Content uploaded by Alexandros Stavdas
Author content
All content in this area was uploaded by Alexandros Stavdas on Oct 11, 2023
Content may be subject to copyright.
An SDN-enabled Path Computation Element for
Autonomous Multi-Band Optical Transport Networks
E. KOSMATOS,1* R. CASELLAS, 2 K. NIKOLAOU, 1 L. NADAL, 2 D. UZUNIDIS, 1
C. MATRAKIDIS, 1 J.M. FÀBREGA, 2 M. SVALUTO MOREOLO2 AND A. STAVDAS 1
1OpenLightComm Europe, Holešovice, 170 00, Praha 7, Czech Republic
2 CTTC/CERCA, Av. Carl Friedrich Gauss n7, 08860 Castelldefels, Barcelona, Spain
*Corresponding author:
Received XX Month XXXX; revised XX Month, XXXX; accepted XX Month XXXX; posted XX Month XXXX (Doc. ID XXXXX); published XX Month XXXX
Abstract
This paper reports on the design, implementation, and validation of an SDN control plane for Multi-Band Optical Networks with externalized Path
Computation. The SDN control plane relies on extending current open and standard interfaces to support dynamic service management and
decoupled path computation services on Multi-Band optical networks while accounting for physical layer impairments, which is critical for successful
service provisioning. We detail the Multi-band Optical Network Resource management and optimization engine for transparent and translucent
networks. The system is experimentally validated in a 22-ROADM BT network with emulated hardware showing the performance of the control
plane and the considered workflows.
http://dx.doi.org/10.1364/JOCN.99.099999
1. INTRODUCTION
Networked Intelligence, a term that designates the convergence of
Information, Communication and Operation Technologies, is the driver
behind the revolutionary changes Industry 4.0 evangelizes [1]. As part
of Networked Intelligence, a vast number of seemingly unrelated
technological breakthroughs converge to create a front of
interdependent innovations that reshape the way we live, work, and
produce. Specifically for the communications sector, the EU expert body
reckons that the diversity of technological domains required for future
communication infrastructures highlights the relevance of multiple
innovation domains [2].
In this regard, 6G radio and wireline Access technologies play a
fundamental role to interconnect all different forms and grades of
intelligence (human, humanoid or those in development within
Datacenter facilities) in a pervasive and ubiquitous way. However, this
progress puts the transport networks into a significant strain in terms
of capacity and connectivity [3]. To overcome the dangers of capacity
crunch and system gridlock in the network transportation segment, the
unused spectrum of the single mode optical fiber (SMF) should be
untapped by means of Multi-Band Wavelength Routed Networks (MB-
WRNs).
In MB-WRNs, multiple spectral bands as for example C-, L-, S-, E, O- or
even U- bands, span a total bandwidth of approximately 450nm [4],[5].
The exploitation of additional spectral bands not only increases the
capacity of point-to-point connections but also it allows to develop
hierarchically flatter network architectures in the context of an ‘optical
continuum’. In this case, a larger number of direct connections between
nodes is possible, limiting the number of O/E/O interfaces, and
aggregation/grooming stages needed to complete an end-to-end path.
In turn, this simplification lowers the total cost of ownership through the
reuse of the available and deployed fiber optical infrastructure,
enhances manageability and improves the end-to-end low-latency
performance. In this context, only transparent and translucent paths are
considered in this work, disregarding opaque networks as the latter
only increase point-to-point capacity without offering further
simplification [6].
The MB-WRNs capitalize on the framework of Elastic Optical
Networks (EONs) [7] to efficiently use of the available spectrum and
deploy new services cost-effectively. This way, the overall network
capacity and spectral efficiency is increased since channels are packed
using much narrower spectral guard bands [8].
To make the most of the MB-WRNs, it is needed to enhance network
manageability, thus programmability and automation need to be
introduced. Along this line, a Software-Defined Networking (SDN)
control plane plays a pivotal role to dynamically manage, scale and
optimize the network to meet the changing traffic demands. An SDN
controller is responsible for provisioning network services and to
implement complex automation tasks, such as closed-loop automation
processes where certain network conditions or events trigger further
actions in the network. Moreover, the decoupling of the Path
Computation function out of the rest of the control plane operations
provides network operators with additional flexibility when defining
network management policies to optimize routing and provisioning
separately from connection management and maintenance operations.
Consequently, an Optical Path Computation Element (PCE) is a
dedicated entity that is tasked to compute paths given a set of
constraints and the network status. Such a Path Computation function
can be made accessible to clients (e.g., the optical SDN controller) using
an open and standard interface. We refer to such entity as a SDN-
enabled Optical PCE.
In this work, we present the design, implementation, and validation
of an SDN control plane for multi-band optical networks with
externalized path computation function. Special focus is given to
present the intertwine between routing operations and physical layer
impairment constraints. In Section 2, we introduce the notion of
automation, and we present an overview of the MB-PCE architecture.
We highlight the shortcomings of transparent networks, and we
elaborate on the need to consider translucent paths. Section 3 presents
the resource management and optimization engine, and in particular
the algorithms upon which the routing engine is built. It also details the
input parameters, and it makes the case for the translucent paths.
Section 4 presents the closed-form mathematical expressions used in
this work to model the physical layer phenomena that degrade the
performance of multi-band transmission systems; these models are
further validated against the results from a numerical simulation tool.
Section 5 and 6 are dedicated to detail the implementation and
validation at the control plane level of the approach using open and
standard interfaces in an environment that represents production
systems, showing key aspects such as service provisioning or path
computation latency.
2. AN SDN-ENABLED OPTICAL PCE FOR MB-WRNS
As it is highlighted in the introduction, network automation has a
cardinal role to play in MB-WRNs to support dynamic cloud service
deployment and to increase operator's turnover by reducing the time to
new services.
Specifically, automation is key for network operators as it relates to
the process by which a network is reconfigured upon network state
changes with little to no human intervention. Macroscopically, it
involves network configuration and reconfiguration (such as service
provisioning) as well as performance monitoring and streaming
telemetry, to implement a closed-control loop where events or patterns
observed in the network trigger further actions. In particular, the
provisioning of connectivity services is delegated to an SDN-based
system that is logically centralized, as described in Section 5.
As mentioned, a particular aspect of relevance is the decoupling of the
path computation function from the rest of the control plane operations,
allowing additional flexibility to the selection of advanced routing and
path allocation policies and algorithms. The PCE is then tasked to
compute a path between two endpoints in the network based on an
abstracted representation of this network and its connectivity including
fiber propagation effects and transmission system specifications. This
task is an inherently multi-layer operation. The PCE retains information
for the status of the network and exploits suitably developed algorithms
that carry out optimizations aiming to adapt network resource usage to
changing conditions in an optimal manner.
The abstract models upon which the PCE is built hide the complexity
of the underlying network technologies. For this reason, these models
should balance accuracy with simplicity and the computational time
needed for their estimation, so the PCE is able respond to network
events in a timely manner. As it is shown in sections 3 to 5, the selection
of the appropriate abstraction methodology is key to reduce
computation times.
Figure 1: A schematic layout of the MB-PCE functionality
In this work, our goal is to build a PCE where the (transmission)
physical layer impairments of the underlying the optical layer, are
embedded into the decision-making process of path-selection. For this
reason, the PCE is materialized by means of constraint-based routing
algorithms that ensure:
• The routing engine makes the most of the available optical
spectrum while connection blocking is minimized by curtailing
the rejected connection requests. This is made possible by means
of an efficient spectrum and modulation-format assignment.
• The candidate optical paths are not falling short against suitably
chosen Quality-of-Transmission (QoT) target values.
• Diverse objectives and network management policies from the
higher management layers are met. To do so the PCE needs to
provide network connectivity while it utilizes the available
resources in the most efficient way without trading capacity
against connectivity requests.
These conditions may not be directly applicable to all transparent
end-to-end paths for several reasons. This may happen when i) the
attainable physical layer performance of a link is not sufficient to ensure
a high-quality connection; ii) the same number of contiguous spectral
slots are not available across a multi-hop end-to-end path; c) an end-to-
end path is crossing different administrative, operational and
ownership network domains where different transmission and
switching technologies may apply. If any of the above apply, the
connection is blocked.
Regarding (i) above, the physical layer degradations (or
impairments) are the result of distinguishable but intertwined
transmission linear and nonlinear physical effects. The accumulation of
Amplified Spontaneous Emission (ASE) noise, nonlinear effects that
depend on the power each channel is launched and/or the total power
in the fiber, transmission system configuration parameters as well as
fundamental technology constraints limit the transparent length.
Finally, the attainable transparent length of an end-to-end paths is also
limited due to the “optoelectronic bottleneck” which was used in legacy
WDM optical transmission systems to describe the disparity between
the data throughput a fiber-optic system could transmit and the speed
the electronic systems could process data [9]. The “optoelectronic
bottleneck” makes itself present again in EONs, this time as the
Optical Multi-Band SDN Controller
REST APIs (ONF Transport API extension)
L-band network
Impairment
Models
PHY Abstractions
QoT Models
Modulation Format Models
Transceiver Models
Channel Specifications
S-band network
E-band network C-band network
Multi-band
Optimization
PHY Aware Multi-band
Routing Modulation and
Spectral slot Assignment
Network DB (links,
paths, resources,
topology)
Multi-Band Network
(links, paths, nodes,
resources, topology)
Databases
Resource Management
Optical Multi-Band Path Computational Element (MB-PCE)
Network Management Policies Network Optimization Applications
discrepancy between the attainable line-rate and the symbol-rate of the
digital-to-analog converter (DAC) and analog to digital converter (ADC)
as it is clearly shown e.g. in Fig.3 of [10]. As a result, to attain a much
higher line-rates than the native DAC/ADC rates, higher cardinality
modulation formats are employed eventually limiting the attainable
transparent length [10-12].
To overcome these limitations, translucent networks allow signal
regeneration and/or spectral conversion in sparse but strategic
locations in the network. The corresponding flows may not necessarily
undergo a L2 aggregation, grooming and switching as sliceable,
bandwidth variable transmitter and receiver [13] can be interfaced
back-to-back to provide the requested functionality [14]. Nevertheless,
a network with translucent paths, which incorporates shorter length
end-to-end transparent paths, comes with an additional CapEx cost
associated to these spare transceivers that need to be allocated in key
network nodes. Finally, as multi-band transmission systems are more
prone to physical layer degradations compared to C-band-only systems,
a MB-PCE should provision both transparent and translucent paths to
enhance resource utilization and network manageability.
In this work, we report a MB-PCE that exploits a Physical Layer
Impairment-aware Routing, Modulation and Spectrum Assignment
(PLI-aware RMSA) routing engine. A schematic layout of the MB-PCE is
presented in Fig.2 where the physical layer abstractions and routing and
optimization algorithms are employed to maximize the utilization of the
available resources and to reduce network blocking as detailed in
sections 3 and 4. An open API is used to interface the MB-PCE to the
Optical SDN controller detailed in section 5.
3. A MULTI-BAND OPTICAL NETWORK RESOURCE
MANAGEMENT AND OPTIMIZATION ENGINE
3.1. A MB-PCE: Key Elements and Concepts
A translucent end-to-end lightpath is decomposed in two or more
disjoint but consecutive transparent paths where the channel is o/e
regenerated at the edges of these paths. Therefore, as a translucent path
is a more general category that incorporates the notion of a transparent
path, a PLI-aware RMSA that is limited to networks with fully
transparent paths only as in [15], finds its niche here too. In this section,
the RMSA algorithm of [15] is extended to incorporate the more general
case of translucent paths.
The operations of the multi-band routing engine are completed in
four consecutive stages as below and the corresponding chart flow is
illustrated in Fig.2.
Stage-I: Network Topology Definition and System Specifications
At this stage the details of the networks and the transmission systems
under consideration are inserted as Input parameters to the routing
engine. These are:
• Network topology definition: the geospatial positioning of the
nodes and the connectivity pattern between them are specified;
the number of edges are defined. The links interconnecting the
nodes are stated together with the location of key elements like
the optical amplifiers. Finally, the k-shortest paths for all network
node pairs are defined and stored.
• Optical system parameter definition. These include the definition
of high-level parameters, like the available optical bands, the size
of the elementary Frequency Slot Unit (FSU) as of [7], the number
of FSUs per band and the total capacity per band, as well as the
operational parameters of the deployed transmission systems.
Fig.3a illustrates the parameters that define an arbitrary point-to-
point link in this work, while the tables in Fig.3b associate the per
link parameters we use in our routing engine (left-hand side) to
the corresponding list of parameters of the IETF standard [16]
(right-hand side).
• Traffic matrix definition: This requires designating the demands
between all nodes and their edges as well as the available line-
rates and their allocation to the corresponding demands. Also,
the average time duration of the demands and the average inter-
arrival time between two consecutive demands.
Figure 2: The flow chart of a multi-band PCE operation for translucent
paths.
Stage-II: Formation of the Spectral and Modulation Format Assignment
(SMA)
This stage is completed in two steps: In the first step, i) a preliminary
spectrum and modulation format assignment (SMA) is made for several
the k-shortest paths, and ii) the QoT metric, Optical Signal to Noise plus
Interference Ratio (OSNIR) (see section 4) for these shorter paths is
estimated by means of closed-form/analytic mathematical expressions.
In the second step, the two processes above are pipelined to allow the
algorithm to either approve or reject the creation of a lightpath. A path
is rejected if at least one of the following reasons apply:
• Spectral slot unavailability: no contiguous spectral slots are
available in any optical band to support a transparent end-to-
end connection.
• Inadequate physical layer performance: either the OSNIR of the
candidate lightpath falls short of the QoT estimator threshold or
the OSNIR of at least one of the already established lightpaths
performs below the QoT threshold due to the presence of this
candidate lightpath. This estimation is made for an arbitrary
Inputs
network topology (nodes, links, distances, amplifier
positions), traffic matrix, request arrivals/departure
parameters, candidate Tx/Rx parameters, available
bands, # FSU per band, max # of k-shortest paths
New Request
k-shortest path estimation
Route the Request
Assign Spectrum temporarily
Calculate Path OSNIR
Is the calculated OSNIR higher
than the minimum Tx OSNIR?
YES
Block Request
YES
Initialization Phase Spectrum check phase
NO
Assign Request
Allocate modulation format,
routing path, spectrum and power
PL check phase
Assignment phase
Request and traffic
generation
Apply Path Split Algorithm
Select the next available band
NO
NO
YES
YES
All available bands are
checked?
NO
YES
All k-paths are checked?
Select the next available k-path
NO
Path selection
Select the next continuous
available FSUs using First-Fit
algorithm
Are spectrum resources
available along the path in
the selected band?
YES All FSUs are checked?
NO
All path split alternatives
are checked?
Select the next path split
alternative
Split the request into two or
more new requests
(segments)
NO
YES
Is the request segment of
a split request?
Are all the other segments
accepted
Assign all the segments
of the split request
Allocate modulation format,
routing path, spectrum and
power
YES
Assign this segment as
accepted
Select the next request segment
in the list of segments
NO
Are all the other segments
checked
NO
YES
All modulation formats are
checked?
NO
Select the next modulation
format
YES
number of channels present in the link. Moreover, the QoT
estimator may or may not include suitably chosen operational
margins as detailed in [15].
(a)
(b)
Figure 3: (a) The input system parameters to the multi band routing
engine; (b) the parameters (on the left) in relevance to those from the
IETF standard (on the right) [16].
In either case, the rejected lightpath is assigned the next available
path from the sorted list of k-shortest paths and then it is re-iterated. If
these paths are all rejected, the first step is repeated using, if possible, a
lower cardinality SMA values. If no path is retained, the routing engine
registers a blocking condition, and it designates this lightpath as a
candidate for Stage-IV.
Stage-III: Finalization of Optically Transparent Paths
During this stage, the MB-PCE finalizes the transparent lightpaths
[15]. The final assessment on network’s throughput is completed and
the lightpath is successfully pertained triggering an update of the
corresponding arrays for each link across the path, e. g., the arrays of
active connections, launch power, modulation format, consumed
frequency slots etc.
Stage-IV: Formation of Translucent Paths
A lightpath that has been rejected during Stage-II, it is forwarded to
the “Path-Split Routine” (PSR) (Fig.2) for further iteration. At this stage,
the algorithm splits a rejected transparent optical path into two or more
shorter-length transparent paths followed by full o/e regeneration at
intermediate nodes along the original path (translucent path
generation).
This way, the two (or more) shorter length transparent paths may
exploit a more suitable combination of a higher cardinality modulation
formats and a lower symbol-rate source. It is possible to select a
different modulation format/symbol-rate pair for the newly candidate
transparent paths as these are decided independently.
To understand the benefits arising from this path split, it is reminded
that the number of optical slots ‘consumed’ by a lightpath is defined by
the symbol-rate employed. To be more specific, the selected symbol-
rate defines the optical bandwidth this channel consumes. Therefore,
the number of FSUs a particular DAC/ADC system requires is given as
the ratio of the optical bandwidth needed to the FSU spectral width
(which is 12.5 GHz throughout this work). Therefore, by downgrading
the symbol-rate in the candidate shorter-length transparent paths, the
total number of optical slots consumed in these new paths is reduced,
reducing the blocking probability due to spectral unavailability.
Moreover, given the shorter transparent length of the two (or more)
paths that emerged from the original path, the chance for a better QoT
performance is improved (although this cannot be guaranteed as the
selection of the aforementioned parameters play a key role). Section 3.2
further provides the details of the corresponding PSR algorithms in
relation to the network management policies of Fig.1.
3.2. The Path-Split Routine and the Related Algorithms
The PSR is the algorithmic procedure that splits a path P from a node
S (source) to a node D (destination) into two or more consecutive paths
based on constraints/criteria associated to higher-level objectives like:
a quest to maximize spectral efficiency, the need to retain the network
in a not blocking state, the necessity for a connection to possibly cross
other administrative domains etc. It is pointed out that this is an
exclusively L1 operation so no grooming or any other L2 process takes
place. The Path Split (PS) algorithm provides as ‘Output’ a set of disjoint
but consecutive paths out of the initial, longer, path. These solutions are
the alternative candidate ‘split paths’ that meet the requested criteria.
The corresponding algorithms are using the parameters and the
variables summarized in Table 1.
Table 1: Variables and Parameters for the Path Split Algorithms
Variable
Description
G
network topology graph
N
set of network nodes
E
set of bidirectional optical fiber links (edges)
B
set of active optical bands
CB
set of capacities (number of frequency slot units - FSUs) for
each optical band in the set B
p
Path between a source node and a destination node
ps
source node of path p
pn
destination node of path p
Np
set of network nodes included in the path p, ordered from
source to destination
Ep
set of edges included in the path p, ordered from source to
destination
Sg
set of path-split alternatives for a specific path p as an
outcome of the path split algorithms
b
optical band
RB
set of available transmitter types for each optical band in
the set B, ordered with increasing value of required FSUs
for each transmitter
r
transmitter type
rd
max distance supported by transmitter type r
rf
number of consequent frequency slot units (FSUs) required
by transmitter type r
ei,j
Edge between node i and node j
u
Utilization of a specific edge e calculated as the ratio
between all available FSUs and occupied FSUs.
INPUT: The initial path p from a source node ps to destination node pn.
The set of network nodes Np and the set of edges Ep included in the path
p from the source to destination nodes. The band b in which the path will
be allocated. The utilization threshold uth is for those algorithms need to
use such a parameter.
OUTPUT: A set of path-split alternatives sg for the path p, sorted from
the mostly to least preferred candidate.
The PS algorithm prioritizes (sorts) the candidate path splits based
on a second set of criteria. These latter criteria are deduced by means of
the following algorithms which are integrated into the MB-PCE’s routing
engine:
Tx .
.
.
Rx
.
.
.
L km
TDFA
EDFA
TDFA
EDFA
PDFA
BDFA
Transmitter Parameters
1. Operational mode
2. Modulation Format
3. Baud rate
4. Bandwidth
5. Channel spacing
6. Data Rate
7. Transmitted power range
8. Emission frequency range
Channel Parameters
1. Fiber type
2. Fiber length
3. Amplifier type
4. Amplifier operational mode
5. Flat amplification range
6. Noise Figure
7. Small signal Gain
8. Maximum output power
9. Band filter attenuation
Receiver Parameters
1. Operational mode
2. Received power range
3. Receiver’s frequency range
4. FEC type
5. FEC code rate
6. FEC threshold
7. DSP type
Transmitter Parameters
Organization identifier ○
Operational mode ○
Modulation Format○
Baud rate○
Bandwidth○
Channel spacing○
Data Rate○
Transmitted power range○
Emission frequency range○
Receiver Parameters
Organization identifier○
Operational mode○
Received power range○
Receiver’s frequency range○
FEC type○
FEC code rate○
FEC threshold○
DSP type○
omin-central-frequency
omax-central-frequency
ocentral-frequency-step
otx-channel-power-min
otx-channel-power-max
orx-channel-power-min
orx-channel-power-max
orx-total-power-max
oSupported modes
oline-coding-bitrate
oBit rate
omax-polarization-mode-dispersion
omax-chromatic-dispersion
ochromatic-and-polarization-dispersion-penalty
omax-diff-group-delay
omax-polarization-dependent-loss-penalty
oavailable-modulation-type
omin-OSNR
omin-Q-factor
oavailable-baud-rate
oroll-off
omin-carrier-spacing
oavailable-fec-type
ofec-code-rate
ofec-threshold
Transceiver Parameters
Link Parameters
Fiber type○
Fiber length○
Amplifier type○
Amplifier operational mode○
Flat amplification range○
Noise Figure○
Small signal Gain○
Maximum output power○
Band filter attenuation○
otype-variety
olength
oloss-coef
ototal-loss
oPmd
oconn-in
oconn-out
oConcentratedloss
otype-variety
ooperational
ofrequency-range
oactual-gain
otilt-target
oout-voa
oin-voa
ototal-output-power
ochannel-power
opower-spectral-density
oraman-direction
oraman-pump
oroadm-maxloss
Link Parameters
• High Modulation Format First Path Split (HighMFFirst)
• Utilization based Path Split (UtilBasedSplit)
• Combined Path Split with Utilization First (CombUtilFirst)
• Combined Path Split with Modulation Format First
(CombMFFirst)
1. HighMFFirst: The HighMFFirst algorithm splits the initial path into
two (or more) separated paths so as the highest feasible modulation
formats are assigned in the newly created consecutive paths.
The main objective of this algorithm is to maximize spectral
efficiency. This is done by using the highest permissible cardinality
modulation format for the longest of the subsequent translucent paths
and for the highest number of the incoming requests to Stage-IV.
Specifically, the performed operations are detailed below:
Set ;
while RB is not empty do
Consider the first transmitter type ;
Set ; Consider the first node ;
while do
Consider the next node ;
while distance(ni,nj)<rd do
Consider the next node ;
end while
Add nj into Gc; Set ni=nj;
Repeat the above steps starting from the last node up to the first
node;
Remove r from ;
end while
if
Add Gc into Sg; Set ;
end if
end while
To further elaborate the operation, the functionality of the
HighMFFirst algorithm is graphically depicted in Fig.4. In this example,
we assume it is requested to route a 400Gb/s connection from node 1
to node 8 that spans a path of 1,200 km in length. If the connection is
transparent, a path via nodes {1-2-3-4-6-7-8} is established as shown in
Fig.4.
With the aid of Fig.3 in [10] where use is made of the case (c) with
two-zone OSNIRs (zone-1: E and S1-bands; zone-2: S2, C and L bands)
and for a transmission system employing QPSK with a 128 Gbaud
symbol-rate DAC/ADC (FSU = 12.5 GHz), 11 FSUs need to be consumed
in every edge across the transparent path during Stage-II.
However, if there are no contiguous spectral slots available across the
entire path during Stage-II, the candidate connection is forwarded to
Stage IV where the potential for a translucent path is considered. In the
translucent mode, the HighMFFirst algorithm splits the initial path {1-2-
3-4-6-7-8} into the consecutive paths {1-2-3}, {3-4-6} and {6-7-8}. Each
of these paths is assigned a set of contiguous but spectrally disjoint
spectral slots i.e. each set of spectral slots is centered at different carrier
wavelength (which in Fig.4 they are denoted with different colors)
spanning a shorter path distance of e.g., 400 km. For these newly created
paths, a 32QAM modulation format and a DAC/ADC with 48 Gbaud
symbol-rate is used instead, consuming 4 FSUs in every edge across the
initial path, giving the chance to accommodate this connection. This
algorithm is used primarily during the connection set-up phase.
Figure 4: A use-case for the HighMFFirst algorithm
2. UtilBasedSplit: The UtilBasedSplit algorithm splits the initial path
based on link utilization considerations. The main objective of the
algorithm is the relaxation of the frequency continuity constraint near
(before or after) the network edges as slots might be unavailable only
between two specific edges and not across the entire path. The
UtilBasedSplit is used when the objective is to minimize the number of
FSUs assigned in such highly utilized links. The related algorithm
operations are:
Set ; Set ; Set ;
Consider the first edge ;
while do
Consider the next node ;
Consider the edge ei,j between ni and nj;
if
Add ni into Nc; Add nj into Nc; Set ni=nj;
end if
end while
while Nc is not empty do
Consider the first node ;
Add n into Gc; Add Gc into Sg; Set
end while
The functionality of the UtilBasedSplit algorithm is graphically
depicted in Fig.5. Using the details of the previous example, we assume
that the path {6,7} is highly utilized so only 5 FSUs are available. Under a
transparent mode of operation, the connection is blocked due to the
unavailability of FSUs along this link as the QPSK format consumes 11
FSUs.
In the translucent mode of operation under the UtilBasedSplit
algorithm, the initial path {1-2-3-4-6-7-8} is split into two paths: the {1-
2-4-6} of 800 km and the {6-7-8} of 400 km. Regarding the former path,
a 64 Gbaud symbol-rate and a 16QAM modulation format is assigned,
consuming 6 FSUs in every edge across this path. Regarding the second
and most troublesome path, a 32QAM modulation format and a
DAC/ADC with 48 Gbaud symbol-rate is used instead, consuming 4
FSUs. In this case, the connection can be granted under the translucent
mode.
Finally, two more path-split algorithms (CombUtilFirst and
CombMFFirst) are introduced that combine the functionality of the
aforementioned algorithms, to maximum the gains from the available
pool of transponders. These algorithms find their niche when the PCE is
tasked to maintain and defragment the network during the in-
operational phase.
Request: 400Gb/s, Path distance: 1,200Km, Source: 1 Destination: 8
1
7
643
2
8
5
1
7
643
2
8
5
MF: 32QAM, FSUs: 4 in total
MF: 32QAM
FSUs: 4
MF: 32QAM
FSUs: 4
MF: QPSK, FSUs: 11 in the total path
Transparent Mode
Translucent Mode
(HighMFFirst)
= Transponder pool
Figure 5: A use-case for the UtilBasedSplit algorithm
3. CombUtilFirst: The CombUtilFirst algorithm integrates the two
previous algorithms, while during the final classification, priority is
given to splits offering higher FSU utilization. The main objectives of this
variant are: a) if a network is underutilized, it would maximize spectral
efficiency; b) if a network is congested it would both relax the frequency
continuity constrain and it will maximize spectral efficiency. The related
algorithm operations are:
Set ; Apply HighMFFirst algorithm;
Consider the outcome of METF-PS algorithm;
Add into ; Apply UBF-PS algorithm;
Consider the outcome of UBF-PS algorithm;
while is not empty do
Consider the first path split alternative ;
if
Add Gc into Sg;
end if
Remove Gc from ;
end while
4. CombMFFirst: The CombMFFirst algorithm integrates the first two
algorithms, but in this case, priority is given to path splits that allow the
higher modulation format cardinality while simultaneously lowering
the demand from the corresponding symbol-rate. The main objective of
the algorithm is, like with the CombUtilFirst, to maximize the spectral
efficiency while relaxing the frequency continuity constraints.
4. THE ABSTRACTION MODEL FOR THE OPTICAL
FIBER PROPAGATION IMPAIRMENTS
As detailed in sections 2 and 3, abstract models of the physical layer
phenomena which are held responsible for performance degradation in
multi-band transmission systems have been readily incorporated into
MB-PCE’s routing engine. The abstraction methodology is the same as
in [15] while the quantity OSNIR is the metric we have selected to weight
the QoT. The OSNIR accounts for ASE noise accumulation and for fiber
nonlinearities. The latter refers to both intra-band effects, like cross-
channel interference (XCI- eq.4 in [15]), and self-channel interference
(SCI- eq.3 in [15]), as well as inter-band effects, like Stimulated Raman
Scattering (SRS). The SRS, although it is not a dominant degradation
effect in C-band only transmission systems, may lead to a severe
performance degradation in MB-WRNs. In fact, the power transfer from
lower to higher wavelength channels due to SRS results in considerable
power depletion of the former. Accordingly, the OSNIR is given by eq.
(1).
,,
1
,
s
ch i
N
SRS j
j
NLI intra,i
ASE i
OSNIR
PG
PP
=
−
=+
(1)
where Pch,i is the power of the ith channel at node ingress/egress, PASE,i
and PNLI-intra,i denote the powers due to ASE accumulation and the intra-
band nonlinearities for the ith channel at the end of the path, respectively,
while the GSRS,j accounts for the SRS Gain/Loss effect in the jth fiber span.
In eq. (1) the intra-band and inter-band effects are considered as two
independent phenomena with no cross-coupling between them i.e., the
transmission system operates in the ‘weak nonlinearity’ regime where
the power transfer due to SRS does not significantly affect the evolution
of XCI/SCI. As it has been shown in [17-19] and in Fig. 4b of [21], as long
as the total power in the optical fiber does not significantly exceed the
+21 dBm, the assumption of no cross-coupling between nonlinearities
holds. In section 4.2 it will be made clear that the ‘weak nonlinearity’
conditions do apply when the operational parameters of the
transmission system are duly optimized.
In [20], the SRS is estimated for a maximum channel spacing of up
to 15 THz in-line with the so-called ‘triangular approximation’.
However, the true Raman curve is extended in frequency up to
approximately 35 THz (see Fig.6). In this work, the impact of this
extended spectrum is taken into consideration following the
formalism of [22] where the true Raman spectrum beyond a 15
THz channel spacing is approximated as a logarithmic tail.
Figure 6: True and approximated Raman spectrum for a standard single
mode fiber.
Because the MB-PCE is carrying out consecutive operations of
considerable computational complexity, it is imperative the physical
layer abstraction models to complete the necessary calculations in a
timely fashion. To ensure this is the case, the corresponding physical
effects are abstracted by means of closed-form mathematical
expressions is in [15], [22]. It is also possible to consider other semi-
analytical approaches, such as [23].
We opted to the closed-form formalism of [15] to estimate the impact
of physical layer effects for MB transmission over other closed-form
alternatives of the ”Gaussian-Noise” family like [24]-[26] as: a) the
formalism we have adopted takes into account the impact of SRS over a
spectrum of 30-35 THz which, as shown in Fig.6, has not a negligible
contribution to the overall power exchange between channels
especially when three or more amplification bands are considered; b)
our formalism predicts very accurately the performance of links with a
span length shorter than 30 km. In BT’s network there are links down
to 2 km in length and this is an important consideration in our analysis.
The closed-form abstraction models we use here are derived under
several approximations which are summarized in section 4 of [15].
These approximations apply here, too.
Request: 400Gb/s, Path distance: 1,200Km, Source: 1 Destination: 8
1
7
643
2
8
5
1
7
643
2
8
5
MF: 16QAM, FSUs: 6in the path
MF: 32QAM
FSUs: 4 in the link
MF: QPSK, FSUs: 11
High utilized link
(5 FSUs free)
Blocking
High utilized
link
No Blocking
0 5 10 15 20 25 30 35
True Spectrum
Triangular Approximation
Proposed Approximation
Raman Gain [a.u.]
Relative Frequency [THz]
4.1. Validation of the Spectrally Extended SRS Model
The spectrally extended SRS model was validated in [22] by
benchmarking the results from the closed-form abstract model to those
from a commercially available simulation tool (VPI). The latter employs
the Split Step Fourier Method (SSF) to numerically solve the
corresponding nonlinear propagation equations in an optical fiber by
making use of the true Raman spectrum.
Here we provide further evidence for the validity of this formalism by
presenting the results from simulations with systems a) having a higher
number of channels; and b) with a wider range of values for the total
power traversing the fiber link.
The merit function against which the benchmarking is made is the
quantity GSRS, which estimates the gain (or the depletion) in channel’s
power at the end of a single link due to the SRS effect. With reference to
Fig.3a, the link consists of a fiber span plus the corresponding multi-
band amplification stage. In the latter, the different xDFA terminals are
individually regulated so their gain just compensates for the losses due
to propagation along the preceding fiber span plus the losses due to the
band mux/demux which were assumed to be of the order of 2 dB. The
GSRS is ratio of the output power, as it is measured at the output of the
band multiplexer which is the zero (gain x loss) point for the combined
losses, over the launch power for a single channel. By excluding the
impact of all other non-linear effects, the higher/lower channel power
that is estimated/measured at the output of the link is due to SRS.
We consider two multi-band transmission systems/scenarios with
305 Nyquist channels in total (with 61 channels per band) and 505
Nyquist channels in total (with 101 channels per band) in E, S, C and L-
bands. Each channel is operated at a 100G line-rate with 32 Gbaud
source rate and a QPSK modulation format. We consider a single link of
50 km in length with the parameters of Table 2 for the central
wavelength (λ) of each band, the values for the fiber attenuation (α),
dispersion (D), nonlinear coefficient (γ), effective area (Αeff) per band
and noise figure (NF) of the different amplifiers (see Fig.3a), as these
fiber parameters are wavelength dependent. To compensate for the
losses, xDFA amplifiers are used with characteristics as in [10]: a
Bismuth-DFA for the E–band while the S band is segregated in two sub-
bands (S1-band and S2-band) to ensure the corresponding amplifiers
provide sufficient power per channel [15] so two separate Thulium-
DFAs are used for the S1 and S2-bands. Finally, two Erbium-DFAs are
optimized for the C and L bands. The values of the corresponding NFs
are as in [10]. By means of these parameters, the GSRS is always estimated
for the central channel in each band.
Table 2: The Parameters of the Multi Band Transmission Systems
E band
S1 band
S2 band
C band
L band
λ (nm) -
central
1416.5
1466.7
1496.7
1546.9
1594.6
α (dB/km)
0.280
0.246
0.229
0.211
0.210
D
(ps/nm/km)
8.63
12.06
13.97
16.96
19.60
γ (1/W/km)
1.65
1.50
1.44
1.32
1.24
Αeff (μm2)
70
74
76
80
83
NF (dB)
6
5.5
5.5
5.5
6
Regarding the numerical simulation tool, the following parameters
were used: the fiber type was the “universal” one to ensure that the
power exchange due to SRS between any pair of channels across the
entire spectrum, from the E to the L band, is accounted for in all cases;
the “sample-mode bandwidth” is set for 32.7 THz while 8,192 bits were
used per polarization. The wavelength-dependent physical layer
parameters of Table 2 e.g. a, D, Aeff, were manually inserted using text
files. The remaining configuration parameters of the link were the
following: at the end of each fiber segment, a “WDM_DEMUX” is inserted
to split the spectrum in bands so each band is amplified separately. This
demux had “rectangular” shaped filters, while the “WDM_MUX” module
which is placed in tandem mirrors the demux (see Fig.3a). Under these
specifications, for a system with 305 channels across the E, S, C and L-
bands, the spectrum is as in Fig.7.
Figure 7: The spectrum with 305 channels across the E, S, C and L-bands
To calculate the GSRS in the context of the numerical simulation, we
measured the optical power of each channel at fiber, input which was
then compared against the power at the output of the band multiplexer.
(a)
(b)
Figure 8: Benchmarking of the spectrally extended SRS model using the
GSRS vs power per channel against the corresponding numerical results
for a 50 km link with (a) 305 channels; and (b) 505 channels.
In Fig.8a and 8b we benchmark the the GSRS vs. Pch values from the
analytical and the numerical tools for a multi-band system with 305
E-band S1-band S2-band C-band L-band
-5.0
-4.0
-3.0
-2.0
-1.0
0.0
1.0
2.0
3.0
4.0
-8 -7 -6 -5 -4 -3 -2 -1 0
Numerical E BAND
Analytical E BAND
Numerical S1 BAND
Analytical S1 BAND
Numerical S2 BAND
Analytical S2 BAND
Numerical C BAND
Analytical C BAND
Numerical L BAND
Analytical L BAND
Power Per Channel (dBm)
GSRS [dB]
-5.0
-4.0
-3.0
-2.0
-1.0
0.0
1.0
2.0
3.0
4.0
-10 -9 -8 -7 -6 -5 -4 -3 -2
Numerical E BAND
Analytical E BAND
Numerical S1 BAND
Analytical S1 BAND
Numerical S2 BAND
Analytical S2 BAND
Numerical C BAND
Analytical C BAND
Numerical L BAND
Analytical L BAND
Power Per Channel (dBm)
GSRS [dB]
channels and 505 channels in total, respectively. In particular, in Fig.8a
the launch power is iterated from -8 dBm per channel to – 1 dBm and in
Fig.8b from -8 dBm per channel to -3 dBm per channel, respectively,
values that correspond to a total optical power in the fiber of the order
of ~250 mW. It is important to point out that all channels are
simultaneously set to the same power level as the power is iterated at
the system’s ingress.
For both systems in Fig.8, the maximum deviation between the
analytical and the numerical methods is of the order of 1 dB when the
total power in the fiber is up to +24 dBm. These findings are in-line with
the limits set in the previous section on the allowable total power within
a fiber segment for the corresponding transmission system to operate
in the weak-nonlinearity regime. Evidently, from Fig. 8 the impact of the
SRS for a single fiber span becomes increasingly critical as the number
of channels in the system and/or the power per channel increases. As a
result, considerable power is transferred from E- and S-bands to L and
C-bands as the number of cascaded fiber segment increases.
4.2. Optimal Conditions for the Physical Layer Abstraction
Model in the Context of the MB-PCE
The results of Fig.8 were obtained under the hypothesis that all
channels in all bands are launched at the same power level. Although
this is a reasonable assumption when the quest is to pinpoint the
physical phenomenon of SRS in isolation, it is a sub-optimal method to
deduce the operational conditions in the context of a PLI-aware RMSA
algorithm where other linear and nonlinear phenomena contribute too
to the final performance. This point has been analyzed in [10],[15],[22].
To overcome this obstacle additional design and algorithmic steps are
necessary to identify the optimal operational parameters.
4.2.1 Power Launch Optimization Strategy:
Different methods have been proposed in the literature to identify
the optimal launch power per channel in multi-band transmission
systems. In [27], power slopes and offsets are used to construct an
optimum distribution of launch powers for the C+L+S bands only.
However, the estimation of the optimal power per channel in a MB
system is an NP-hard problem, as the power of each channel affects the
power of all other channels spaced up to 35 THz. Another important
consideration is that the operators very seldom allow the per link
utilization to be above 70% while, as shown in Fig. 11 of [15], many links
remain under-utilized at lower loads even if central links are utilized to
the maximum. So, if the optimal power per band is estimated when the
links are fully loaded, the returned value will be sub-optimal at other
operational conditions. Moreover, the SRS gain depends on the specific
channel allocation, so the amplifiers need to be regulated every time a
new channel is added stretching the capabilities of the SDN control-
plane. To overcome these limitations, other researchers proposed
methods based on iterative searching or brute-force searching that
optimize the launch power per channel/band to improve the overall
transmission capacity [28]-[29], but these methods are time consuming.
The approach we adopt here is reported in [15], [10] according to
which, the different Pch,{band}, representing the power all channels in a
bands are launched, are used as free variables of a co-optimization that
tailors the OSNIR performance per band (or the requested transparent
length) to the policies issued by the higher layer management
(designated as “network management policies” in Fig.1). Our method
manifests a very good balance between complexity and accuracy, it
applies to an arbitrary number of bands, while it does not stress the SDN
control-plane with frequent adjustments and tweaks on the
corresponding optical amplifiers.
As an example, we are presenting two out of the three alternative
band exploitation schemes reported [10]:
• In the first scheme, case (b) in Fig.3 of [10], the OSNIR is requested
to have a ‘flat’ (less than 1-dB) variation across the entire E-band
to L-band spectrum. This is made possible by trading the OSNIR
performance in C and L bands in favor of E and L bands. This way,
all channels from E to L band have the same optical reach.
• In the second scheme, case (c) in Fig.3 of [10], the OSNIR
performance in C, L and S2 bands is maximized at the expense of
the performance in E and S1 bands. As it is illustrated in Fig.2a of
[10], the result of this optimization leads to a ‘step-like’ OSNIR
performance where the aim is to employ C, L and S2 bands to
serve long-haul connections while E and S1 bands are used for
shorter-length connections only.
It is reminded that in both band exploitation schemes, the different
Pch,{band} are optimization variables and it is only their values that change
according to the policies issued by the higher layer management for the
functionality of the bands (or the requested transparent length). These
Pch,{band}, values are deduced from a suitably chosen optimization
algorithm which is our case is based on simulated annealing [30]. This
algorithm is a separate and external to the algorithm illustrated in Fig.2.
The system’s energy is given by eq. (2)
(2)
where the constants Cband allows to perform the aforementioned
relative tuning of the OSNIR performance per band, while the OSNIRband
values are those calculated using eq. (1).
The operations of the simulated annealing algorithm are shown
below:
Initialize Pch per band values;
Set E = current system energy;
Set T = start temperature;
while T > end temperature do
Modify one band’s Pch value;
Set E’ = current system energy;
Set ΔE = E’ – E;
if ΔE < 0 or e-ΔE / (Κ * T) > some uniform random number
Set E = E’;
else
Undo earlier Pch modification;
end if
Multiply T by the cooling rate;
end while
Given that the model of eq. (1) internally works with a discrimination
of the power levels down to the individual channel, this approach can be
further expanded towards a more fine-grained power optimization,
partitioning each band in smaller segments, but this variation is not
explored further here. Moreover, this optimization is applicable for an
arbitrary number of channels per band.
4.2.2. Further validation of the proposed band exploitation scheme: As
mentioned in section 4.1, the impact of the SRS for a single fiber span
becomes increasingly critical as the number of channels in the system
and/or the power per channel increases. Therefore, despite the
introduction of the optimization method of 4.2.1, the impact of SRS on
signal power evolution should be countered at regular intervals. Where
this compensation should take place and which methods should be used
to do so, is still an open issue in literature. In [31] it is suggested to
compensate the effect of SRS per fiber span while in [10],[15],[22] the
compensation of SRS takes place at a node based on Wavelength
Selective Switch (WSS) elements. Indeed, Liquid Crystal on Silicon
(LCoS) WSSs have an inherent ability to regulate the power level per
channel and per band to the desired level. As such, the latter SRS
compensation method is a lower CapEx solution with a reduced
computational complexity and a lower burden for the SDN control play.
The different band exploitation plans, like the flat/step-like OSNIR
presenting in the previous section, are made feasible by combining the
optimal power launch strategy of section 4.2.1 and the SRS
compensation at the WSS-based optical nodes along an end-to-end path.
In the rest of this work, we solely focus on the band exploitation plan
featuring a flat-OSNR performance across the E to L-bands spectrum.
Here we extend the work reported in [22] to provide further evidence
for its robustness.
Following the example of section 4.1, the results from our closed-
form physical layer model are benchmarked against those from a
commercially available simulation tool. We consider an optical path
layout as in Fig.9 where an optical node is placed at the end of every
third fiber span of 50 kms to perform (in our case) power level
compensation due to the SRS effect. As such, an inter-node distance is
150 km. In [22] we reported a link of 75 channels (15 in each band) and
a path length of 450 and 900 km; here we assess the OSNIR
performance of a link with 155 channels across the E to L-band
spectrum and a total length of 1,200 km.
Each channel is operated at a 100G line-rate with 32 Gbaud source-
rate and a QPSK modulation format while the optical launch powers per
band are as of Table 3. The optical bandwidth for the estimation of ASE
noise is set to 32 GHz (equal to the baud rate).
Figure 9: A schematic representation of the abstraction model that
emulates the corresponding transmission system. SRS is compensated
at the Optical, WSS-based, Node.
With our closed-form models, the OSNIR values are directly deduced
from eq.(1) and eq.(2) with the addition of the SRS gain/loss
compensation at the optical nodes in regular intervals.
Table 3: Launch powers per band for a ‘flat’ OSNIR
E-band
(dBm)
S1-band
(dBm)
S2-band
(dBm)
C-band
(dBm)
L-band
(dBm)
-1.97
-6.84
-8.2
-9
-8.14
To deduce the OSNIR performance by means of the numerical
simulation tool, the chromatic dispersion and the polarization mode
dispersions are mitigated via the in-built Digital Signal Processing (DSP)
module at the receiver. Then the OSNIR values are deduced from the
corresponding BER values by means of the eq.(5) in [32]. Finally, the
results from the two simulation methods are illustrated in Fig.10.
Evidently, the abstraction method we propose in this work, that is based
on the closed-formed expressions, allows us to predict the transmission
performance of an optical multi-band transmission system with
sufficient accuracy.
Moreover, the proposed channel launch power optimization method
together with the SRS compensation per node does, indeed,
demonstrate a ‘flat’ OSNIR performance across the entire E-band to L-
band spectrum which means that the entire pool of channels (155
channels) from E to L band have the same optical reach.
4.2.3. Discussion: Setting different weights in eq.(2) for the importance of
a particular band, different band-management objectives are feasible. In
this way it is made possible to directly translate high-level network
management objectives to specific network operational parameters.
Regarding the flexibility of the two simulation methods, the closed-
form expression and the numerical tool, there is vast difference in the
computational time between the two methods: the execution time for
the abstraction method that is based on closed-form expressions is less
than second while the script from the numerical tool was running for
over two days. This confirms that in the context of MB-PCE operations,
only the former method may return results in a timely manner.
Given that experimental data at that scale we report here are not
currently available in the literature, a numerical simulation tool is a very
powerful asset to assess the performance in multi-bands transmission.
However, the computation power that is needed to complete multi-
band performance estimations becomes an increasingly harder
challenge to overcome. For example, we have been unable to complete
the benchmarking of a system with 305 channels and 900 km in length
as we reached the limit of our computing capabilities. Moreover, we
have excluded the O and U-bands from this study as given that a wider
simulation bandwidth is traded with the attainable number of channels,
the exploitation of these bands would limit the number of channels the
numerical simulation tool is able to process.
Figure 10: The OSNIR performance for a link of 1,200 km with 155
channels from the E-band to the L-band (100G per λ/QPSK).
5. THE SDN CONTROL-PLANE ARCHITECTURE
The integration of the MB-PCE to the SDN control-plane architecture
follows the approach of [33] for disaggregated optical networks. In this
case, the control and configuration of the network elements is carried
out by means of an Optical Multi-Band SDN (OMB-SDN) controller, that
orchestrates the data connectivity services.
This controller is deployed either as a single SDN controller or as a
dedicated controller for the optical line system (OLS) using, for example,
partial disaggregation [33].
As the MB-PCE shall carry out consecutive operations of considerable
computational complexity, the functions of the routing engine detailed
in sections 3 and 4 are externalized [34] as it is shown in Figs.1 and 11.
The OMB-SDN may request path computation services to the MB-PCE
via an open standard interface. Under this framework, a key challenge is
to retrieve the status of the network (the elements shown in Fig.3 as well
as the active services at any time) and to keep the MB-PCE updated. This
synchronization is implemented either by means of polling
(periodically, or on demand) or by means of streaming telemetry and
notifications [33].
Optical
Node
x Ns,L spans
50 km
x NLlinks
Tx .
.
.
Rx
.
.
.
TDFA
EDFA
TDFA
EDFA
BDFA
x Nch channels x Nch channels
OSNIR (dB)
Channel Index
Figure 11: SDN Control plane architecture for a disaggregated multi-
band optical network with externalized MB-PCE.
The OMB-SDN’s North Bound Interface (NBI) as well as the East
Bound interface to the MB-PCE follow the TAPI yang data models, as
detailed in the Reference Implementation Agreement (RIA) [35]. In
particular, for the aforementioned aspects of Network topology
definition these entail the parameters of section 3.1 like: the nodes and
their connectivity pattern, the number of edges, the associated links and
their MB amplifiers, along with the optical transmission system
parameters. Moreover, it includes the available optical bands, the size of
the FSU, the number of FSUs per band and the total capacity per band.
All these are included in the extended TAPI context as detailed next.
Figure 12: Sample of the TAPI extensions including specification of
amplifiers per-band information or terminal operation modes.
5.1. Transport API Extensions for the MB-PCE:
We developed the necessary extensions to the current ONF
Transport API v2.1.3 photonic media layer models to support the
dynamic provisioning of services of MB-WRN networks exploiting a
MB-PCE. The development of these extensions is a challenging task as it
requires to provide extensions a larger number of system parameters
like those listed in Fig.3 and the variables listed in Table 1. These
extensions address mostly the TAPI common, topology, connectivity,
and path computation modules (see Fig.12 for a selected fragment).
5.2. Control plane Provisioning Workflow
The operator requests a Digital Signal Rate (DSR) service (see Fig.13)
between transceiver’s client ports. The request contains the identifiers
of the Service Interface Points (SIPs); the requested bit rate (e.g., 100G,
200G, 400G) and any applicable routing and topological constraints. The
SDN controller delegates the computation to the externalized routing
engine of the MB-PCE that explores the PLI-aware RSMA platform of
Sections 3 and 4. Upon successful completion, it provides the number of
required OTSi and, for each OTSi, the path in terms of links, the number
of frequency slots and the selected transmission system parameters
(note that in this work we have considered single OTSi). The SDN
controller instantiates the DSR, OTSi and MC layers connection and
connection end point (CEPs) objects, it configures the transceivers’
operational modes and the media channels in the ROADM devices via
the SBI interface.
Figure 13: SDN workflow for the provisioning of a DSR connectivity
service. Note the intermediate request of the TAPI context (including
topology): this corresponds to the on-demand polling to synchronize
state between the Optical SDN controller and the MB-PCE.
6. EXPERIMENTAL VALIDATION OF THE SDN-BASED
AUTOMATION
SDN Agent / Node
Controller
Disaggregated
Node
Disaggregated
Node
TAPI v2.1
Digital Signal Service (DSR)
Service Provisioning
PointToPoint Connectivity TAPI DSR Connectivity Request
Optical Connectivity TAPI Photonic Media Layer (Network Media Channel).
SDN Agent / Node
Controller
e.g., OpenConfig
Model
OMB SDN Controller
Node Controller
Routing Engine
MB-PCE
Path Computation
e.g., OpenROADM
TAPI v2.1
Path Computation
SBI
NBI
SDN Agent /
Device Controller
Open Terminal
Device
SDN Agent /
Device Controller
Open Terminal
Device
USER SDN-CTRL Routing Engine
MB-PCE
TAPI 2.1.3 DSR Service provision
Input: SIPs, Bit Rate,
Topology Constraints
HTTP/1.1 POST TAPI 2.1.3 Path Computation
Input: SIPs, Bit Rate,
Topology Constraints
HTTP/1.1 POST
TAPI 2.1.3 Context retrieval
HTTP/1.1 GET
TAPI 2.1.3 Context (22 nodes,...)
HTTP/1.1 OK
TAPI 2.1.3 Path reply
HTTP/1.1 OK
TAPI 2.1.3 DSR Service Delete
HTTP/1.1 DELETE TAPI 2.1.3 Path Delete
Input: Path uuid
HTTP/1.1 POST RPC delete path
TAPI 2.1.3 Path Delete
HTTP/1.1 OK
TAPI 2.1.3 Service Delete
HTTP/1.1 OK
TAPI 2.1.3 Service
HTTP/1.1 OK
Path Computation Latency
Provisioning Latency
This section details the experimental evaluation of the SDN-enabled
MB-PCE, supported over a distributed control plane testbed that consist
of the CTTC optical SDN controller (with IP, e.g., 10.8.0.2) and the OLC-
E’s MB-PCE (with IP, e.g., 10.8.0.6) which are connected via tunnels over
the public Internet. The emulated network is BT’s optical mesh shown
in Fig.14 that consists of 22 ROADM nodes, 56 amplifiers, 28 terminal
devices (106 network elements in total) and 238 unidirectional links.
For the scope of this experiment, it is assumed that each link may
support E, S, C and L bands.
Figure 14. BT’s optical mesh with 22 ROADM nodes, 56 amplifiers, 28
terminal devices (106 network elements in total) and 238
unidirectional links
We have developed hypothetical scenarios where multiple
connection requests need to be provisioned between the node_ROADM1
and the node_ROADM 2 in Fig.15. To do so, the optical SDN controller
delegates the path computation task to the MB-PCE which, upon
request, retrieves (if needed) the network topology and the list of active
services/connections from the optical SDN controller. Subsequently, the
MB-PCE executes the PLI-aware RMSA algorithm and for each of these
requests it returns to the optical SDN controller the selected path as well
as the configuration details and their operational parameters for each
network element along this link.
In the following three scenarios, the routing engine of the MB-PCE is
instructed to check the potential of the available bands to support
service connection requests. For such scenarios, the MB-PCE is
instructed to iterate available bands with the same order of preference
as {C, E, L, S1, S2} so if the C-band is not available, the routing engine
proceeds to the second best which is the E-band and so on. We wish to
investigate the effectiveness of the MB-PCE and its behavior/response
against legitimate and illegitimate instructions from the OMB SDN
controller. In the latter case, the misconfiguration could be either due to
an erroneous data exchange between the MB-PCE elements or even the
result of deliberate/malicious acts.
6.1. Scenario A
In this first hypothetical case, we wish to investigate the impact of a
deliberate or accidental physical layer misconfiguration instruction on
the completion (or not) of service requests in the case of fully
transparent paths.
The 400G DSR service between two transceivers at the source-
destination nodes, node_101 and node_102 in Fig.15, is indicated by
means of the corresponding Service Interface Points (SIPs) and Fig.16
shows the provisioning of a connection request using the TAPI
connectivity model. The 400G line-rate is based on a 16QAM, 48 Gbaud
symbol-rate transceivers. We proceed to a deliberate misconfiguration
of PHY parameters, so the optimal launch power algorithm of section
4.2.1 is turned off. Then, the optical SDN controller sends service
requests, and it instructs the MB-PCE to launch the power of the C-band
channels at +1 dBm per channel while the channels in E-band are set to
optimal launch power. The selected powers render the C-band unusable
as the RMSA algorithm returns no route for the connection requests. As
a result, the algorithm proceeds to assess the potential of the E-band to
support the connection service (second band in the list) and it, indeed,
grants these requests to the E-band. It is critically important for network
operators that no misconfigured paths for the connection requests are
selected.
Figure 15: The different scenarios are considered requesting services between ROADM-1 and ROADM-2 using multiple transceivers.
Figure 16. Wireshark capture of the RESTCONF HTTP/POST operation
for the creation of the connectivity service.
6.2. Scenario B
In the second scenario, the deliberate misconfiguration of the PHY
parameters is less dramatic. The procedure of Scenario A is repeated
and the optical SDN controller sends sequentially 3 service requests but
now, the MB-PCE is instructed to launch the power of the C-band
channels at +0.6 dBm per channel (the E-band uses the optimal launch
power). As it is shown in Fig. 17, the first two requests are granted in the
C-band albeit the PLI-aware RMSA algorithm selects a larger guard-
band between the channels serving these two successive requests. In
particular, the two requests are allocated to channels spectrally space by
162.5 GHz (or 13 FSUs) which corresponds to over three times the
channel’s optical bandwidth. At such wide guard-band, the effect of XCI
is minimized.
However, the third request is not allocated to the C-band as the
overall efficiency in FSU utilization is low in this band under these
conditions. Therefore, the third connection request is allocated to the E-
band as it is illustrated in Fig. 17. It is pointed out that the distances
between node_101 and node_102 in Fig.15 are short raising no concerns
for the attainable transparent length in either band.
The important conclusion is that the PLI-aware RMSA can identify
operational parameters even under adverse conditions and is able to
‘switch’ between bands in the case of blocking.
Figure 17: Graphical representation of the spectrum usage of the link
between ROADM1 and ROADM2 showing the gap between services in
the C-band and a third service in the E-band.
6.3. Scenario C
In this third scenario, the same processes are repeated but this time
the optical SDN controllers sends to the MB-PCE, five service requests
sequentially designating a -3.7 dBm launch power for all of them which
is with acceptable range of powers (albeit non-optimal). Given the light
traffic load, the PLI-aware RMSA does not observe any OSNIR
degradation, so all service requests are granted spectrally successive
channels in the C-band. This is clearly illustrated in Fig. 18.
Also, of interest is to measure the response time (latency) needed to
complete the provisioning processes in the three scenarios. Fig.19
shows the exchange of messages between the optical SDN controller
and the MB-PCE. These messages include the messages sent by the
optical SDN controller to the MB-PCE related to the TAPI operations to
synchronize and update the MB-PCE on the current network status. It is
pointed out that as part of further optimization of the control-plane
operations, only the initial synchronization is mandatory; beyond that,
only the synchronization for the status of the active services is
necessary.
Figure 18. Graphical representation of the spectrum usage of the link
between ROADM 1 and 2, showing 5 services using the C band
This is understood as follows: as a service provisioning operation
may fail during the path computation phase, there might be a
misconception between the optical SDN controller and the MB-PCE for
the actual state of the network, a so regular message exchange between
the two entities is necessary.
Figure 19. Wireshark capture of the Path Computation exchange,
showing a path computation latency of approximately 2.5 seconds,
including the dynamic retrieval of the TAPI context from the controller.
The MB-PCE latency for scenario C was measured to be in the range
between 1.8 – 2.2 seconds. This value depends on whether the MB-PCE
is also tasked to retrieve the network topology as explained above (first
request in the series). On the contrary, this latency is two orders of
magnitude lower, ranging between 17ms and 36 ms, when the network
status was already up to date (assuming state synchronization). In
contrast, for scenarios A and B, the deliberate degradation of the OSNIR
rendered a large number of the listed frequency slots as
‘unavailable/void’, so the RMSA algorithm had to execute the PHY layer
validation process thousands of times resulting to considerable delays
for the MB-PCE to return results. For these two scenarios A and B, the
latency is measured to be between 2.5 and 3.2 seconds.
These results confirm the assumption of section 3 and 4, i.e., the
importance to incorporate to the routing engine a PLI-aware RMSA
based on closed-form expression as this is the only way to attain service
provisioning within a time frame compatible to the dynamicity of the
events that take place in the F6G transportation network. This is
particularly true as the projected capacity for the F6G transportation
network would entail several hundred connection requests. Moreover,
to ensure the scalability of the MB-PCE and the optical SDN controller, it
is crucial to use rapidly converging algorithms to determine the
optimization launch power per band or even per channel. Finally, the
resiliency of the MB-PCE against misconfigurations is proved.
6. CONCLUSION
The deployment of an SDN control plane is of clear interest for
network operators, notably in highly dynamic scenarios and as a key
enabler for network automation. Recently, the complexity of path
computation in optical networks (and, in particular, in multi-band
optical transport networks with non-negligible physical layer
impairments) has motivated the introduction of split architectures with
the decoupling of the path computation function. In this work, we have
introduced the design of such MB-PCE, validating the set of abstractions
that have been made, and showing its applicability in a few scenarios
with selected backbone networks. Regarding service provisioning, the
latency introduced is within accepted ranges, where hardware
configuration latencies are typically larger.
Acknowledgment. This work was supported by the EC H2020 B5G-
OPEN (101016663) as well as Spanish MINECO OPTRAN-CONTELEM
(TSI-063000-2021-22) funded by Spanish Ministry of Economic Affairs
and Digital Transformation and the Next Generation EU under the PRTR
programme, and RELAMPAGO grant PID2021-127916OB-I00 funded
by MCIN/AEI/10.13039/501100011033 and by ERDF A way of
making Europe.
References
1. A. Stavdas; "Networked Intelligence: A Wider Fusion of Technologies
that Spurs the Fourth Industrial Revolution", World Review of
Political Economy, Part I: Foundations, Vo.12 (2), p.220-235, 2021;
Part II: The Transformation of Production Systems, Vo.12 (2), p.236-
254, 2021
2. NetworldEurope; “Strategic Research and Innovation Agenda 2022
networks”. Online: https://www.networldeurope.eu/sria-2022-
announcement/
3. ITU-T G.Sup66 -201907, “5G wireless fronthaul requirements in
passive optical network context”, 07/2019
4. A. Stavdas; “Architectural solutions towards a 1,000 channel ultra-
wideband WDM network”. SPIE Optical Networks Magazine, Vol.2
No.1, pp. 51-60, 2001
5. T. Hoshida, V. Curri, L. Galdino, D. T. Neilson, W. Forysiak, J. K. Fischer,
T. Kato, and P. Poggiolini, “Ultrawideband systems and networks:
Beyond C and L-band,” Proc. IEEE 110, 1725–1741 (2022)
6. B. Ramamurthy, H. Feng, D. Datta, J. P. Heritage and B. Mukherjee;
"Transparent vs. opaque vs. translucent wavelength-routed optical
networks," Optical Fiber Communication Conference, 1999, San
Diego, CA, USA, 1999, pp. 59-61 vol.1
7. Spectral Grids for WDM Applications: DWDM Frequency Grid,
document ITU-T G.694.1, Oct. 2020. [Online]. Available:
https://www.itu.int/rec/T-REC-G.694.1/en
8. A. Lord, S. J. Savory, M. Tornatore and A. Mitra, "Flexible Technologies
to Increase Optical Network Capacity," in Proceedings of the IEEE,
vol. 110 (11), pp. 1714-1724, November 2022
9. A. Stavdas (Editor); “Core and Metro Networks”, John-Wiley (March
2010), ISBN: 978-0-470-51274-6.
10. D. Uzunidis, C. Matrakidis, E. Kosmatos, A. Stavdas, P. Petropoulos, A.
Lord, “Connectivity Challenges in E, S, C and L Optical Multi-Band
Systems”; European Conference on Optical Communications (ECOC)
2021, Bordeaux, 2021
11. P. Poggiolini et al., "Analytical and Experimental Results on System
Maximum Reach Increase Through Symbol Rate Optimization," in
Journal of Lightwave Technology, vol. 34, no. 8, pp. 1872-1885, 15
April15, 2016, doi: 10.1109/JLT.2016.2516398.
12. T. Gerard et al., "Relative impact of channel symbol rate on
transmission capacity," in Journal of Optical Communications and
Networking, vol. 12, no. 4, pp. B1-B8, April 2020, doi:
10.1364/JOCN.12.0000B1.
13. Laia Nadal, Michela Svaluto Moreolo, Josep M. Fàbrega, and F. Javier
Vílchez, "SDN-Enabled Multi-Band S-BVT Within Disaggregated
Optical Networks," J. Lightwave Technol. 40, 3479-3485 (2022)
14. E. Kosmatos, T. Orphanoudakis, C. Matrakidis, A. Stavdas and A. Lord;
" Switchless Elastic Rate Node (SERANO) Architecture: A Universal
Node for Optical Grooming and Adaptive Networking", IEEE/OSA
Journal of Optical Communications & Networking, Vol. 8(7), A162-
A170, 2016
15. D. Uzunidis, E. Kosmatos, C. Matrakidis, A. Stavdas and A. Lord,
"Strategies for Upgrading an Operator's Backbone Network Beyond
the C-Band: Towards Multi-Band Optical Networks," in IEEE
Photonics Journal, vol. 13, no. 2, pp. 1-18, April 2021.
16. IETF, “A YANG Data Model for Optical Impairment-aware Topology”,
version 12. Available at: https://datatracker.ietf.org/doc/draft-ietf-
ccamp-optical-impairment-topology-yang/12/ (March, 13, 2023).
17. L. Galdino et al., "Optical Fibre Capacity Optimisation via Continuous
Bandwidth Amplification and Geometric Shaping," in IEEE Photonics
Technology Letters, vol. 32, no. 17, pp. 1021-1024, 1 Sept.1, 2020.
18. B. J. Puttnam, R. S. Luís, G. Rademacher, Y. Awaji and H. Furukawa,
"319 Tb/s Transmission over 3001 km with S, C and L band signals
over >120nm bandwidth in 125 μm wide 4-core fiber," 2021 Optical
Fiber Communications Conference and Exhibition (OFC), 2021, pp.
1-3.
19. B. J. Puttnam, R. S. Luís, G. Rademacher, M. Mendez-Astudilio, Y. Awaji
and H. Furukawa, "S, C and Extended L-Band Transmission with
Doped Fiber and Distributed Raman Amplification," 2021 Optical
Fiber Communications Conference and Exhibition (OFC), 2021, pp.
1-3.
20. R. Casellas et all; “An SDN Control Plane for Multiband Networks
Exploiting a PLI-aware Routing Engine”. In Optical Fiber
Communication Conference 2022, W1F-2, 2022
21. Daniel Semrau, Robert I. Killey, and Polina Bayvel, "The Gaussian
Noise Model in the Presence of Inter-Channel Stimulated Raman
Scattering," J. Lightwave Technol. 36, 3046-3055 (2018).
22. D. Uzunidis, K. Nikolaou, C. Matrakidis, A. Stavdas, A. Lord; “Closed-
form Expressions for the Impact of Stimulated Raman Scattering
Beyond 15 THz”, European Conference on Optical Communications
(ECOC) 2022, Basel, 2022.
23. A. D’Amico et al., “Scalable and Disaggregated GGN Approximation
Applied to a C+L+S Optical Network,” Journal of Lightwave
Technology, vol. 40, no. 11, pp. 3499–3511, Jun. 2022, doi:
10.1109/JLT.2022.3162134.
24. D. Semrau, E. Sillekens, R. I. Killey and P. Bayvel, “A Modulation
Format Correction Formula for the Gaussian Noise Model in the
Presence of Inter-Channel Stimulated Raman Scattering,” in Journal
of Lightwave Technology, vol. 37, no. 19, pp. 5122-5131, 1 Oct.1,
2019.
25. C. Lasagni, P. Serena and A. Bononi, "Modeling Nonlinear
Interference With Sparse Raman-Tilt Equalization," in Journal of
Lightwave Technology, vol. 39, no. 15, pp. 4980-4989, Aug.1, 2021.
26. Alessio Ferrari, Mark Filer, Karthikeyan Balasubramanian, Yawei
Yin, Esther Le Rouzic, Jan Kundrát, Gert Grammel, Gabriele
Galimberti, and Vittorio Curri, "GNPy: an open source application for
physical layer aware open optical networks," J. Opt. Commun. Netw.
12, C31-C40 (2020)
27. Huaijian Luo, Jianing Lu, Zhuili Huang, Changyuan Yu, and Chao Lu,
"Optimization strategy of power control for C+L+S band
transmission using a simulated annealing algorithm," Opt. Express
30, 664-675 (2022).
28. F. Hamaoka, M. Nakamura, S. Okamoto, K. Minoguchi, T. Sasai, A.
Matsushita, E. Yamazaki, and Y. Kisaka, “Ultra-wideband WDM
transmission in S-, C-, and L-bands using signal power optimization
scheme,” J. Lightwave Technol. 37(8), 1764–1771 (2019).
29. B. Correia, R. Sadeghi, E. Virgillito, A. Napoli, N. Costa, J. Pedro, and V.
Curri, “Power control strategies and network performance
assessment for C+ L+ S multiband optical transport,” J. Opt. Commun.
Netw. 13(7), 147–157 (2021).
30. S. Kirkpatrick, C.D. Gelatt Jr and M.P. Vecchi, “Optimization by
simulated annealing”. Science, vol. 220, no. 4598, pp. 671-680,
May 13, 1983.
31. A. Ferrari et al., "Assessment on the Achievable Throughput of Multi-
Band ITU-T G.652.D Fiber Transmission Systems," in Journal of
Lightwave Technology, vol. 38, no. 16, pp. 4279-4291, 15 Aug.15,
2020, doi: 10.1109/JLT.2020.2989620.
32. D. Uzunidis, C. Matrakidis, and A. Stavdas, "Application of a simplified
FWM expression in mixed-fiber links," in 2016 24th
Telecommunications Forum (TELFOR), 2016.
33. R. Casellas et al., "Advances in SDN control and telemetry for beyond
100G disaggregated optical networks [Invited]," in Journal of Optical
Communications and Networking, vol. 14, no. 6, pp. C23-C37, June
2022, doi: 10.1364/JOCN.451516.
34. K. Ishii et al., "Two-Level Abstraction Approach for SDN-based
Service Provisioning in Open Line Systems Featuring TAPI
Externalized Path Computation," 2020 European Conference on
Optical Communications (ECOC), Brussels, Belgium, 2020, pp. 1-3,
doi: 10.1109/ECOC48923.2020.9333136.
35. R. Casellas, A. Mazzini, N. Davis, editors, ONF TR-547 v1.1, "TAPI
Reference Implementation Agreement", December 2021, online
https://opennetworking.org/wp-content/uploads/2021/12/TR-
547-TAPI_ReferenceImplementationAgreement_v1.1.pdf