Zami, and N. Brogard, “Interest of an
adaptive margin for the quality of transmission estimation for lightpath
establishment,” in Proc. IEEE/OSA OFC/NFOEC 2009, Paper OWI6,
Mar. 2009.
[19] D. Geisler et. al., “Experimental demonstration of flexible bandwidth
networking with real-time impairment awareness,” OSA Opt. Express,
vol. 19, no. 26, pp. B736-45, Dec. 2011.
[20] L. Liu et. al., “Field and lab trials of PCE-based ONSR-aware dynamic
restoration in multi-domain GMPLS-enabled translucent WSON,” OSA
Opt. Express, vol. 19, no. 27, pp. 2656877, Dec. 2011.
[21] CIAN (Center for Integrated Access Networks): http://cian-erc.org
[22] C. P. Lai et. al., “Demonstration of QoS-Aware Packet Protection via
Cross-Layer OSNR Signaling, ” in Proc. IEEE/OSA OFC/NFOEC 2010,
Paper OTuM2, Mar. 2010.
[23] H. Takeshita, K. Fukuchi, S. Shioiri, and E. L. T. de Gabory, “In-service
Monitoring of PMD Induced Optical Signal Degradation Using SOP
Vector Trajectory on the Poincare Sphere for High-Speed Reconfigurable
Optical Networks,” in Proc. ECOC 2009, Paper 9.3.1, Sep. 2009.
[24] Z. Rosberg, Hai Le Vu, M. Zukerman, J. White, “Performance analyses
of optical burst-switching networks,” IEEE J. Sel. Area. Comm., vol. 21,
no. 7, pp. 1187-1197, Sep. 2003.
[25] O. Pedrola et. al., “JAVOBS: a flexible simulator for OBS network
architectures,” J. Netw., vol. 5, no. 2, pp. 256-264, Feb. 2010.
