Optimization Techniques for Incremental Planning of Multilayer Elastic Optical Networks

Abstract

As optical networks become increasingly flexible and software driven, network operators need to reconsider their present mode of network planning and operation, which traditionally relies on long planning periods, performed independently for the IP edges (logical topology) and the optical transport layer (physical topology). Network planning assuming fully loaded end-of-life conditions fails to follow traffic evolution and results in capacity overprovisioning, underutilized equipment, and stranded investments. We argue that it would be beneficial to have shorter upgrade cycles and a multiperiod network planning approach that accounts jointly for the upgrade of the optical but also the IP edges of the network. We formulate the incremental multilayer planning problem of an IP over elastic optical network and propose an integer linear programming (ILP) algorithm to solve it. The ILP model leverages the reconfigurability of both network layers to delay equipment deployment and benefit from cost erosion. Our objective is, through repurposing of existing network resources, to deploy in a period the minimum additional network equipment (capital expenditures) to cope with traffic changes from the previous period but also to minimize changes in transitioning between the two periods (operational expenditures). The proposed planning approaches are validated through simulations based on realistic network scenarios, where we also study the effect of the upgrade period duration.

Full text

309131
1
1
Abstract— As optical networks become increasingly
flexible and software driven, network operators need
to reconsider their present mode of network planning
and operation, which traditionally relies on long
planning periods, performed independently for the IP
edges (logical topology) and the optical transport
layer (physical topology). Network planning assuming
fully loaded, end-of-life conditions, fails to follow
traffic evolution and results in capacity
overprovisioning, underutilized equipment and
stranded investments. We argue that it would be
beneficial to have shorter upgrade cycles and a
multi-period network planning approach that
accounts jointly for the upgrade of the optical but also
the IP edges of the network. We formulate the
incremental multi-layer planning problem of an IP
over elastic optical network (EON), and propose an
integer linear programming (ILP) algorithm to solve
it. The ILP model leverages the reconfigurability of
both network layers to delay equipment deployment
and benefit from cost erosion. Our objective is,
through repurposing of existing network resources,
to deploy in a period the minimum additional
network equipment (CapEx) to cope with traffic
changes from the previous period, but also to
minimize changes in transitioning between the two
periods (OpEx). The proposed planning approaches
are validated through simulations based on realistic
network scenarios, where we also study the effect of
the upgrade period duration.
Index Terms—Elastic optical networks; incremental
capacity planning; joint multi-layer planning.
I. INTRODUCTION
he continuous growth of IP traffic and the emergence of
new services are leading to a huge increase of traffic
volume [1]. Future 5G networks will engender a wide range
of new services such as ultrahigh-definition video streaming,
Manuscript received October 12, 2017; revised December 23, 2017
(Doc. ID 309131).
P. Papanikolaou, K. Christodoulopoulos, and E. Varvarigos are
with the Electrical and Computer Engineering Department,
National Technical University of Athens, Athens, Greece.
(e-mail: ppapanikola@mail.ntua.gr)
augmented and virtual reality, cloud gaming, smart homes,
etc. The considerable challenge faced by telecom operators is
to cater for higher capacity and efficiency but also for high
unpredictability and dynamicity. This requires an agile
network infrastructure, spanning from the access towards
the metro/regional and core segments of the network [2].
Optical transport networks used today in metro/regional
and core networks are designed and operated in a static
manner. The optical transport network is planned assuming
long upgrade periods. In order to ensure that the resulting
network will cope with increasing traffic until the next
upgrade, future needs are forecasted and capacity is
overprovisioned. The extra capacity allocated results in
underutilized equipment and unnecessary investments for
long periods of the network lifecycle. The longer the upgrade
periods, the more the overprovisioning, and the higher the
unnecessary investment paid upfront. Moreover, such
approach fails to capture traffic evolution and technology
maturation (equipment cost decreasing or new more
advanced equipment becoming available). Finally, another
factor that contributes to overprovisioning is that the
different segments of the network are upgraded
independently. To make things worse, overprovisioning is
performed not only at the network capacity level but also at
the physical layer. The current practice is to establish
lightpaths (optical connections) and estimate their Quality of
Transmission (QoT) so that it remains acceptable until their
End-of-Life (EOL). Since the QoT deteriorates with time due
to equipment ageing, increased interference from new
connections, etc, QoT estimation is done with high margins
that are too pessimistics in the early years [3].
The advent of Elastic Optical Networks (EONs) combined
with optical transport platforms that facilitate the setting up
and tearing down of lightpaths within minutes or even
seconds [4] create all the necessary conditions to achieve a
truly programmable and flexible networking environment.
Moving towards this direction, EONs can exploit the
bandwidth variable transponders (BVTs) to reconfigure the
lightpaths to meet dynamic traffic requirements [5].
Combining EONs with the IP layer reconfigurability can
facilitate a pay-as-you-grow approach, where little
equipment is installed and continuously re-optimized and
upgraded. However, this has to be done in a coordinated
manner for both the IP and optical segments [6]. Thus, a
multi-period multi-layer network planning approach with
short periods is required. This would closely capture the
Optimization Techniques for
Incremental Planning of Multilayer
Elastic Optical Networks
P. Papanikolaou, K. Christodoulopoulos, and E. Varvarigos
T

309131
2
traffic evolution and avoid overprovisioning by incorporating
smaller but more frequent network updates. The required
upgrades could be limited by exploiting the optical and IP
(multi) layer reconfigurability and would also benefit from
technology maturation that would be missed in longer
upgrade periods.
Towards this end, we propose an integer linear
programming (ILP) model that jointly considers multi-layer
and incremental multi-period planning. The challenge is to
capture the effects of time dependent planning parameters,
such as traffic evolution, technology maturation and
equipment cost decrease. These are not known in the long
term and the longer we predict them the more uncertainty
we would introduce. So in our model we assume a given
period duration and we optimize the deployment of the
additional network equipment and the changes imposed by
the transition between the current and the next state.
Our results indicate that combining multi-layer and
incremental network planning significantly reduces total
cost of ownership. Cost-efficient network solutions are
obtained by exploiting the reconfigurability capabilities of
flexible equipment so as to adapt to the time evolution of
planning parameters. We also verify that additional savings
are achieved by adopting shorter network upgrade periods.
The rest of the paper is organized as follows. Section II
presents the related work, while in Section III we formally
state the incremental planning of a multi-layer IP over EON
problem. Section IV describes the joint incremental and
multi-layer planning techniques, while the ILP formulation
for solving the optimization problem follows in Section V.
Performance results are presented in Section VI. Our
conclusions follow in Section VII.
II. RELATED WORK
Significant research has focused on multi-layer network
optimization [7]-[9], with multi-period network planning
[10]-[14] also receiving recent attention. Regarding IP over
(elastic or fixed) optical networks, the authors in [7]
highlight the role played and the significant cost savings
achieved by a design process that optimizes the base IP
topology introducing router bypass. In order to reduce the
aggregation level of incoming flows, the authors in [8] exploit
EON technology’s finer granularity to allow grooming at the
optical layer. To this end, they propose a new architecture for
national IP/multiprotocol layer switched (MPLS) networks
interconnected through an EON core. The authors in [9]
examine the planning problem of a multi-layer IP over EON
from the perspective of capital expenditure minimization by
accounting for modular IP/MPLS routers at the optical
network edges along with BVTs.
Multi-period planning aims to optimize the cost of
transport networks over a long time frame. There are two
approaches for multi-period planning, (i) global optimization
assuming knowledge of the traffic and equipment
characteristics/prices for all periods [10]-[11], and (ii)
incremental planning [12]-[14]. Authors in [10] incorporate
multi-layer and multi-period planning in a single
optimization step. In their attempt to study the migration
scenario from a networking point of view, authors in [11]
propose a single-layer ILP model. Multi-period planning is
used to study the migration of the network from single- to
mixed-line-rate and investigate the deployment of an
optimal channel mix based on reach and equipment prices.
To quantify the degree of traffic dynamicity and growth
that would justify higher initial investment in (flex-rate)
BVT technology, the authors in [12] propose an ILP model for
multi-period analysis that accounts for hardware
provisioning requirements over multiple periods of
increasing traffic. To achieve savings over the current
provisioning practice of using End-of-Life physical layer
margins, the authors in [13] present an algorithm to
provision lightpaths based on actual physical performance
and use it in a multi-period planning scenario for Just in
Time equipment deployment. In a similar concept, [14]
models the progressive ageing of the transmission channel
and quantifies the benefits of dynamically adjusting the BVT
to the physical network quality.
In this paper, we take an incremental planning approach
for the joint planning of a multi-layer IP over EON network.
The incremental approach is motivated by the increased
traffic dynamicity and unpredictability resulting from the
advent of new services and 5G technology. Under such
conditions, it seems hard to have a priori knowledge of the
exact traffic volume and pattern for the entire network
lifecycle, while it is possible to have rather good forecasts of
short-term traffic growth. Our objective is to deploy at each
period the minimum amount of additional network resources
required to cope with traffic changes from the previous
period, optimizing both the capital expenditure (CapEx) of
the equipment used and the operational expenditure (OpEx)
associated with the changes imposed by the transition
between the two periods.
Even though multi-layer planning and incremental
multi-period planning have been extensively researched, to
the best of our knowledge no other work apart form [15] gives
a formal description and optimal solution to the combination
of these planning approaches. In [15] we provided the formal
description and optimal solution to the problem only
considering optical layer reconfigurability. In this paper we
extend [15] and propose techniques that exploit both optical
layer reconfigurability and IP layer grooming capabilities,
examining the impact of each layer in the incremental
planning process. Additionally, we consider non-uniform
traffic evolution scenarios (to account for the impact of traffic
dynamicity) and we also study the impact of the upgrade
period duration.
In summary, the main novelties of this work are the
following: Firstly, we formulate and provide an optimal
algorithm to solve the aforementioned problem, leveraging
the reconfigurabilty of network equipment at both layers to
avoid capacity overprovisioning and improve the cost
efficiency of the network. Secondly, the proposed model
introduces a penalty on the reconfiguration of existing
connections at both layers (lightpaths and IP/MPLS label
switched paths (IP-LSPs)), to restrict the extent of
modifications performed between periods and associated
costs and disruptions. In this way we obtain a trade-off
between the equipment added (CapEx) and the changes
performed (OpEx) between successive periods (which might
require manual intervention or service disruption). Thirdly,

309131
3
we use the proposed algorithm to examine the effect of the
upgrade period durations and we verify that small and
frequent upgrades can yield significant savings, which can
be further increased by considering technology maturation
and cost erosion of network equipment.
III. PROBLEM STATEMENT
A. Network Architecture
We assume an EON domain comprised of optical switches
and fiber links. The fiber links consist of single-mode fiber
(SMF) spans and erbium-doped fiber amplifiers (EDFAs).
The optical switches function as Reconfigurable Optical Add
Drop Multiplexers (ROADMs) employing flex-grid
technology, and support lightpaths of one or more contiguous
12.5 GHz spectrum slots. Note that the solutions to be
proposed will also be valid for fixed-grid WDM networks (50
GHz wavelengths), which can be viewed as a special and
simpler case of EONs. At each optical switch, zero, one or
more IP/MPLS routers are connected, comprising the edges
of the optical domain. An IP/MPLS router is connected to the
ROADM via BVT transponders that transform the client
signal for optical long-haul transmission. We also assume
that the optical nodes can be equipped with bandwidth
variable regenerators (BVRs) of similar specifications with
BVTs that can be used to regenerate the optical signal.
The source BVT transponder, functioning as a
transmitter, converts the electrical packets coming from the
IP source router to optical signals (E/O conversion). Then the
traffic entering the ROADM is routed over the optical
network along the established lightpaths. We assume that a
number of transmission parameters of the BVTs and the
BVRs are under our control, affecting the rate and reach at
which they transmit. The lightpath passes transparently or
translucently (if BVRs are required to restore signal quality)
through intermediate ROADM and reaches the destination
ROADM where it is dropped. The signal is converted back to
electrical at the destination BVT that operates as the optical
receiver (O/E conversion), and the packets are forwarded to
the corresponding IP/MPLS router. This forms a virtual or IP
link between the lightpath source and destination IP/MPLS
routers. Note that lightpaths are assumed to be bidirectional
and thus in the above description an opposite directed
lightpath is also installed, and transponders act
simultaneously as transmitters and receivers. An IP/MPLS
router that is reached can be the final destination or an
intermediate hop in this domain. If it is the final destination
in the IP domain, the packets are forwarded to the next
domain. If it is an intermediate hop on the virtual topology,
the packets are routed back to the optical network, over a
new lightpath and traverse more intermediate IP/MPLS
router hops to reach the domain destination. The IP links
that comprise the IP/MPLS path in the domain are called the
virtual (IP) path or the IP-label switched path (IP-LSP).
From the optimization point of view, the network consists
of two layers, the IP (or virtual or logical) layer and the
optical (or physical) layer. The optical lightpaths are
installed on the physical topology to create the virtual
topology, on top of which the IP-LSPs are installed. A virtual
link is implemented as a lightpath (or a sequence of
lightpaths). Lightpaths are assigned routes, modulation
format and spectrum (RMFS) on the physical topology, while
IP/MPLS packets are routed over the IP-LSPs, i.e, paths of
the virtual topology.
B. Incremental Multi-layer Planning
Due to the rather static and inflexible nature of current
optical transport networks, the planning process uses long
planning periods. Aiming to avoid capacity overprovisioning
and unnecessary investments that affect cost efficiency we
introduce a periodic re-optimization process that can
facilitate a pay-as-you-grow approach. Through periodic
re-optimization of the network, operators are able to detect
early signs of QoT degradation, equipment ageing, and
capacity exhaustion. Note that the periods’ length
determines how closely traffic evolution and technology
maturation will be captured.
We implement the concept of periodic re-optimization by
adopting an incremental planning approach. We assume
that the upgrade process of the multi-layer network is
performed periodically and takes decisions on how to support
the traffic for the next planning period, given the current
state of the network and the equipment availability and
prices. So, the assumption is that this process is performed
successively and separately for each period, having the
knowledge (forecast) of the traffic only of the next period and
no further future knowledge.
Fig. 1. Incremental multi-layer planning process.
In the initial planning period (Period t0) both (IP and
optical) layers are simultaneously optimized with the
objective being the minimization of the cost. Algorithms such
as those in [16] can be used for this step. At the start of a new
period tN, the incremental model takes as input the new
traffic, the current equipment availability and prices, the
previous state of the network at tN-1, including the state of
the resources (established lightpaths and IP-LSPs), and
information about physical resources (installed/available
equipment and its location). The optimization process
considers jointly the IP and physical layers and the previous
network state and aims at minimizing both the added
network equipment (CapEx) and the equipment
displacements and reconfiguration between the two
successive network states (OpEx).
As shown in Fig. 1 the proposed model exploits the
flexibility of BVTs that can be used in numerous different
configurations to carry client traffic. This allows an initial
design that is scalable through the years, since it is possible
to increase client’s port rate by increasing, when feasible, the

309131
4
number of optical carriers, or by using higher order
modulation formats with the possible addition of BVRs
(since higher order modulation formats entail a decrease in
the optical reach) and the possible displacement of the
already installed ones. Additionally, network resources can
be made available by re-optimization of the previous
network state exploiting the grooming/reconfiguration
capabilities of both layers to enable spare capacity
utilization. Our model takes into account the traffic
dynamics, cost evolution and technology development to
achieve greater cost savings.
IV. INCREMENTAL MULTI-LAYER PLANNING
OPTIMIZATION TECHNIQUES
We assume, in our study, that accurate QoT and reach
estimation is available at the start of each period.
Mechanisms for this purpose are developed in the
ORCHESTRA project, where physical layer information
obtained from software Optical Performance Monitors is
processed using data analytics/correlation methods to yield
accurate physical layer knowledge [17]. This can be used in
dynamic use cases to re-optimize the network following the
observe-decide-act control loop, or for planning purposes, as
done here.
Three techniques, based on the reconfigurability of IP and
optical equipment, are useful in leveraging multi-period and
multi-layer planning simultaneously: The first technique
uses IP grooming capabilities to minimize the cost of added
equipment at both network layers. The second technique
exploits optical layer reconfigurability to delay equipment
deployment and benefit from cost erosion to minimize the
network cost over a long period of time. The third technique
considers in a single optimization step the joint
minimization of (i) IP and Optical network layers equipment
cost and (ii) cost of the changes (e.g. IP rerouting, lightpath
tear-down, BVTs reconfiguration, setup of new lightpaths,
etc) required for the transition between two periods.
A. Virtual layer re-optimization (VTR)
The incremental planning technique based on virtual
topology re-optimization (VTR) focuses on the grooming
capabilities of the IP layer, in order to exploit the flexibility
the IP layer provides in a multi-period planning scenario.
Through IP rerouting and re-grooming, VTR tries to use
efficiently the spare capacity of the lightpaths and IP-LSPs
established in the previous period. When the spare capacity
of existing IP-LSPs is inadequate to serve the current state
demands, new equipment is added.
Fig. 2 presents an illustrative example of the examined
multi-layer network, where four IP/MPLS routers comprise
the IP (virtual) layer, and four flex-grid optical switches
together with BVTs supporting different transmission tuples
comprise the optical (physical) layer. More specifically, in
Fig. 2(a), three lightpaths (A↔B, B↔D, and C↔D) have been
established at the optical layer, and three IP-LSPs (from
node RA to node RB, from node RB to node RD and from node
RC to node RD) have been set up at the IP layer. In Fig. 2(b)
the transition between two network states corresponding to
periods tN-1 and tN is depicted. The capacity of some IP-LSPs
may be inadequate due to high congestion on some virtual
topology links, in which case a virtual topology
reconfiguration has to be performed to groom and balance
the traffic avoiding the heavily utilized resources. In the
example presented in Fig. 2(b), upon transition between
state tN-1 to tN, the remaining capacity of IP-LSP RB-RD (200
Gb/s) is unable to serve the demands dA-D (140 Gb/s) and dB-D
(90 Gb/s). The virtual topology can be re-optimized by
establishing a new IP-LSP (RA-RC) that allows grooming part
of dA-D (30 Gb/s) and dC-D (70 Gb/s) demands (dotted red line
in Fig. 2(b)). The objective of VTR is to exploit the IP
grooming capabilities in order to minimize the cost of added
equipment at both layers of the network, which in this
example was to add 100 GB/s lightpath A↔C and the
corresponding RA-RC IP-LSP.
a)
b)
Fig. 2. Example of IP layer reconfiguration in an incremental
planning scenario: (a) previous network state (tN-1) and (b) current
network state (tN).
B. Optical layer re-optimization (OLR)
The incremental planning technique based on optical
layer re-optimization (OLR) exploits the reconfigurability of
optical equipment in order to delay new equipment
deployment and decrease the total cost of the network over
multiple periods. In this technique, the IP grooming
capabilities are limited, since our goal is to understand the
impact of the optical layer flexibility in the evolution of the
network lifecycle.
The advantages of this technique stem from the use of
BVTs and flex-grid ROADMs. BVTs can adjust their
bandwidth, by changing the modulation format, baudrate
and number of carriers (according to the particular BVT’s
specifications). This is typically combined with a spectrum
reconfiguration, making use of the flex-grid ROADMs.
Through the use of BVTs, we avoid the future purchase of
many different transponders—one can deliver a wide range
of required capacities, as depicted in the example of Fig. 3. A
somehow related benefit of using BVTs is their ability to
trade-off reach for capacity. Reallocating transponders to
connections and making use of this trade-off we can
accommodate abrupt traffic changes and postpone or avoid
equipment investments. The main benefit of investment
delay is that technology maturation usually leads to
reductions in equipment price.

309131
5
As an illustration of the OLR technique, consider Fig. 3,
where three lightpaths (B↔A, A↔C, and C↔D) and three
IP-LSPs (from node RA to node RB, from node RB to node RD,
and from node RC to node RD) have been established at the
optical and the IP layer, respectively. In Fig. 3(a) a 100 Gb/s
demand (dB-C) is allocated a lightpath of 100 Gbps
polarization division multiplexed quadrature phase shift
keying (PDM-QPSK). This leaves a large capacity margin,
since 200 Gbps could be transmitted over the same path
using PDM-16 quadrature amplitude modulation (QAM).
Assuming that the deployed transponder is rate-flexible
(BVT) and 200 Gbps-capable, when the demand (dB-C) grows
(in period tN) to 160 Gb/s, the extra capacity can be allocated
without replacing the transponder, by simply changing its
modulation format from PDM-QPSK to PDM-16QAM (Fig.
3(b)). Demands dA-D and dC-D are served following a similar
process for the other two lightpaths (A↔C, and C↔D). Note
that OLR is unable to reroute the already established
IP-LSPs. Thus, it exploits only BVT reconfigurability to
serve growing demands, while when the capacity margins of
the BVTs are depleted, new equipment is deployed.
a)
b)
Fig. 3. Example of optical layer re-optimization in an incremental
planning scenario: (a) previous network state (tN-1) and (b) current
network state (tN).
C. Joint multi-layer re-optimization
The incremental planning technique based on joint
multi-layer re-optimization (JMR) fully exploits the
reconfigurability of the optical network equipment and the
grooming capabilities of the IP layer. In this technique we
jointly consider both the capital expenditure (CapEx) of the
equipment used in both layers of the network and the
operational expenditure (OpEx) associated with the changes
(e.g. IP rerouting, lightpath tear-down, BVTs reconfiguration,
setup of new lightpaths, etc.) imposed by the transition
between two network periods.
Using JMR the optical and IP layer of the network are
utilized in a coordinated manner in order to increase its
capacity, extend its life, decrease deployment cost, and
minimize the required manual interventions.
An illustration of JMR is presented in Fig. 4, where in Fig.
4a, three lightpaths (A↔C, B↔C, and C↔D) and three
IP-LSPs (from node RA to node RB, from node RB to node RC,
and from node RC to node RD) have been established at the
optical and the IP layer, respectively. In this example we
demonstrate that choices made for one connection in the
early planning periods, e.g. to serve it over a specific path by
placing regenerators at specific nodes so as to avoid
congestion over another path, could be changed in
subsequent periods, when the chosen path becomes
congested while the avoided path turns to be relatively
empty. More specifically, the optical connection B↔C, which
is already established at network period tN-1, requires
intermediate BVR (node E). The growth of the traffic
demands and the addition of a new demand (dB-D) [in period
tN] triggers the JMR technique, which combines traffic
grooming and the reconfigurability of the BVTs to reroute
lightpath B↔C and eliminate regeneration. The
establishment of four low rate optical connections (A↔B,
B↔A, B↔D and D↔C) facilitates the rerouting of the
increased demands in order to remove bottlenecks and avoid
network congestion. Fig. 4(b) presents the re-grooming of
demands dA- D and dB-C, where part of the demand dA-D (50
Gb/s) and demand dB-C (150 Gb/s) are groomed and allocated
to a 200 Gb/s BVT (lightpath A↔C). In node C, grooming
demand dC-D (100 Gb/s) and part of demand dA-D (50 Gb/s)
leads to a 150 Gb/s BVT (lightpath C↔D). The rest of the dA-D
demand (80 Gb/s) and dB-C demand (70 Gb/s) and the demand
dB-D (50 Gb/s) are groomed and allocated to lightpaths A↔B,
B↔D and D↔C, with a capacity of 100 Gb/s, 200 Gb/s and
100 Gb/s respectively.
a)
b)
Fig. 4. Example of joint multi-layer re-optimization in an
incremental planning scenario: (a) previous network state (tN-1) and
(b) current network state (tN).
V. MATHEMATICAL FORMULATION
In the ILP model to be presented in this section,
multi-layer and incremental planning are jointly considered
in a single optimization step. For each period, both network
layers are simultaneously optimized by taking into account
not only equipment cost (CapEx) but also the cost of changes
from the previous network state (OpEx). The extend to which
the current state will commit to the previous one
(equivalently, the trade-off between CapEx and Opex for the

309131
6
Optical and the IP layer) is controlled through parameters
Wo and Wf, passed as input to the model.
The network is represented by a graph G(V,L), with V
being the set of nodes and L the set of bidirectional fiber
links connecting nodes. The nodes of the graph correspond to
the optical & IP nodes of the network where we account for
the cost of both layers. We assume that we are given the
traffic matrix Λ, where Λsd corresponds to IP demanded
capacity between source-destination pair (s,d).
We are also given the models of the BVT transponders and
BVR regenenerators. We represent by B the set of BVT, and
we also assume that for each BVT bB there is an equivalent
BVR, represented by rb. The transmission option of a BVT or
BVR are described in what are called transmission tuples.
Each transmission tuple t represents a specific configuration
of the BVT (rate, spectrum) and is related to a specific
transmission reach, using a QoT estimation model (e.g. [18]).
To be more specific, transmission tuple t = (Dt,Rt,St)
represents a feasible transmission at distance Dt, with rate
Rt (Gpbs), using St spectrum slots. The network designs are
based on pre-calculated paths. In particular, we assume that
for each pair of nodes (i,j) we precalculate k-paths from i to j,
which define the set Pij. The algorithm decides to serve
connections using a specific BVT (and if needed BVR), a
tranmission tuple, and a path. This is represented by a
feasible path-transmission tuple pair (p,t), where the BVT
transponder is represented by its related transmission tuple.
If the length of path p is higher than the reach of tuple t, we
assume that we place BVR (with the same configuration t as
the BVT) over the path at the node before QoT becomes
unacceptable. The term feasible is used for this reason, to
indicate that QoT is accounted for. We denote by Cp,t the cost
of the path-transmission tuple pair (p,t) which includes the
cost of the BVT and BVR (if required).
Finally, we are also given a model for an IP/MPLS router.
We assume that a router is a modular device, built out of
(single or multi) chassis. A chassis provides a specified
number of bi-directional slots with a nominal transmission
speed. Into each router slot, a linecard of the corresponding
speed can be installed. Each linecard provides a specified
number of client ports at a specified speed and occupies one
slot of the IP/MPLS router. A client port is connected with an
equivalent BVT and we assume that for every BVT there is
an available tunable linecard type. The scalable
multi-chassis core router has up to NLCC line card chassis
and there are NFCC router slot capability per chassis.
A problem instance is described by the following inputs:
 the network topology, represented by graph G(V,L);
 the maximum number Z of available spectrum slots (of
12.5 GHz);
 the traffic, described by the traffic matrix Λ;
 The sets of paths Pij for all pairs of nodes (i,j)
 the sets B and R of available transponders (BVT) and
regenerators (BVR).
 the set T of availiable transmission tuples for all
transponders. The set Tb respresent transmission tuples of
transponders bB.
 the set of feasible path-transmission tuple pairs (p,t) and
their cost Cpt, which includes the cost of the BVT
transponder and BVR regenerator(s) (if required).
 the set of line-cards represented by H, where the
line-cards for transponder bB are represented by the set
Hb. A linecard hHb is represented by tuple h=(Nh,Ch),
where Nh is the number of transponders of type b that the
line-card supports and Ch is the cost of the line-card;
 the IP/MPLS router cost model. We assume that an
IP/MPLS router consists of line-card chassis of cost CLCC,
that suport NLCC line-cards each, and fabric card chassis of
cost CFCC, that suport NFCC line-card chassis;
 the weighting coefficient, WC, taking values between 0 and
1. Setting WC = 1 minimizes solely the CapEx whereas
setting WC ≈ 0 minimizes the maximum spectrum used;
 weighting coefficients Wo and Wf, taking values between 0
and 1. Setting Wo = Wf = 1 minimizes solely the current
state cost ignoring the previous network state, whereas
setting Wo ≈ 0 maintains the previous state’s lightpaths
(optical layer equipment), while setting Wf ≈ 0 maintains
the previous state’s IP-LSPs (IP layer equiment). Thus, Wo
(or Wf) controls the tradeoff between CapEx and Opex for
the Optical layer (or the IP layer, respectivey).
Variables:
psd
f
Float variable, equal to the capacity of the IP-LSP
from IP source s to destination d that passes over a
lightpath (virtual link) that uses path p.
xpt
Integer variable, equal to the number of lightpaths
with path-transmission tuple pair (p,t) used.
ynh
Integer variable, equal to the number of line-cards
of type h at node n.
θnb
Integer variable, equal to the number of used
transponders of type b at node n.
vnb
Integer variable, equal to the number of deployed
transponders of type b at node n.
qn
Integer variable, equal to the number of line-card
chassis at node n.
on
Integer variable, equal to the number of
fabric-card chassis at node n.
O
pt
d
Integer variable, equal to the number of torn down
lightpaths from the previous state, counted as the
number of removed (p,t) path – transmission
tuples pairs.
F
psd
d
Boolean variable, identifying if the IP-LSP from IP
source s to destination d that passes over a
lightpath that uses path p was affected by the
transition between two network states.
z
Integer variable, equal to the maximum indexed
spectrum slot.
zl
Integer variable, equal to the total number of
spectrum slots used in each bidirectional fiber
link.
c
Float variable, equal to the capital expenditure of
the added network equipment.
ω
Integer variable, equal to the number of lightpaths
torn down in the transition between the previous
and the current state.
φ
Integer variable, equal to the number of affected
IP-LSPs in the transition between the previous
and the current state.
Constants:

309131
7
psd
F
Float constant, equal to the IP traffic of
end-nodes s to d that is transferred over optical
path p in the previous network state.
pt
X
Integer constant, equal to the number of
lightpaths of path-transmission tuple pairs (p,t)
used in the previous network state.
nb

Integer constant, equal to the number of
transponders of type b at node n used in the
previous network state.
M
Float constant, a big number that is used to
form big-M constraints (e.g. M>max(Λsd)).
Objective:
   
 
 
1
min (1 ) 1 f
o f C C o W
W W W c W z W

  
        
(1)
 Capital expenditure calculation constraints:

|( , )feasible
.
ij
h nh
pt pt
i V j V p P n V h H
t T p t
LCC n FCC n
n V n V
c C y
Cx
C q C o
    








   
  


(2)
 Operational expenditure calculation constraints:
|( , )feasible .
.



  
    



 
  
ij
ij
O
pt
t T p t
i V j V p P
F
psd
s V d V i V j V p P
d
d
(3)
 IP flow continuity constraints:
 
2
, ,,s d V Vn
,
,.
0, ,
in nj
sd
psd psd sd
i V p P j V p P
ns
f f n d
n s d
   




   






   
(4)
 Path-transmission tuple assignment constraints:
2
( , ) ,i j V
2|( , ) feasible
( ).
  


   
ij ij
psd t pt
p P p P t T p t
sd V
f R x
(5)
 Previous state constraints (optical layer):
feasible( ),p,t
.
O
pt pt pt
d X x
(6)
 Utilized transponders constraints:
, ,n V Bb
|
.

  

ni b
nb pt
i V p P t T
x
(7)
 Deployed transponders constraints:
, ,n V Bb
,.
nb nb nb nb
vv

  
(8)
 Previous state constraints (IP layer):
 
 
22
,,,, , ij
s d V i j V pP 
.
F
psd psd psd
M d F f  
(9)
 Number of line-cards per node constraints:
, , ,
b
n V b B h H   
/ .
nh nb h
h
y v N
(10)
 Number of line-card chassis per node constraints:
,nV
/.
b
n nh LCC
hH
q y N


(11)
 Number of fabric card chassis per node constraints:
,nV
/.
n n FCC
o q N
(12)
 Estimation of maximum spectrum slot used constraints:
2
,( , )l L i j V  
,
| |( , ) feasible
( ).
  


ij
lt pt
p P l p t T p t
z S x
(13)
,lL
.l
zz
(14)
.zZ
(15)
The joint multi-layer planning ILP formulation presented
above dimensions the network for the next period. An IP
demand between s and d is served by a single lightpath
(when s=i and d =j), or by a series of lightpaths that compose
the IP-LSP. The IP-LSP paths are identified by the values of
the IP flow variables fpsd, showing the amount of IP traffic of
end-nodes s to d that is transferred over optical path p.
Variables xpt represent the lightpaths; a lightpath between
source-destination optical nodes i,j are chosen among k
(pre-calculated) optical paths Pij (Eq. (5)). Note that some
transponders that were deployed in a previous period might
remain unused in the current period. To account for this, we
use two types of variables, vnb that corresponds to the
deployed transponders (Eq. (8)), including idle ones,
and θnb that corresponds only to the used ones in the current
period (Eq. (7)). Since linecards and subsequently chassis are
matched and calculated based on deployed
transponders vnb such variables distinction is not required
for that equipment (Eqs. (11)-(13)).
The cost of the IP/MPLS routers is captured through
variables ynh, qn and on. The objective (Eq. (1)) is to minimize
a weighted sum of the maximum spectrum used in the
network, the CapEx of the equipment used in both layers
(Eq. (2)) and the reconfigurations of lightpaths and IP-LSPs
(Eq. (3)). Constraints (4)-(5) and (11)-(13) deal with the joint
multi-layer (optical and IP) planning problem, constraints
(6)-(10) address the incremental planning problem, while
constraints (13)-(15) address the spectrum usage.
In order to reduce the model complexity and obtain
optimal results for realistic network sizes, the ILP does not
perform spectrum assignment. It calculates an estimation of
the slots used per link zl, by summing the spectrum of the
lightpaths that cross the link, thus neglecting the spectrum
continuity constraint (requiring the use of the same
spectrum slots over all links of the lightpath). Based on those
it minimizes the estimation of the maximum spectrum used
in the network z (Eq. (1)), which is constrained to be within
the available spectrum slots range Z (Eq. (15)). The model
can be extended to jointly perform spectrum allocation as
well, but the gains in the objective were observed to be

309131
8
significantly small. For the purposes of simplicity and to
enable running the ILP model for large network instances,
the spectrum assignment was performed, with respect to the
spectrum continuity constraint, in a subsequent step using a
modified Hungarian method [19].
VI. ILLUSTRATIVE RESULTS
In this section we evaluate the performance of the
incremental multi-layer planning techniques presented in
Section IV. In particular we distinguished the following
three scenarios:
 the planning from scratch scenario (denoted as ML),
where the whole network is designed at each period without
taking into account the previous network state. This scenario
provides the optimal benchmark for the comparisons, since
planning the network from scratch without considering the
previous network state obviously leads to the optimum
(lowest) CapEx; it is not, however, a realistic approach since
it maximizes OpEx and disruption [Wo= Wf =1 in Eq. (1)];
 the static incremental scenario (denoted as Inc), where
the network is incrementally planned without being able to
perform any change from the previous network state, thus
fully respecting it. This restriction applies to both layers of
the network, limiting the BVT reconfiguration and IP
grooming capabilities, this scenario provides the pessimistic
benchmark for the comparison [Wo= Wf =0 in Eq. (1)]; and
 the joint multi-layer incremental planning scenario,
where both layers of the network are upgraded in a
coordinated manner. The objective of this scenario is to
optimize both the added equipment (CapEx) at each period
and the number of changes made (OpEx), using the proposed
techniques (Section IV). We examined three scenario
variations, by varying the parameters Wo and Wf that control
the ability to deviate from the previous network state: Wo=0
and Wf=1 (denoted as VTR - IV.A), Wo=1 and Wf=0 (denoted
as OLR- IV.B), and Wo=0.5 and Wf=0.5 (denoted as JMR -
IV.C). When Wo=0 and Wf=1 (VTR), the re-optimization
model is based solely on the grooming capabilities of the IP
layer and is not able to perform any change in the
established lightpaths of the previous network state.
Respectively, when Wo=1 and Wf=0 (OLR) the
re-optimization process exploits only the reconfigurability of
the optical layer equipment. When Wo=0.5 and Wf=0.5 (JMR)
the re-optimization incremental planning technique utilizes
both IP grooming capabilities and BVT flexibility in order to
equally minimize both the CapEx of the added equipment
and the OpEx associated with the transition changes
between the two states.
In our simulations we used two reference network
topologies with different characteristics in terms of number
of nodes, link lengths and load: the Deutsche Telekom (DT -
[20]) and the Telefónica (TID - [20]) topologies, so that the
results obtained are representative of real networks. For
these networks we also used realistic traffic matrices. The
traffic matrices of the DT and TID networks used in our
simulations were based on input by the related operators
reported in IDEALIST project [20] for past years. We
projected the traffic of these networks for 10 years, with a
step of 2 months. In order to emulate the dynamic evolution
of traffic we assume random growth rates for every
demand/entry of the traffic matrix. More specifically, for
both networks, we categorized each demand into three
groups consisting of small, medium and large demands. For
every demand of each group a set of traffic growth
parameters is randomly generated per year. For the DT
topology the yearly scaling factor of the large demands varies
between 1.35-1.4, while for the medium demands varies
between 1.3-1.35 and for the small demands between
1.25-1.3. Respectively for the TID topology the yearly scaling
factors vary between 1.15-1.3. Following [20], the average
yearly increase of the total traffic for the DT network is 1.35,
while for TID it is 1.25.
In Fig. 5 we present the traffic growth for three random
demands of each group, and the uniform traffic growth for
the DT network (1.35 yearly traffic growth). Our main goal
through the aforementioned traffic patterns is to emulate
dynamic traffic evolution and provoke network congestion.
Fig. 5. Illustration of the traffic demands profile of the DT topology
used in our study.
We assume two types of BVTs (and equivalent BVRs), the
first with maximum rate of 400 Gbps and the second of
1Tbps, with the latter being made available after year 2020.
The transmission configurations (tuples) of the BVTs are
presented in TABLE I. We assume that for every BVT there
is an available tunable linecard type. We also consider a
scalable multi-chassis core router, with up to 72 chassis
(NFCC), and a 16 router slot capability per chassis (NLCC). The
cost of BVTs and of the IP/MPLS routers are based on cost
models defined by IDEALIST project [20]. The reference cost
unit (c.u.) is defined as the cost of a 100 Gb/s coherent
transponder. In our CapEx model we also considered the cost
of the fibers assuming that they are rented, with relative cost
0.004 c.u./km/year). In our study we account for equipment
cost erosion over the network lifecycle, assuming a cost
erosion of 10% per year for all types of equipment.
TABLE I
BANDWIDTH VARIABLE TRANSPONDERS
BVT 1
BVT 2
CAPACITY
(GB/S)
REACH
(KM)
DATA
SLOTS
COST
(C.U.)
CAPACITY
(GB/S)
REACH
(KM)
DATA
SLOTS
COST
(C.U.)
100
2000
4
1.76
500
950
7
2*
150
1350
4
600
800
8
200
1050
5
700
700
9
250
950
5
800
650
11
300
700
6
900
550
12
350
600
6
1000
450
14
400
450
6
* AVAILABLE FROM 2020
A. Cost evaluation and spectral impact
In this section we compare the different planning
techniques with respect to the resulting CapEx (Fig. 6) and

309131
9
spectral resource utilization (Fig. 7) for the two reference
networks. We use the planning from scratch (ML) technique
as an optimal benchmark for the comparison, while the static
incremental planning (Inc) technique is used as a pessimistic
benchmark, since it is unable to exploit the reconfigurability
of the IP and optical equipment, exhibiting in all periods the
worst performance. The joint multi-layer incremental
planning (JMR - IV.C) technique leverages the flexibility
provided by both network layers to achieve noteworthy
CapEx and OpEx savings. Virtual topology re-optimization
(VTR IV.B) and Optical layer re-optimization (OLR-IV.B)
techniques achieve limited savings by focusing solely on the
re-optimization of the virtual and optical layer, respectively.
More specifically, Fig. 6 depicts the cost evaluation of the
entire network assuming incremental planning with
12-month increments. In both network topologies, JMR
marginally underperforms the optimal benchmark (ML),
while it exhibits significantly higher efficiency (ranging
between 10% and 48%) when compared to the pessimistic
benchmark (Inc). The increased efficiency results from the
limited reconfiguration capabilities of the Inc technique at
both layers of the network. In the DT network, VTR performs
similarly to OLR, while it clearly underperforms the latter
for the TID network. This is indicative of the impact each
layer has in the re-optimization process of the network and
comes as a result of the different traffic profiles (higher
traffic growth for DT) and topologies (in the DT topology all
optical nodes are interconnected with IP routers, while in
TID topology, there are several optical transit nodes not
connected to IP routers with no traffic terminating/initiating
at those nodes).
a)
b)
Fig. 6. Capital expenditure of (a) DT and (b) TID topology for
different optimization options and 12-month network planning
periods.
Fig. 7 presents the results regarding spectrum utilization
for the two reference networks. All planning approaches
perform similarly with respect to spectrum utilization as to
the CapEx metric, even though we observe slightly lower
spectrum savings for the DT network, due to the deployment
of more regenerators for the DT topology, which provide
wavelength conversion possibilities. In contrast, TID
topology is characterized by lower traffic demands leading to
the deployment of lower order modulation format BVTs that
are able to exploit the trade-off between spectral efficiency
and reach. In cases where the link lengths of the network are
small enough, we are utilizing modulation formats that
increase the spectral efficiency of the network. In cases,
where the available spectrum is consumed (Fig. 7(a)), we
assume that extra fibers are installed. The cost of the fibers
and the equipment required for the installation of new fibers
is included in the calculation of the network cost.
a)
b)
Fig. 7. Maximum spectrum used for the (a) DT and (b) TID topology
for 12 month planning periods.
B. Lightpath re-configuration analysis
In this subsection we focus on the trade-off between
CapEx minimization of the equipment used in the current
state and the minimization of OpEx associated with the
optical equipment displacements and reconfigurations
between network states. Fig. 8(a) and Fig. 8(b) present the
number of reconfigured and added lightpaths per period
(considering a 12-month period), respectively. As already
stated, ML is agnostic to the previous state of the network,
leading to the optimum CapEx achieved through an
extensive reconfiguration of already established lightpaths.
Fig. 8(a) shows that the proposed joint incremental
multi-layer (JMR) approach limits the number of lightpath
reconfigurations and establishing of new lightpaths, and
consequently controls the corresponding OpEx. The JMR
technique achieves a significant reduction, of the order of
50%, of the reconfiguration processes, while maintaining a
relatively small number of added lightpaths per period,
which is only 18% larger than the one achieved by the
benchmark planning technique (ML - Fig. 8(b)).

309131
10
a)
b)
Fig. 8. (a) Reconfiguration overhead in the optical layer and (b)
number of added lightpaths per period for different incremental
planning techniques (12-month period).
C. Cost breakdown and impact of each layer in the
total CapEx
In Fig. 9(a) we present the cost breakdown for four
planning techniques examined in our study. As expected the
JMR technique exhibits the best performance, since it
exploits the flexibility of both network layers to minimize the
cost of the network equipment. The static incremental
approach (Inc) increasingly underperforms, due to its
inability to use the reconfiguration capabilities of the
network equipment. As time advances the bad choices made
by Inc aggregate and are not corrected at any point of the
network lifecycle.
Fig. 9(b) illustrates that under medium (2022) and heavy
(2026) traffic load, JMR balances the cost contributions of
both network layers to achieve a cost efficient solution, while
the performance of OLR and VTR is affected by the
single-layer re-optimization capabilities. In particular, OLR
exhibits significant increase in the IP layer costs while VTR’s
inability to exploit the reconfigurability of the optical layer
equipment affects the savings that can be achieved through
traffic grooming.
a)
b)
Fig. 9. (a) Network equipment cost breakdown, and (b) impact of
each network layer in the total cost of the network for different
planning techniques.
D. Duration of network planning periods
The duration of the network planning periods determines
the required investment to be made in order to ensure that
the resulting network design can cope with future traffic
until the next upgrade period. By adopting short network
planning periods we are able to avoid the deployment of
unnecessary equipment, and benefit from technology
maturation and corresponding price reductions.
This becomes evident in Fig. 10(a), where we use the JMR
technique assuming upgrade periods of different duration.
The 60-month upgrade period, leads to 77% higher CapEx
when compared to the 2-month network period, in the
beginning of each network planning period (2017a and
2022a). We obtain similar results when comparing with the
12 (28% - 39% higher CapEx) and 24 (42% - 48% higher
CapEx) months upgrade periods. The same pattern emerges
when examining the Telefonica network topology (Fig.
10(b)).
a)
b)
Fig. 10. Impact of the duration of the network planning periods to
the capital expenditure of the (a) DT and (b) TID topology, using
JMR incremental planning technique.
It is noteworthy that by adopting short network planning
periods we not only delay equipment deployment but we also
improve cost efficiency over the entire network lifecycle.
From Figures Fig. 10(a) and Fig. 10(b), and especially when

309131
11
examining the end of the 60-month period (2021_f and
2026_f), it becomes evident that we can obtain significant
savings by closely capturing traffic evolution and exploiting
technology maturation through short network planning
periods.
VII. CONCLUSION
The inevitable growth of the traffic to be transported by
optical transport networks, which may also be non-uniformly
directed, accentuates the need for planning methods that
have the ability to repurpose existing network equipment. In
view of this, we proposed planning techniques that account
jointly for the upgrade of the optical and the IP edges of the
network in an incremental manner. Through an ILP
formulation we exploited optimally the reconfigurability of
optical and IP equipment, with the objective being the
minimization of the equipment added at each period (CapEx)
and the equipment reconfigurations (OpEx) required
between two consecutive periods. We evaluated incremental
planning performance under realistic network scenarios and
quantified the impact of the reconfiguration capabilities of
each layer on the total network cost over the entire the
network lifecycle. Additionally, we verified that short
network periods are able to closely capture the effects of
traffic dynamicity and technology maturation, resulting in
significant cost savings.
ACKNOWLEDGMENT
P. Papanikolaou was supported by IKY PhD Scholarship
and K. Christodoulopoulos by IKY postdoc Scholarship,
which were funded from resources of “Human Resources
Development, Education and Lifelong Learning” 2014-2020,
co-funded by the European Social Fund and the Greek State.
REFERENCES
[1] “Cisco visual networking index: Forecast and methodology,
2015–2020,” Cisco White Paper, 2016.
[2] I. Tomkos, F. Effenberger, J. K. K. Rhee, “Introduction to the
special issue on optical networking for 5G mobile and wireless
communications,” IEEE/OSA J. of Opt. Com. and Netw., 2016.
[3] Y. Pointurier, "Design of low-margin optical networks," in
IEEE/OSA Journal of Optical Communications and
Networking, vol. 9, no. 1, pp. A9-A17, Jan. 2017.
[4] ITU-T Rec. 8080, “Architecture for the Automatically Switched
Optical Network,” 2012.
[5] M. Jinno, H. Takara, B. Kozicki, Y. Tsukishima, Y. Sone and S.
Matsuoka, "Spectrum-efficient and scalable elastic optical path
network: architecture, benefits, and enabling technologies,"
IEEE Com. Mag., 2009.
[6] O. Pedrola, A. Castro, L. Velasco, M. Ruiz, J. P.
Fernández-Palacios and D. Careglio, "CAPEX study for a
multilayer IP/MPLS-over-flexgrid optical network," in
IEEE/OSA Journal of Optical Communications and
Networking, vol. 4, no. 8, pp. 639-650, Aug. 2012.
[7] O. Gerstel, C. Filsfils, T. Telkamp, M. Gunkel, M. Horneffer, V.
Lopez, A. Mayoral, “Multi-layer capacity planning for IP-optical
networks”, IEEE Com. Mag., 2014.
[8] L. Velasco, P. Wright, A. Lord, G. Junyent, “Saving CAPEX by
extending flexgrid-based core optical networks toward the
edges [invited],” IEEE/OSA J. of Opt. Com. and Netw., 2013.
[9] V. Gkamas, K. Christodoulopoulos. E. Varvarigos, "A Joint
Multi-Layer Planning Algorithm for IP Over Flexible Optical
Networks," IEEE/OSA J. of Light. Techn., 2015.
[10] E. Palkopoulou, C. Meusburger, D. Schupke, L. Wosinska, T.
Bauschert, “Combining multi-period and multi-layer network
planning: Ignored potential?,” ECOC, 2010.
[11] C. Meusburger, D. A. Schupke, A. Lord, “Optimizing the
Migration of Channels With Higher Bitrates,” IEEE/OSA J. of
Light. Techn., 2010.
[12] A. Eira, J. Pedro, J. Pires, “Optimal multi-period provisioning
of fixed and flex-rate modular line interfaces in DWDM
networks,” in IEEE/OSA J. of Opt. Com. and Netw., 2015.
[13] P. Soumplis, K. Christodoulopoulos, M. Quagliotti, A. Pagano,
E. Varvarigos, “Actual Margins Algorithm for Multi-Period
Planning”. OFC, 2017.
[14] J. Pesic, T. Zami, P. Ramantanis, S. Bigo, "Faster return of
investment in WDM networks when elastic transponders
dynamically fit ageing of link margins," OFC, 2016.
[15] P. Papanikolaou, K. Christodoulopoulos and E. Varvarigos,
"Incremental planning of multi-layer elastic optical networks,"
2017 International Conference on Optical Network Design and
Modeling (ONDM), Budapest, 2017, pp. 1-6.
[16] P. Papanikolaou, K. Christodoulopoulos, E. Varvarigos,
“Multilayer flex-grid network planning,” ONDM, 2015.
[17] http://www.orchestraproject.eu/
[18] I. Sartzetakis, K. Christodoulopoulos, C. P. Tsekrekos, D.
Syvridis and E. Varvarigos, "Quality of transmission
estimation in WDM and elastic optical networks accounting for
space–spectrum dependencies," in IEEE/OSA Journal of
Optical Communications and Networking, vol. 8, no. 9, pp.
676-688, September 1 2016.
[19] A. Schrijver, “Combinatorial Optimization: Polyhedra and
Efficiency,” Springer, 2003.
[20] IIDEALIST Project, “Elastic Optical Network Architecture:
Reference scenario, cost and planning,” Deliverable D1.1, 2013
[Online].Available: http://cordis.europa.eu/project/rcn/105820_en.html