Document template for MAINS - CORDIS · Web viewAt low frequencies, the driving voltage can be...

IST IP IDEALIST (Industry-Driven Elastic and Adaptive Lambda

Infrastructure for Service and Transport Networks)

Elastic Optical Network Architecture: reference

scenario, cost and planning

D1.1 - Elastic Optical Network Architecture: reference scenario, cost and planning

Status and Version: Elastic Optical Network Architecture: reference scenario, cost and planning. Draft 2

Date of issue: 24.6.2013

Distribution: Project Internal

Author(s): Name Partner

Andrew Lord (Editor) British Telecom

Juan Fernandez-Palacios Telefonica I+D

Oscar González Telefonica I+D

Victor Lopez Telefonica I+D

Luis Velasco UPC

Jaume Comellas UPC

Gabriel Junyent UPC

Marco Quagliotti TI

Paul Wright BT

Aristotelis Kretsis University of Patras

Polyzois Soumplis University of Patras

Emmanouel (Manos) Varvarigos University of Patras

Kostas Christodoulopoulos University of Patras

Annalisa Morea Alcatel Lucent

Alexandros Stavdas UoP

Daniel Fonseca Coriant

Ori Gerstel Cisco

Matthias Gunkel Deutsche Telekom

Michael Parker Lexden Technologies

Norberto Amaya-Gonzalez University of Bristol

Alexandros Stavdas University of Peloponese

Michela Svaluto CTTC

Checked by: Juan Pedro Fernandez-Palacios TID

Page 1 of 99





Abstract

D1.1 of Idealist covers the initial ground on all of the relevant sub topics within WP1 of Idealist. This includes operator reference networks and static traffic matrices for analysis. A summary of the state of the art for relevant flexgrid algorithms and definition of future work is provided as well as discussion of plans for the prototype planning tool. There is a definition of the CAPEX and OPEX models (including operational and power consumption considerations) to be used in the techno-economic analysis of the different data and control plane alternatives identified in WP2 and WP3. The deliverable also covers the key Use Cases so far identified for extensive modelling.

Page 2 of 99





Contents1 EXECUTIVE SUMMARY....................................................................................................................... 5

1.1 Purpose of Idealist.............................................................................................................................51.2 Methodology.....................................................................................................................................51.3 Benefits of EON: Use Cases...............................................................................................................51.4 Reference Networks..........................................................................................................................81.5 Techno-economic modelling.............................................................................................................91.6 Network Planning, modelling and optimisation..............................................................................111.7 Conclusions and further work.........................................................................................................12

2 INTRODUCTION................................................................................................................................ 14

2.1 Purpose and Scope..........................................................................................................................142.2 Reference Material..........................................................................................................................14

2.2.1 Reference Documents.................................................................................................................142.2.2 Acronyms....................................................................................................................................182.2.3 Definitions...................................................................................................................................19

2.3 Document History...........................................................................................................................19

3 NETWORK ARCHITECTURE AND USE CASES.......................................................................................20

3.1 Rationale behind Elastic Optical Networks......................................................................................203.2 Use cases for Elastic Optical Networks............................................................................................24

3.2.1 ST-1: Multi-Layer Restoration......................................................................................................243.2.2 ST-2: IP over Elastic/FixRate optical networks.............................................................................263.2.3 ST-3: Flexgrid Optical Networks for DataCentre Federations......................................................273.2.4 ST-4: The disaster recovery (DR)..................................................................................................293.2.5 ST-5: Flexgrid in metro-regional networks - serving traffic to BRAS servers................................30

3.2.5.1 Current Metro architecture................................................................................................................303.2.5.2 Scenario A: evolutionary approach....................................................................................................313.2.5.3 Scenario B: Fixed-based approach with adaptation layer...................................................................313.2.5.4 Scenario C: Flexgrid-based approach with Sliceable BVTs at the remote BRAS..................................323.2.5.5 Evaluation of the Scenario C use-case................................................................................................33

3.2.6 ST-6: Scalable core networks with Architecture on Demand nodes.............................................33

4 REFERENCE NETWORKS.................................................................................................................... 36

4.1 Introduction....................................................................................................................................364.2 Telecom Italia National Reference Core Network...........................................................................374.3 Deutsche Telekom national reference IP core network..................................................................38

4.3.1 Topology and Network Architecture...........................................................................................384.3.2 A and B network split..................................................................................................................394.3.3 Traffic matrix and forecast..........................................................................................................40

4.4 Telefónica National Transport Network..........................................................................................404.5 British Telecom Reference Networks..............................................................................................424.6 Telecom Italia Sparkle European Network......................................................................................434.7 Summary.........................................................................................................................................44

5 TECHNO-ECONOMIC ANALYSIS......................................................................................................... 46

5.1 Introduction....................................................................................................................................465.2 CAPEX Model...................................................................................................................................46

5.2.1 Summary of STRONGEST model..................................................................................................465.2.2 Idealist CAPEX model..................................................................................................................47

5.3 Target cost for Sliceable Bandwidth Variable Transponders...........................................................57

Page 3 of 99





5.3.1 Case study definition...................................................................................................................575.3.2 Case study results........................................................................................................................59

5.4 OPEX model.....................................................................................................................................615.4.1 Cost of floor space.......................................................................................................................625.4.2 Field operations and repair model...............................................................................................62

5.5 Energy model for high data rate transponders................................................................................645.5.1 Energy model for fixed data rate devices....................................................................................645.5.2 Energy model for elastic transponders........................................................................................70

6 NETWORK PLANNING....................................................................................................................... 74

6.1 State of the Art................................................................................................................................746.1.1 Clustering of nodes for hierarchical traffic grooming..................................................................746.1.2 Off-line RSA models.....................................................................................................................756.1.3 Dynamic RSA...............................................................................................................................756.1.4 Spectrum Reallocation................................................................................................................766.1.5 Elastic Spectrum Allocation for Variable Traffic..........................................................................766.1.6 Available Planning tools..............................................................................................................77

6.2 Algorithms for network planning.....................................................................................................786.2.1 Algorithms for off-line network planning....................................................................................786.2.2 Algorithms for network clustering...............................................................................................806.2.3 Transmission configurations selection and RSA under physical layer impairments.....................826.2.4 Specifically-designed recovery for Flexgrid..................................................................................836.2.5 Large-scale optimization techniques...........................................................................................836.2.6 Algorithms for in-operation network planning............................................................................83

6.3 Architecture of network planning tool............................................................................................856.3.1 Off-line network planning tool (MANTIS)....................................................................................866.3.2 In-Operation network planning tool (PLATON)............................................................................88

6.4 Problems to be implemented in PLATON........................................................................................916.4.1 Single Layer Flexgrid Network Design Problem...........................................................................916.4.2 After Failure Repair Optimization (AFRO)....................................................................................926.4.3 Spectrum defragmentation (SPRESSO)........................................................................................92

7 CONCLUSIONS.................................................................................................................................. 94

8 Annex 1 - Detailed reference network information.................................................................................96

Page 4 of 99





1 Executive summary

1.1 Purpose of IdealistElastic optical networks are more flexible than existing, fixed alternatives: they offer the scope to use different signal modulation formats and different spectrum allocations, even in a dynamic way. The flexible options being discussed are varied and as such EON covers a broad range of solutions ranging from mixed line rates (MLR) over fixed grid to sliceable bit rate variable transponders (SBVTs) over a fully flexible optical spectrum.

The Idealist EU project intends to find out if EONs can be beneficial to carriers, and if so, under which network scenarios and applications or Use Cases. Additionally, Idealist will pinpoint the optimum EON for each case, quantifying how much benefit will be gained, in terms of CAPEX and a range of OPEX measures. The final outcome of Idealist will be a clear recommendation of the value of EON, the most fruitful situations to consider using it and the actual benefits in doing so.

D1.1 is the first deliverable from Work Package 1 of Idealist and as such, sets out the roadmap to reach this goal. In this Executive Summary, a complete overview of the deliverable is presented, including methodology, work accomplished so far, and outlook.

1.2 MethodologyTo carry out its plan, Idealist is adopting the following methodology:

(i) Seek the most likely opportunities in carriers’ networks, known as Use Cases

(ii) Define a range of national and international reference networks including traffic profiles (both static and dynamic).

(iii) Develop techno-economic modelling facilities to enable both CAPEX and OPEX-based assessment of each of the solutions in a comparative way.

(iv) Development of a wide range of modelling tools to plan the network off-line and provision resources in real time. In this way, both the elastic and non-elastic approaches will be compared fairly.

(v) Bring (i) – (iv) together in comprehensive modelling to do a broad technology comparison between EON, flexgrid and more conventional alternatives.

1.3 Benefits of EON: Use CasesAlthough there is perhaps an acceptance that EON has benefits, there is a debate about whether a full flexgrid implementation is required to achieve those benefits. Given this, a key, current industry debate relates to when is the most appropriate time for carriers to: (a) install flexgrid ready components, and (b) enable them and start using the technology. The key flexgrid components are the liquid-crystal on silicon (LCoS) based wavelength selective switches (WSS’s), which allow arbitrary spectrum demarcation, although they can be readily used in fixed grid mode, and the flexgrid capability can be software enabled (and paid for!) at a later date, when required. This is a useful alternative for carriers, who might only refresh their main transmission infrastructure every few years.

Page 5 of 99





An alternative strategy is to continue with a fixed grid 50 GHz solution but use mixed line rates and higher modulation formats to carry demands. For high rates of 400 Gb/s and above, the fixed grid option requires the use of inverse multiplexing, in which the total rate is split into smaller chunks. For example, 400 Gb/s could be transmitted as 100 Gb/s, DP-QPSK in four, not necessarily adjacent fixed grid spectrum slots.

It is commonly accepted though, that eliminating the arbitrary 50 GHz boundaries, allow a more efficient use of spectrum, and this results in an increase in network capacity. One early Idealist study, based around the Spanish network, shows the impact on increasing capacity, in terms of delaying further network build. The key result, in the figure below, shows that fixed grid upgrades are required in 2019, whereas flexgrid can support traffic growth until 2024. If current fixed grid networks have sufficient capacity until 2019, this suggests that although flexgrid is seen to have significant advantages, there isn’t an immediate need for it, and so flexgrid ready components could be installed when carriers next refresh their DWDM capability.

2014 2016 2018 2020 2022 20240

10

20

30

40

50

WSON Case FLEXI-GRID Case

num

ber o

f new

Fib

er L

inks

(cum

mul

ated

)

Year

Figure : Number of new Fiber Links WSON vs. flexgrid evolution models in Telefonica Spain reference network

However, one issue hidden by this result is the number of transponders required, and this becomes significant in fixed grid, where large demands have to be inverse multiplexed as described earlier. This implies that there is a cost impact in delaying the move to flexgrid, especially if we start to see the development of cost effective 400 Gb/s and 1 Tb/s transponders.

This example shows the complexities of the question, the answer to which depends on the specific scenarios and solutions being compared. Consequently it is impossible to get a definitive answer in the short published papers usually published in the journals or conferences.

The only sound way to tackle a question as complex as this, is to do it within a large, multi-partner project such as Idealist. In this context, it is possible to construct a full range of scenarios, reference networks, EON variants, planning and operational algorithms, techno-economic models – and then combine then to thoroughly compare the options and draw clear conclusions. One key element is the definition of Use Cases that span the full range of applications of potential interest. This section will summarise these Use Cases.

Page 6 of 99





A range of Use Cases have been compiled, with extensions being added as the project progresses. The existing Use Cases fall broadly into two categories – medium and longer term. The following table summarises them and indicates the status of each one.

Use-Case Summary Timescale Status Contributor

Multi-Layer Restoration

Resilience mechanisms relying on the IP layer exclusively are not efficient.

Restoration is a multi-layer problem to be triggered from IP routers and their TE functionality.

The introduction of CoS in an EON allows to adapt the line-rate to a given restoration path and allow a fast recovery of the high-priority traffic

ST To be studied in WP1

DT

IP-o-EON

In an IP-o-EON, multi-layer planning and operation is an essential feature.

A joint optimization of the packet flow in both, the working and the backup paths over the IP/MPLS layer in association to RSA and CoS differential is an important design asset.

ST To be studied in WP1

ALU

Federated Data-Centres based on EONs

Federated Data-Centres is emerging as an important part of the IT infrastructure.

Static bandwidth provisioning is using inefficiently in a high-CAPEX infrastructure.

EONs facilitate to dynamically adjust bandwidth based on the actual traffic flow needs in real time.

ST Ongoing study

UPC

EONs in disaster recovery

Disaster recovery mandates a flexible and reconfigurable network to provide maximum (but typically not full) service recovery.

EONs play a critical role in DR plans since the BVTs allows one to optimize the line-rate for a given distance, and they provide cost benefits compared to fixed-grid solutions with given

ST Ongoing study

CISCO

Page 7 of 99





regenerator placement.

EON in Metro Networks

IP functionality is implemented in the Broadband Remote Access Servers (BRAS) which are usually located at the second level of aggregation

The introduction of the flexgrid approach in the Metro Area Networks (MAN) would support support BRAS centralization and hence IP router machinery reduction.

The work would compare a flex-grid solution in a realistic metro-regional network scenario to traditional approaches based on fixed-grid WDM systems that are used today

ST Ongoing study

TID

Scalable core networks with Architecture on Demand nodes

The current WSS solutions have scalability problems

The AoD concept is proposed where an optical backplane facilitate to obtain a node architecture where the specific functionality comes from the optical sub-systems that are selected at will.

LT Ongoing study

UoBristol

1.4 Reference NetworksThe general characteristics of the Reference Networks collected in WP1 are summarized in the first table below. At the present stage the IDEALIST project relies on six networks of different types and geographic scopes, from Nationwide to European Continental. All the networks are available with topological details and often with the characteristics of the fibers and other valuable features (for instance optical amplifier positions and span length) that allows us to perform an accurate network design. Topological features like node degree and link length (average and maximum values) are reported in the second table below. For most of the reference networks the traffic demand is also available but limited to the static version, as this is related to information from transport networks today. Dynamic traffic can be generated by traffic engineering modelling, whilst waiting for real data that is expected to be collected during the progress of the project.

Main Features of Reference Networks

Operator Location Segment covered Main features

TI Italy Core (National)Flat National 44 nodes network, mainly but not exclusively for carrying IP backbone traffic; mainly G655 and G652 and few G653.

DT Germany Core (National) Flat 12 PoPs National Core, physically installed twice (12+12 nodes) to serve exclusively the IP core network, fiber is wholly

Page 8 of 99





G.652.

TID Spain Core (National part) Two levels National optical network, 30 nodes National Core, G.652 fiber.

TID Spain Core (Regional part) 5 Regional networks (30 nodes each); G.652 fiber everywhere.

BT Great Britain Core, Metro and Aggregation

1113 nodes network connected by a G652 fibre infrastructure. No inherent hierarchy but sites classified as Core, Metro or Aggregation.

BT Great Britain Core (National) 22 nodes flat core network with G652 fiber links.

TI Europe Core (Continental) Flat 49 nodes network. Fibers on the links are G.652 or G.655.

Topological characteristics of Reference Networks

Operator Location Nodes Nodal degree Links Link length [km]average max average max

TI Italy 44 3.2 5 70 174 482DT Germany 12 3.3 5 20 243 485

TID Spain (National) 30 3.7 5 56 148 313

TID Spain(5 Regions)

30(each Reg.) 3.5 5 53 73 185

BT Great Britain 1113 3.5 17 1956 24 295BT Great Britain 22 3.2 4 35 147 686TI Europe 49 2.9 5 69 393 1212

1.5 Techno-economic modellingIdealist has taken as its starting point for techno-economic analysis, the CAPEX model constructed in the STRONGEST project. This contains cost figures for a wide range of transport functions and allows a good costing exercise for an existing network. In D1.1 this is updated with cost information for Layer 3 components and there has been a change of cost baseline from 10 Gb/s non-coherent to 100 Gb/s coherent transponder.

The main techno-economic interest for Idealist is to achieve a detailed cost comparison of flexgrid / elastic networks as compared to fixed grid. This suggests that most of the activity will be focused on the optical layer, but it is felt strongly that the IP client layer will have a significant role to play. This will particularly be the case when we consider IP-over-flexgrid type architectures, potentially using (Sliceable) Bit Rate Variable Transponders, which give the opportunity to share IP bandwidth very flexibly.

The modelling here also assumes that technology will move forwards and also reduce in cost as it sells in volume. To this end, Idealist is focusing on 3 timeframes – 2013, 2015 and 2018. Cost estimates beyond this time are impossible. Transponders with data rates from 100 Gb/s to 1 Tb/s are assumed to appear as this timeline unfolds, initially with fixed bit rate, but eventually with bit rate variability and ultimately sliceability emerging.

The table below gives the first glimpse of the cost figures for flexgrid transponders, although, as seen from the table, there is clearly a lot more work to do in conjunction with WP2, who are actually developing the flexible transponders alluded to here.

Page 9 of 99





Bandwidth Variable Transponders in flexgridInterface type Specification Available Cost (ICU) Required slot

Transponder 1 100G, 50GHz, 2000km AUA* 40G, 50GHz, 2500km 2013 1.44 2

Transponder 2 400G, 75GHz, 500km AUA 200G, 75GHz, 2000km AUA 100G, 75GHz,

2500km 2015 1.76 4Transponder 3 1000G, 175GHz, 500km AUA 500G,

175GHz, 2000km 2018 2.00 6Transponder 400G 400G, 100GHz, 1000km 2018 1.20 2100G Muxponder, 2 x 40G + Transponder1100G Muxponder, 10 x 10G + Transponder1

Specification Available Cost (ICU)Required slot

400G Muxponder, 10 x 40G + Transponder2

T.B.D. 2015 T.B.D. T.B.D.

400G Muxponder, 4 x 100G + Transponder2

T.B.D. 2015 T.B.D. T.B.D.

400G Muxponder, 10 x 40G + Transponder 400G

T.B.D. 2018 T.B.D. T.B.D.

400G Muxponder, 4 x 100G + Transponder 400G

T.B.D. 2018 T.B.D. T.B.D.

* AUA = Also Useable As

The deliverable also provides the basis for a range of OPEX related models that will be further developed in IDEALIST. These include cost of accommodation / floor space, cost of operations and repair in the field, cost of maintenance including spares, and the cost of using energy. With respect to energy usage, D1.1 presents an in-depth analysis of the energy consumption in various kinds of fixed and flexgrid transponders – information that will be essential to allow fair comparison between them. An example of the kind of modeling and analysis attempted is given in the table below which shows energy consumption values for 400 Gb/s and 1 Tb/s transponders based on coherent technologies, including details of the components required to make them.

Component

400Gb/s transponder 1Tb/s transponder

UnitPower

consumption (W)

UnitPower

consumption (W)

Client sideClient card (@10Gb/s) 4 24 10 24

Framer/Deframer 1 100 1 200

E/O modulation

Drivers 2x4 2 4x4 2

Laser 2x1 6.6 4x1 6.6

O/E receiver

Local oscillator 2x1 6.6 4x1 6.6

Photodiode +TIA 2x4 0.4 4x4 0.4

Page 10 of 99





ADD 2 80 4 90

Management power 20% total power 20% total power

1.6 Network Planning, modelling and optimisationWP1 has a strong algorithm and simulation focus, with a wide range of tools available and under development to solve a wide range of problems. The key problems to be solved here relate to the assignment of optical spectrum across specific network routes to meet traffic demands. This Routing and Spectrum Assignment (RSA) is an NP complete problem, and therefore highly computationally intensive. It therefore requires carefully crafted heuristic approaches to give solutions in realistic time scales, especially as the number of network nodes increases.

RSA divides into two distinct categories of problem:

Off-line planning. Here there is no requirement for decisions in real time. We have a network with some traffic requirements and we wish to optimise the location of equipment and the routing and spectrum allocations for the various traffic demands. Time can be taken to explore different scenarios to yield the best solution for the given requirements.

On-line modelling. Here, the network is up and running, and new real-time RSA decisions need to be made. Time is limited, and only one or a small number of new demands are to be scheduled.

One critical issue that determines which category of problem is required, relates to the dynamicity of the traffic. Static traffic works well with slower algorithms, although even static traffic grows and provisioning the new circuits brings in an element of dynamics, albeit small. Dynamics can of course arise from quite natural causes, such as time of day traffic variations and restoration following link or node failures. Often these variations cause the traffic to either grow or reduce everywhere simultaneously, and so don’t necessarily require RSA operations.

It is fair to say that carriers are not currently experiencing traffic that is dynamic enough on the timescales that would impact on the optical layer, but a future dominated by cloud, data centre and ultra-high definition TV services could change that situation, or at least modify it. Nevertheless, Idealist is assessing the benefits that an elastic optical network can bring to dynamic Use Cases, and this requires development of suitable algorithms.

Looking beyond the nearer term Use Case opportunities, an important flexgrid research question involves looking at End-to-End architectures and re-optimising them for flexgrid. This is important because flexgrid might perform best in flatter architectures, because it is better able to handle a large range of optical demands. An early part of this architectural study involves segregation of the whole network into metro clusters interconnected via core nodes. Other early work is looking at incorporating physical layer limitations (both noise / reach and nonlinearity related) into the RSA algorithms. Also, a study of large-scale optimisation techniques has begun, to deal with the extremely high numbers of variables in some of the large problems involving many degrees of freedom and many nodes. Finally, a range of approaches to solving the online problems has been initiated – ranging from real time RSA through to defragmentation problems.

Page 11 of 99





Simulation tools are being developed to handle this huge computational challenge in a robust way. There are two distinct tools to handle the offline and online processing paradigms:

MANTIS. Predominantly offline processing tool. Provides a repository for networks and algorithms – thus allowing easy comparison of different approaches and a ready-made benchmarking tool. Potential availability as a Cloud service via a web interface.

PLATON. Online processing tool. Addressing problems such as flexgrid design, post-repair optimisation and spectrum defragmentation.

Whilst a version of MANTIS existed pre-Idealist, PLATON is a new tool, being developed within the Idealist project to specifically address problems requiring real time decisions. The following table shows the intended development plan for PLATON.

Task %Done May

20

13Ju

n 20

13

Jul 2

013

Aug

2013

Sep

2013

Oct

201

3

Nov

20

13Ye

ar #

2

Year

#3

PLATON 18% Cluster Manager 60% Requests Database 100% Manager 100% Management Web Server 100% Web-services Server 0% PCE Server 0% HPC Agent 13% Communications 0% Optimization Framework 25% Deliverable D1.2 0% Algorithms year #2 17%

Single Layer Flexgrid Network Design Problem 50%

After Failure Repair Optimization (AFRO) 0% Spectrum Defragmentation (SPRESSO) 0% Algorithms year #3 0% Define Algorithm Set 0% Implement Algorithms 0%

1.7 Conclusions and further workThe first deliverable has provided all of the information needed to carry out a thorough investigation into the benefits of flexgrid and more generally Elastic Optical Networking. Reference networks, applications or Use Cases, modelling tools and algorithms and techno-economic data are all provided here.

Of course these project inputs will, to some extent, require refinement. Although the reference networks will remain stable now for the remainder of the project, there will be an on-going adjustment to the Use Cases. It is likely that further Use Cases will be added to the list, as the project digests the results of the simulations and modelling.

Regarding CAPEX, the status is mature, at least regarding existing network technologies. But it will be required to describe and to assign the cost parameters to new devices when they will be more clearly defined (SBVTs and flexgrid node parts, especially the ones on A/D side allowing the connection with SBVTs). In addition it could be useful to add to the

Page 12 of 99





model an integrated OTN with fixed grid WDM as a competitor to the all-optical flexgrid option.

Regarding OPEX, there is clearly a great deal more that can be done on all of the OPEX related aspects, from automation to energy consumption and space saving. Energy consumption has an impact both from a network architecture perspective, but also at the level of individual components: energy savings should come directly from the additional flexibility provided. There has already been strong input on energy savings to be expected from adaptive data-rate transponders, but further work is expected here.

Finally, the scene has been set for a great deal of intensive algorithm-based work to provide the key tools to carry out the comparisons required throughout Idealist. One area of particular interest is the research on large-scale optimisation, which stands to shed light on the largest problems, e.g. modelling BT’s huge network of over 1000 nodes, and with many other simultaneous variables to be optimised.

EON and flexgrid offer the potential for significant bandwidth efficiency increases and this will enable operators to squeeze significantly more life out of their network infrastructures – one estimate presented here suggests up to 5 years. Although this is a large benefit, and although all carriers are experiencing continuing large traffic volume increases, the non-elastic fixed grid solutions can meet these capacity needs for several years to come. In the meantime, there is a lot of discussion about when carriers should make their networks EON-ready and how they should migrate.

Additionally, there is increasing thought being given to other benefits that might emerge, as well as the basic capacity boost. Already in Idealist, we are seeing interest crystallising in the nearer term towards multi-layer protection applications, and in the longer term towards the use of technologies such as the sliceable bit rate variable transponder. The current perception is that elastic and flexible concepts have much to offer, but we need to continue the creative and analytical processes to find the application sweet spots.

For example, with SBVTs, a comprehensive techno-economic study is required, both for the longer term dynamic traffic scenario, but also for a nearer term more pragmatic case in which the traffic grows, but in a predictable, relatively static way. Are BVTs useful – and if so, what are the additional advantages of sliceability? The SBVT architecture (including the structure of the device and the architecture of the application in which it will be used) is still in the process of being properly defined, and part of that exercise involves interworking with WP2.

Overall, EON has strong potential to provide capacity increase, flexibility to future traffic dynamics and a reduction in equipment costs and energy consumption. Which version of EON could achieve these advantages, and which carrier applications are most applicable in the medium and long terms, will require the detailed study upon which WP1 is now embarked.

Page 13 of 99





2 Introduction

2.1 Purpose and ScopeThis is the first deliverable from Work Package 1 of Idealist. A succinct Executive Summary has been provided with this deliverable, which gives an overview of the entire document, and should be useful for those seeking a bird’s eye view. The main document is broken into the same logical sections that the work package itself is using:

Architectures and Use Cases

Reference Networks and traffic

Techno-economic modeling

Network planning and operation algorithms

The project has been running for 8 months, and this first deliverable sets the scene for the remainder of the project. It presents the basic areas for study, the networks that will be used, and the wide range of approaches in terms of planning, online operation and finally the various CAPEX and OPEX related tools to enable comparisons.

2.2 Reference Material

2.2.1 Reference Documents

[1] A. Mokhtar and M. Azizoglu, “Adaptive wavelength routing in all-optical networks” IEEE/ACM Transactions on Networking, vol. 6, no. 2, pp. 197-206, April 1998

[2] Gunkel, Autenrieth, Neugirg, Elbers: “Advanced multilayer resilience scheme with optical restoration for IP-over-DWDM core networks - how multilayer survivability might improve network economics in the future”, International Workshop on Reliable Networks Design and Modeling (RNDM), 03.10.2012, St. Petersburg, Russia, 2012

[3] Autenrieth, Elbers, Eiselt, Grobe, Teipen, Grießer: “Evaluation of technology options for software-defined transceivers in fixed WDM grid versus flexible WDM grid optical transport networks”, ITG-Fachtagung Photonische Netze, 07.05.2013, Leipzig, Germany, 2013

[4] P. Winzer, IEEE Commun. Mag., 7:48 (2010)

[5] A. N. Patel, C. Gao, J. P. Jue, X. Wang, Q. Zhang, P. Palacharla, and T. Naito, “Traffic grooming and regenerator placement in impairement-aware optical WDM networks”, in Proc. ONDM 2010, Kyoto, Japan

[6] A. Autenrieth et al, Proc. DRCN ‘03, pp. 333-340 (2003)

[7] S. L. Woodward, Proc OFC ‘10, PDPC8 (2010)

[8] M. Armbrust, et al., “A View of Cloud Computing,” Communications of the ACM, vol.53, pp. 50-58, 2010

Page 14 of 99





[9] M. Mishra, A. Das, P. Kulkarni, and A. Sahoo, “Dynamic Resource Management Using Virtual Machine Migrations”, IEEE Communications Magazine, vol. 50, pp. 34-40, 2012

[10] I Goiri, F. Julià, R. Nou, J.L. Berral, J. Guitart, and J. Torres, “Energy-aware scheduling in Virtualized Datacenters”, in Proc. IEEE International Conference on Cluster Computing, 2010

[11] I. Goiri, J. Guitart and J. Torres, “Characterizing Cloud Federation for Enhancing Providers’ Profit”, in Proc. IEEE International Conference on Cloud Computing, 2010

[12]Cisco, Global Cloud Index, 2012

[13]C. Liou, O. Turkcu, V. López, J.P. Fernández-Palacios, “An Economic Comparison of Cloud Network Architectures” in Proc. PTC 2013

[14]T.Tanaka, A. Hirano, and M. Jinno, "Performance Evaluation of Elastic Optical Networks with Multi- Flow Optical Transponders," in European Conference and Exhibition on Optical Communication, OSA Technical Digest (online) (Optical Society of America, 2012), paper Tu.3.D.2.

[15]K. Christodoulopoulos, P. Soumplis, E. Varvarigos, "Trading off Transponders for Spectrum in Flexgrid Networks", Conference on Optical Fiber Communications (OFC), 2013

[16]K. Christodoulopoulos, P. Soumplis, E. Varvarigos, "Planning Flexgrid Optical Networks under Physical Layer Constraints", submitted to IEEE/OSA JOCN

[17]F. Rambach et al,. “A Multilayer Cost Model for Metro/Core Networks”, J. OPT. COMMUN. NETW./VOL. 5, NO. 3/MARCH 2013, pp. 210-225

[18] ITU-T Recommendation G.694.1: Spectral grids for WDM applications: DWDM frequency grid.

[19]B. Collings, “New Devices Enabling Software-Defined Optical Networks”, IEEE Communications Magazine, March 2013, pp. 66-71

[20]FP 7-STRONGEST project, D2.1 “Efficient and optimized network architecture: Requirements and reference scenarios”. Available at http://www.ict-strongest.eu

[21] Information Resources Management Association, Networking and Telecommunications: Concepts, Methodologies, Tools and Applications.: IGI Global, 2010.

[22]N. Amaya, G. Zervas, and D. Simeonidou, "Introducing Node Architecture Flexibility for Elastic Optical Networks," J. Opt. Commun. Netw. 5, 593-608 (2013).

[23]K. Roberts, D. Beckett, D. Boertjes, J. Berthold, and C. Laperle, “100G and Beyond with Digital Coherent Signal Processing,” IEEE Communication Magazine, Vol. 48, No. 7, pp 62-69, July 2010

[24]G. Bosco, V. Curri, A. Carena, P. Poggiolini, and F. Forghieri, “On the Performance of Nyquist-WDM Terabit Superchannels Based on PM-BPSK, PM-QPSK, PM-8QAM or PM-16QAM Subcarriers,” in Journal of Lightwave Technology, Vol. 29, No. 1, 99. 53-61, January 2011.

[25]Y. Miyata, W. Matsumoto, H. Yoshida, and T. Mizuochi, “Efficient FEC for Optical Communications using Concatenated Codes to Combat Error-floor,” in proceedings of OFC 2008, paper O.Tu.E.4.

Page 15 of 99

http://www.ict-strongest.eu/





[26]A. Morea, S. Spadaro, O. Rival, J. Perello, F. Agraz, D. Verchere, “Power management of optoelectronic interfaces for dynamic optical networks,” in Proceedings of IEEE ECOC 2010, paper Th.10.F.5.

[27]Cisco, data sheet available on http://www.cisco.com

[28]A. Morea, O. Rival, N. Brochier, and E. Le Rouzic, “Datarate Adaptation for Night-Time Energy Savings in Core Networks,” in Journal of Lightwave Technology, Vol. 31, No. 5, pp. 779-785, March 2013.

[29]F. Vacondio, A. El Falou, A. Voicila, C. Le Bouëtté, J.-M. Tanguy, C. Simonneau, E. Dutisseuil, J.-L. Pamart, L. Schoch, and O. Rival, “Real-Time Elastic Coherent Muxponder Enabling Energy Proportional Optical Transport,” in Proceedings of IEEE/OSA OFC 2013, paper JTh2A.51.

[30]T. D. Burd, T. A. Pering, A. J. Stratakos, and R. W. Brodersen “A dynamic voltage scaled microprocessor system,” IEEE J. Solid-State Circuits, Vol. 35, No. 11, pp. 1571-1580, Nov. 2000.

[31]A. Stavdas, et. al., “Dynamic CANON: A Scalable Multi-domain Core Network”, IEEE Commun. Mag., vol. 46, pp. 138-144, 2008.

[32]B. Chen et al., “Clustering for Hierarchical Traffic Grooming in Large-Scale Mesh WDM Networks”, IEEE/OSA Journal of Optical Communications and Networking (JOCN), vol. 2, pp. 502, 2010.

[33]M. Li, B. Ramamurthy, “A Generic Autonomous Clustering-Based Heterogeneous Waveband Switching Architecture in WDM Networks”, in Proc. IEEE/OSA OFC, 2006.

[34]L. Velasco, P. Wright, A. Lord, and G. Junyent, “Designing National IP/MPLS Networks with Flexgrid Optical Technology,” in Proc. ECOC 2012

[35]Q. Liu et. al., “Distributed Grooming in Multi-Domain IP/MPLS-DWDM Networks”, In Proc. of the 28th IEEE conference on Global telecommunications, GLOBECOM, pp. 6384-6389, 2009

[36]J. Wu, L. Guo, W. Hou, “Multi-domain grooming algorithm based on hierarchical integrated multi-granularity auxiliary graph in optical mesh networks”, Photon Netw. Commun., vol. 23, no. 3, pp. 205-216, 2012

[37]T. Orphanoudakis, A. Drakos, A. Stavdas, “Efficient Traffic Grooming Through Node Clustering”, in Proc. IEEE/OSA OFC, 2012.

[38]K. Christodoulopoulos, I. Tomkos, and E. Varvarigos, “Elastic bandwidth allocation in flexible OFDM based optical networks,” IEEE/OSA J. Lightwave Technol., vol. 29, pp. 1354-1366, 2011.

[39]Y. Wang, X. Cao, and Y. Pan, “A study of the routing and spectrum allocation in spectrum-sliced elastic optical path networks,” in Proc. of IEEE INFOCOM, 2011.

[40]M. Klinkowski and K. Walkowiak, “Routing and spectrum assignment in spectrum sliced elastic optical path network,” IEEE Comm. Lett., vol. 15, pp. 884-886, 2011.

[41]L. Velasco, M. Klinkowski, M. Ruiz, and J. Comellas, "Modeling the Routing and Spectrum Allocation Problem for Flexgrid Optical Networks," Springer Photonic Network Communications, vol. 24, pp. 177-186, 2012.

Page 16 of 99

http://www.jdsu.com/





[42]A. Castro, L. Velasco, M. Ruiz, M. Klinkowski, J. P. Fernández-Palacios, and D. Careglio, “Dynamic Routing and Spectrum (Re)Allocation in Future Flexgrid Optical Networks,” Elsevier Computers Networks, vol. 56, pp. 2869-2883, 2012.

[43]F. Cugini, F. Paolucci, G. Meloni, G. Berrettini, M. Secondini, F. Fresi, N. Sambo, L. Potí, and P. Castoldi, “Push-pull defragmentation without traffic disruption in flexible grid optical networks,” IEEE/OSA J. Lightwave Technol., vol. 31, pp. 125-133, 2013.

[44]M. Klinkowski, M. Ruiz, L. Velasco, D. Careglio, V. Lopez, and J. Comellas, “Elastic Spectrum Allocation for Time-Varying Traffic in FlexGrid Optical Networks,” IEEE Journal on Selected Areas in Communications (JSAC), vol. 31, pp. 26-38, 2013.

[45]L. Velasco, P. Wright, A. Lord, and G. Junyent, “Saving CAPEX by Extending Flexgrid-based Core Optical Networks towards the Edges,” submitted to IEEE/OSA Journal of Optical Communications and Networking (JOCN), 2013.

[46]A. Asensio, M. Klinkowski, M. Ruiz, V. López, A. Castro, L. Velasco, J. Comellas, “Impact of Aggregation Level on the Performance of Dynamic Lightpath Adaptation under Time-Varying Traffic,” in Proc. IEEE International Conference on Optical Network Design and Modeling (ONDM), 2013.

[47]T. Gonzalez, “Clustering to minimize the maximum inter-cluster distance”, Theoret. Comput. Sci., vol. 38, pp. 293–306, 1985.

[48]A. Klekamp, R. Dischler, R. Buchali, “Limits of Spectral Efficiency and Transmission Reach of Optical-OFDM Superchannels for Adaptive Networks,” IEEE Photonics Technology Letters, vol. 23, 2011.

[49]K. Christodoulopoulos, P. Soumplis, E. Varvarigos, “Planning Flexgrid Optical Networks under Physical Layer Constraints”, submitted to IEEE/OSA Journal of Optical Communications and Networking.

[50]K. Christodoulopoulos, P. Soumplis, E. Varvarigos, “Trading off Transponders for Spectrum in Flexgrid Networks,” in Proc. IEEE/OSA OFC 2013

[51]A. Castro, L. Velasco, M. Ruiz, and J. Comellas, “On the benefits of Multi-path Recovery in Flexgrid Optical Networks,” submitted to Transmission Systems Journal, 2013

[52]D. King, A. Farrel, “A PCE-based Architecture for Application-based Network Operations,” IETF draft, Jan. 2013.

[53]E. Crabbe, J. Medved, R. Varga, and I. Minei, “PCEP Extensions for Stateful PCE,” IETF draft, Mar. 2013.

[54]A. Castro, L. Velasco, J. Comellas, G. Junyent, “Dynamic Restoration in Multi-layer IP/MPLS-over-Flexgrid Networks,” in Proc. IEEE Design of Reliable Communication Networks (DRCN), 2013.

[55]K. Christodoulopoulos, I. Tomkos, E. Varvarigos, “Time-Varying Spectrum Allocation Policies and Blocking Analysis in Flexible Optical Networks,” Journal on Selected Areas in Communication, vol. 31, 2013.

[56]A. Castro, F. Paolucci, F. Fresi, M. Imran, B. B. Bhowmik, G. Berrettini, G. Meloni, A. Giorgetti, F. Cugini, L. Velasco, L. Potì, and P. Castoldi, “Experimental Demonstration of an Active Stateful PCE Performing Elastic Operations and Hitless Defragmentation,” accepted in ECOC 2013.

[57]Mantis: http://www.mantis-tool.net/

Page 17 of 99

http://www.mantis-tool.net/





[58]GALACTICO: http://www.ict-galactico.eu/

2.2.2 Acronyms

ADC Analog to Digital conversion

AoD Architecture on Demand

AWG Arrayed Waveguide Grating

BRAS Broadband Remote Access Server

BVT Bitrate Variable Transponder / Transceiver

CANON Clustered Architecture for Nodes in an Optical Network

CoS Class of Service

DAC Digital to Analog converter

DSP Digital Signal Processing

DP Dual Polarisation

EON Elastic Optical Network

FEC Forward Error Correction

ICU Idealist Cost Unit (based on a 100G coherent transponder)

LCoS Liquid Crystal on Silica

LSP Label Switched Path

MAN Metropolitan Area Network

MGOXC Multi Granular Optical Cross Connect

MMTTR Minimum Mean Time To Repair

MTBF Mean Time Between Failures

MTTR Mean Time To Repair

OFDM Optical Frequency Division Multiplexing

OLA Optical Line Amplifier

OTN Optical Transport Network

OXC Optical Cross Connect

PLI Physical Layer Impairments

QAM Quadrature Amplitude Modulation

QoS Quality of Service

QoT Quality of Transmission

QPSK Quadrature Phase Shift Keying

Page 18 of 99

http://www.ict-galactico.eu/





ROADM Reconfigurable Optical Add Drop Multiplexer

RSA Routing and Spectrum Assignment

RWA Routing and Wavelength Assignment

SBVT Sliceable Bitrate Variable Transponder

SCU STRONGEST Cost Unit based on 10G Transponder

SSON Spectrum Switched Optical Network

WSON Wavelength Switched Optical Network

WSS Wavelength Selective Switch

2.2.3 Definitions

2.3 Document History

Version Date Authors Comment

Draft 1 20.5.13 Andrew Lord 1st draft after Barcelona Plenary. Contains placeholders for all contributions

Draft 2 17.6.13 Andrew and others Draft after first round of inputs

Draft 3 24.6.13 Andrew and others Draft after main review

Final version 26.6.13 Juan Final review and quality check

Page 19 of 99





3 Network architecture and Use Cases

3.1 Rationale behind Elastic Optical NetworksOptical spectrum optimization, although it is an important benefit, might not be enough to justify the EON approach. Further benefits from elastic optical networking may come from: joint IP and optical transport optimization, multilayer restoration, regeneration minimization, power consumption reduction, transponders optimization (SBVT), operational simplification (programmable transponders) and others.

Optical Spectrum OptimizationIt is well known that one of the benefits of flexgrid is the optical spectrum optimization due to its ability to adjust the spectrum reserved for a channel to the actual spectrum needs of the optical signal. However, a key question is: when will optical spectrum become a problem for network operators?

This section studies the evolution of a national optical transport network, comparing a strategy that follows current fixed grid architecture (WSON strategy) and an alternative strategy where a flexgrid architecture is implemented (SSON strategy). Specifically, it is intended to determine the year when the installed capacity will be exhausted and new optical links would need to be deployed. This information will be useful in order to check if the capacity gain given by the introduction of flexgrid is a necessity in the coming years and if it is worth implementing flexgrid in a real network.

In this study we will use the term WSON to refer to the fixed grid-based architecture. This WSON evolution model represents the continuity of the already deployed infrastructure. On the other side, the SSON model represents a scenario where flexgrid capability is present. There are two possible cases in the SSON model: 1) it is possible to activate the flexgrid functionality in the WSSs by a software upgrade; and 2) the deployed WSSs cannot implement flexgrid capability and would need to be replaced. This can be considered as a greenfield scenario. This section considers the case where WSSs are upgradable.

The reference network is the Spain Reference Transport network publicly available in [10] and described in Section 4. The study assumes an initial ROADM deployment and an initial set of 10 Gb/s and 40 Gb/s given by the reference scenario. After the initial deployment, the network grows, on the one hand, following the WSON model and on the other hand, following the SSON model. The study compares the number of new long-haul links that need to be activated following each strategy.

A heuristic RWA/RSA algorithm for WSON and SSON architectures respectively has been used for the path calculation and resource (wavelength/spectrum) assignment of the optical channels. The algorithm for the WSON model is the Adaptive Unconstrained Routing Exhaustive (AUR-E) [1] RWA algorithm. For the SSON case such an algorithm has been adapted to solve the RSA problem.

Figure 1 shows the number of activations of new links needed up to a given year in both evolution models. It can be seen that the capacity of the deployed network will be exhausted by 2019 if the WSON evolution model is followed. However, by the activation of the flexgrid functionality the lifetime of the network is extended by 5 years. This life extension in the SSON evolution model is due to both the use of high efficiency modulation formats beyond 100 Gb/s and the adaptive spectrum assignment.

Page 20 of 99





2014 2016 2018 2020 2022 20240

10

20

30

40

50

WSON Case FLEXI-GRID Case

num

ber o

f new

Fib

er L

inks

(cum

mul

ated

)

Year

Figure 1: Number of new Fiber Links WSON vs. flexgrid evolution models

Electronic regeneration minimization beyond 100Gb/sA short-term driver for flexgrid might be the appearance of 400 Gb/s client signals and cost effective 400 Gb/s transponders. In this case study, for the Telefónica core transport network, the number of 400 Gb/s demands that could be needed in the following years, with a 30% yearly traffic increase, maintaining the same traffic distribution, is obtained. Figure 2 shows the number of forecast demands for 400 Gb/s channels per year.

Figure 2: Number of new Fiber Links WSON vs. flexgrid evolution models

There are two main choices for 400 Gb/s transmission:

400 Gb/s transmission based on OFDM-DP-QPSK over 125 GHz. The reach of each subcarrier modulation format (DP-QPSK) would enable long haul deployments. However, as 125 GHz is needed, it is only feasible using flexgrid technology.

The other option is to split the demand into two wavelengths with DP-16QAM 200 Gb/s over the standard fixed 50 GHz grid. However, in this case, the reach is lower. Taking into account the length and number of ROADMs on the long haul routes in

Page 21 of 99





the Telefónica of Spain Network, approximately at least 30% of the 400 Gb/s transmission using DP-16QAM would need regeneration. Such regeneration can be avoided by using OFDM-DP-QPSK over flexgrid.

A similar analysis was also done over Telecom Italia network, where different technological scenarios characterized by transponders that use different modulation formats and number of carriers for each channel (e. g. single or multiple carriers organized as superchannels) were investigated. The considered scenarios for this analysis classified as homogeneous (all the traffic is carried by only on type of transponder) are reported in Table 1 where the main characteristic of each scenario is summarized.

Table 1: Transponder used in homogeneous scenarios.

Transpondertype

Client Rate

[Gb/s]

Baud rate

[Gbaud]Format Sub

carriersBW

[GHz]Comb Cap.

SE[bit/s / Hz]

ReachG.652 [km]

Reach G655 [km]

100G Reference Transponder (RT)

RT 100G 100 32 PM-QPSK 1 50 80 2 3500 2000

Nyquist DWDM Transponders based on 100G RT

NY 200G 200 32 PM-QPSK 2 87.5 45 2.25 3500 2000

NY 400G 400 32 PM-QPSK 4 162.5 24 2,40 3500 2000

NY 1T 1000 32 PM-QPSK 10 387.5 10 2.50 3500 2000

Galactico Transponders

GAL 200G 200 32 PM-16QAM 1 50 80 4 1000 600

GAL 400G 400 16 PM-16QAM 4 100(87.5)

40 (45)

4(4.5)

600 400

GAL 1T 1000 32 PM-32QAM 4 200 20 5 400 300

The baseline scenario employs the so called Reference Transponder (RT), which is the state of the art of the long haul transmission technology nowadays (beginning year 2013). RT is a 100 Gbit/s DP-QPSK requiring a bandwidth of 50 GHz (37.5 GHz for the signal plus 12.5 GHz of guard band). DWDM systems in C band using the baseline scenario are characterized by 80 channels and 2 [(bit/s)/Hz] Spectral Efficiency.

“Nyquist DWDM” RT based scenarios (denoted with NY) are made up from more signals, each with the characteristics of the RT (37.5 GHz bandwidth), close together in a single superchannel and with a guard band of 12.5 Ghz. An additional scenario based on devices proposed in GALACTICO project [58] is considered. In this GALACTICO reach optimized scenario, on every ligthpath the transponder type is chosen so to avoid as much as possible the use of regenerators (higher spectral efficiency formats have a short reach and their use on longer paths implies a higher number of regenerators). Actually, as it is well known and confirmed in the results presented in the following, transponders and regenerators give the most significant contribution to the network cost.

Figure 3 shows the CAPEX of baseline scenario for both G652 and G653, for 366 Tb/s traffic matrix. The transponder cost is dominant and regeneration costs are null (G652) or

Page 22 of 99





very small (G655). Node cost is a small fraction of the total (about 10 %) while the cost of optical line amplifiers is definitely negligible.

Figure 3 Breakeven costs of flexgrid transponders.Figure 3 shows both the absolute and the specific breakeven cost of transponders (i. e. the costs that makes the cost of a solution equal to the baseline case cost). For the Nyquist DWDM case the breakeven specific cost is essentially constant for all superchannel type: this means that interesting network CAPEX reduction could come straightforwardly from the cost reduction achieved with device integration.

Therefore we can conclude that a combination of NY DP-QPSK and Flexgrid could enable electronic regeneration minimization a for 400G and 1T channels. According to it, breakeven costs in Telecom Italia network would be much higher (more interesting) than DP-16QAM solutions. These solutions consume more optical spectrum than DP-16QAM but this is not expected to be an issue in the mid-term.

ConclusionsSumming up, the results show that a network based on flexgrid could extend the lifetime of the network 5 years with respect to the legacy WSON. Results also show that current WSON capacity will not be exhausted until 2019. Hence, the pure argument of capacity improvement is not a short-term driver to deploy flexgrid solutions. However, the limitation on the maximum capacity that can be carried on a single wavelength causes a dramatic increase in 100 Gb/s optical channels, in the case where the network grows following a WSON model. If the network migrates towards flexgrid, the number of transponders is reduced, thus justifying the migration of some connections to flexgrid. In the case that the deployed network is based on WSSs that are not flexgrid-capable, a mixed WSON-SSON

Page 23 of 99





strategy has been presented. Such a strategy implies starting to deploy a parallel SSON network when the capacity of the WSON network is exhausted. At the expense of the deployment of the parallel network, the number of transponders is drastically reduced compared to keeping traffic on the WSON network, which needs to grow adding parallel links and increasing the degree of the ROADMs.

Even though in the coming years, capacity in terms of spectrum is not yet an issue, and thus migrating to flexgrid just because of the lack of spectrum, is not urgent, a key driver to migrate to flexgrid is the demand for high speed channels and the availability of cost-effective 400 Gbs and 1 Tbps transponders.

3.2 Use cases for Elastic Optical Networks In this section, the project identifies the EON Use Cases i.e. it elaborates where and in what way flexgrid will impact network architecture. The Use Cases are classified into two large categories: the short-term objectives (ST) and the long-term (LT) ones. The former cover architectural changes through an evolutionary path, whilst the latter are more disruptive, having a profound and deeper impact on network architecture.

3.2.1 ST-1: Multi-Layer Restoration

Internet traffic in national backbones of European operators continues to grow at an annual rate of between about 30 and 35 per cent. In this situation, operators are about to add new transport capacity connecting backbone routers over an agile DWDM layer.

In traditional packet networks, it is the IP layer that reacts to a failure. Following the traditional approach of having reactive resilience exclusively on the packet layer causes a comparably bad utilization of both router interfaces as well as transponders equivalent to lambdas on the optical layer. Though packet traffic is statistically multiplexed onto lambdas, those lambdas may be filled up only to about 50 per cent of the maximum. The remaining capacity is reserved for backup if a failure occurs. Without any dynamic countermeasure on the DWDM layer, optical robustness is conventionally assured by the creation of a second 1+1 disjoint backup path. This comes along with a second transponder interface leading to more wavelengths and more links in the overall network. Directly related to these intrinsic inefficiencies are high capital expenditures inhibiting overall network profitability.

With a recently proposed converged transport network architecture (in the sense of an integration of packet layer devices and optical transport equipment), operators are aiming at building a network that comprises automated multi-layer resilience enabling considerably higher interface utilization. By treating different service classes unequally, capacity for best effort class traffic is optically recovered after a short delay, while high-priority traffic maintains priority and is served in the same way as today. The concept relies on an agile photonic layer where switching on the lambda layer is related with the least cost. Several studies have proved the feasibility and techno-economic superiority of this architecture [2].

Flex-reach ML Resilience: In case of a network failure on the optical layer, such kind of multi-layer (photonic & packet) resilience switching induces of course an unintended and sudden, but sometime unavoidable traffic dynamic. When routing the backup path through the EON, it might exceed the reach specification for the given modulation format. One promising application case of future software-

Page 24 of 99





configurable transceivers is to benefit from the reach-capacity trade-off, i.e. to adjust the modulation format dependent on the reach requirement. Practically, this means a reduction of the original interface capacity1. However, instead of losing the entire interface capacity, automatically lowering an existing flex-rate interface provides at least some part of the original capacity to be further exploited by IP routers accordingly. Thus, this type of photonic elasticity may save the operator a considerable amount of CAPEX when compared to the case of a fixed-reach transceiver with its necessity for intermediate signal regeneration.

Transceiver Sliceability for ML Resilience: Due to the unpredictability of traffic arising from, e.g. cloud services on one hand and the general burstiness of packet traffic on the other hand, there is always a significant amount of stranded capacity in today’s networks, which provide fixed-capacity circuits between nodes. This leads to an under-utilization of interfaces on the optical layer. Especially in the early days of an optical transport network, the utilization of interfaces is inherently low.

In this situation a high-capacity transponder (e.g. at 400 Gb/s or 1 Tb/s) being able to be logically and physically sliced into several virtual transponders targeting at different destinations with electronically adaptive bit rates may be highly beneficial for improving network economics. At day one, even a single sliceable bandwidth variable transponder may provide enough total capacity for serving all destinations at a comparably low bit rate, say 100 Gb/s. Later when the traffic increases, it may serve only a few destinations each at a higher bit rate. Finally, it may support only a single huge flow to a single destination. All this is expected to become electronically controlled and adjustable.

The spectral slicing process is to be monitored and driven by the associated IP routers. They control the different service classes and assign them to spectral bandwidth slices enabling significant provisioning flexibility. In case of a failure in the optical layer, an adapted treatment of service classes is expected to be beneficial. An integrated control plane optimizes the assignment of high-priority and best-effort traffic unequally to optical subcarriers. In addition to parameters like, e.g. path cost, latency, and shared risk groups, the optimization may also be accomplished dependent on various EON-specific parameters like availability of spectral slices or fiber fragmentation status.

Consequently, the next evolution step might be the investigation of the appropriateness of SBVTs within a realistic ML resilience concept based on a service class differentiation.

For both application cases, the network architecture requires two main ingredients at least: Firstly, an agile DWDM layer with colourless, directionless and flexgrid ROADMs provides faster service provisioning. These photonic devices are also used for resilience switching at the physical layer (L0). Secondly, an integrated packet-optical control plane offers a software-based flexibility including routing, spectrum assignment instead of wavelength assignment (RSA vs. RWA), signalling functionalities as well as elastic path selection in case of a network failure. An efficient integrated CP solution also comprises just the right amount of information exchange between the packet and elastic optical layer.

1 Without further limitations (e.g. spectral scarcity) it is supposed here that in real network deployments optical interfaces run at their maximum capacity.

Page 25 of 99





3.2.2 ST-2: IP over Elastic/FixRate optical networks

IP/MPLS over DWDM multi-layer optical network architecture is a promising solution to bridge the bandwidth gap between client packet data flows and ultra-high capacity lightpaths enabled by recent developments on optical transmission systems and advanced modulation formats. Commercial transmission equipment at 100 Gb/s is commercially available to date, and research efforts are already targeting 400 Gb/s and 1 Tb/s [4]. To efficiently exploit the high capacities on optical devices at the WDM layer, multi-layer optical networks groom lower speed client flows (up to 10 Gb/s) onto available optical connection (which capacity can reach hundreds of Gb/s). It has been demonstrated that multi-layer dimensioning, allowing the optical bypass is the more cost efficient solution when compared to the non-bypass (or opaque) solution. Indeed, optical bypass allows to offload intermediate IP/MPLS routers between electrical termination points reducing router capacity requirements, thus network capital expenditures (CAPEX) and/or the network power consumption [5].

With optical connections carrying such high capacities (100 Gb/s and 400 Gb/s), any failure (e.g., fiber cut, transponder or node failure) can lead to catastrophic data losses. These data losses also have associated big economic losses for operators, due to the high downtime costs when serving certain kinds of clients. Therefore, survivability becomes of paramount importance in the design and operation of multi-layer optical networks.

Within Idealist we want to propose a study where IP/MPLS over WDM multi-layer optical networks, survivable against any single link failure scenario using IP/MPLS protection switching is investigated.

This solution is interesting for network operators because:

1. Fast Label Switched Path (LSP) recovery below 100 ms can be achieved [6]; 2. Recovery is entirely performed at the MPLS layer and there is no need for still

immature optical control plane solutions; 3. Recovery can be performed per LSP, allowing differentiated LSP recovery based

on classes if desired.

To allow a dynamic connection of the client ports to any WDM transponder, we assume that optical nodes can change the port-to-wavelength connectivity through a Patch Panel (PP, e.g. as shown in Figure 4) [7], allowing the reuse of router ports during failure scenarios.

The objective of this approach is to design an IP/MPLS over single-line rate, multi-rate and elastic (with/without sliceable bandwidth variable transponders, SBVT) optical network carrying the predicted peak traffic matrix, so that all packet flows are survivable under any single link failure through IP/MPLS protection switching with the minimum CAPEX/OPEX. This entails a joint optimization of the packet flow primary and backup routes over the IP/MPLS layer, together with the Route Wavelength and Rate Assignment (RWRA) of all optical connections required in all virtual links to make the network 1-link failure survivable.

Page 26 of 99





Figure 4: Survivable multi-layer optical network design.

3.2.3 ST-3: Flexgrid Optical Networks for DataCentre Federations

In the Internet of services, information technology (IT) infrastructure providers play a critical role to make the services accessible to end customers. IT infrastructure providers host platforms and services in their datacenters (DCs). The cloud initiative has been accompanied by the introduction of new computing paradigms, such as infrastructure as a service (IaaS) and software as a service (SaaS), that have dramatically reduced the time and costs required to develop and deploy a service [8]. These paradigms are playing a role of paramount importance in the way companies invest their money regarding to IT resources; companies are moving from a model where large amount of capital (CAPEX) is needed to build their own IT infrastructure and additional cost to maintain it (OPEX) to a pure OPEX model where IT resources are requested to cloud providers in a pay as you go model.

Dimensioning DCs is a challenging task since workload mixes and intensities are extremely dynamic; dimensioning DCs for the peak load can be extremely inefficient, whereas reducing its capacity might result in a poor quality of service (QoS) causing service level agreements (SLA) breaches. In addition, the huge energy consumption of DCs requires an elastic resource management, e.g. by turning-off physical machines (PM) when they are not used or turning them on to satisfy increments in the demand. Thanks to virtualization, mixed workloads can be easily consolidated and performance isolated, their consumptions tailored, and placed in the most proper PM according to its performance goals. By encapsulating jobs in Virtual Machines (VM) a cloud resource manager can migrate jobs from one PM to another looking for reducing energy consumption (or some other OPEX objective) while ensuring the committed QoS [9][10].

Creating DC federations, where two or more cloud providers interconnect their DCs, allow further load balancing accommodating spikes in demand [11]. This brings benefits to cloud providers since: i) they can increase their revenue from using IT resources that would otherwise be under-utilized; and ii) it enables them to expand their geographic coverage without building new DCs.

Page 27 of 99





Network operators provide the necessary network infrastructure to create federations interconnecting DCs and linking users and services. In a recent study [12], Cisco forecasts DC traffic to quadruple over the next 5 years, reaching 554 EB per month by 2016. Two main components of traffic leaving DCs can be distinguished: i) traffic among DCs (DC2DC); and ii) traffic between DC and end users (U2DC). The former includes, among others, VM migration and DB synchronization, whereas the latter is associated to the transactions between applications and end users. Like in IT environments, dimensioning connectivity for the peak leads to dramatic inefficiencies if we consider that DC2DC traffic is generated periodically as a result of a set of VMs being migrated.

As shown in [13] for a study of traffic profile in dual DC hubs in Latin America, DC2DC traffic is generated periodically as a result of a set of VMs being migrated and DBs being synchronized (Figure 5). Therefore, as before, dimensioning connectivity for the peak leads to dramatic inefficiencies. However, network providers still maintain a relatively rigid structure currently offering static connection provisioning. Transport networks are currently configured with big static fat pipes based on capacity over-provisioning since they are needed for guaranteeing traffic demand and QoS (Figure 6a). However, Cloud services require new mechanisms to provide reconfiguration and adaptability of the transport network, thus reducing the amount of over-provisioned bandwidth.

Figure 5: Traffic Profile in Dual Datacentre Hubs in Latin America [13].

Dynamic connectivity could allow DCs to manage optical connections to remote DCs, requesting connections as they are really needed so as to perform data transferences and releasing them when all data has been transferred (Figure 6b). Furthermore, the finer granularity and the wider range of bitrates that networks based on the flexgrid technology can provide (in comparison to the fixed grid technology) make flexgrid technology the ideal candidate for inter-DC interconnection. After requesting a connection and negotiating its capacity as a function of the current network availability, the resulting amount of bitrate can be used by scheduling algorithms to organize transferences. The availability of resources, however, is not guaranteed and lack of network resources at requesting time may result in long transference times and even in transfer period overlapping. Note that connection’s bitrate cannot be re-negotiated and remains constant over the connection’s holding time.

Page 28 of 99





Figure 6: Inter-DC connectivity. a) Static connectivity; b) Dynamic connectivity; c) Dynamic and elastic connectivity.

To solve to some extent the unavailability of required connectivity resources at requesting time, on-demand connection elasticity can be used to allow an increase/decrease in the amount of spectral resources assigned to each connection, and thus its bitrate (Figure 6c); if not enough resources are available at the connection request time, more bitrate can be requested at any time after the connection has been set up. Elasticity can reduce transference times, thus minimizing the period of overlapping probabilities compared to the fixed connection bitrate provided by previous models.

3.2.4 ST-4: The disaster recovery (DR) Recovery of the network infrastructure from large scale disasters such as hurricanes, floods, or terror attacks, is becoming increasingly important. This is partly due to the more crucial role the network plays today, and partly due to an increasing prevalence of such disasters. The approach to DR is different from the approach to network failures such as fiber cuts or node failures. The latter can be planned for, by providing sufficient resources to deal with all single network failures. There is no way to anticipate the impact of a disaster and therefore no cost effective way to plan for full recovery from it. A disaster may sever multiple routers, take down entire central offices, or data centers connected to the network. The key to a cost effective disaster recovery is to have a flexible enough network that can be reconfigured to provide maximum (but typically not full) recovery of service. Another crucial element is a network management application that has up to date information about available resources, current traffic conditions, and the failure impact. This application must also support orchestration of the required change in the network.

BVTs play an important role in providing sufficient flexibility for DR. Without BVTs, the reach of a particular transponder is fixed. If the DR plan requires a longer reach, it must identify regenerators to extend the reach. Since regenerators are an expensive and therefore scarce resource in the network, they may not be in place for the required connections to be established during a disaster. By contrast, BVTs allow for extended unregenerated reach at a lower capacity, which may provide a much better alternative than not having the connectivity at all. The ability to slice a BVT and use it in multiple directions is also important as it allow increased connectivity at lower bit rates. Without this property, the surviving BVTS may not suffice in providing the required connectivity during the disaster.

Quantifying the cost benefits of BVTs and SBVTs for DR is not straightforward: one cannot simply compare the cost of a BVT based solution to a non-BVT solution, since in both cases the network is not over-provisioned to deal with disasters. Instead, the two solutions must be compared based on the % traffic being restored after the disaster.

Page 29 of 99





3.2.5 ST-5: Flexgrid in metro-regional networks - serving traffic to BRAS servers

This study aims to describe and evaluate a potential application scenario for the introduction of the flexgrid approach in the Metro Area Networks (MAN) to support BRAS centralization. To do so, we will compare a flex-grid solution in a realistic metro-regional network scenario to traditional approaches based on fixed-grid WDM systems that are used today.

3.2.5.1 Current Metro architectureTypically, metro architectures are composed of two main levels of aggregation:

i) First level of aggregation (named MTU level), in charge of collecting traffic from the OLTs;

ii) Second level of aggregation (named Access level), that aggregates the traffic from the MTUs mainly through direct fiber connections (i.e. dark fiber).

Nowadays, the IP functionality (i.e. traffic classification, routing, authentication, etc.) is implemented in the Broadband Remote Access Servers (BRAS) which are usually located at the second level of aggregation. Therefore, they are distributed in many sites along the regional area that is covered by the MAN, causing a high CapEx impact.

Current network architecture contains connections between MTU switches towards a number of BRAS servers over an optical metro-regional transport network (Figure 25). A representative scenario has 200 MTU switches that generate traffic and access level with 62 nodes and an optical transport network comprising 30 ROADMS that are connected to 2 BRAS servers over that network.

Figure 7: Current Metro architectureSince, in the last years, the main European network operators have been deploying/expanding their photonic mesh to the regional networks, it is proposed to move the BRAS’s (i.e. IP functionality) to the transit level (typically, it is composed of two sites for redundancy purposes). This way, the IP equipment needs (and its associated costs) would be reduced. A conveniently dimensioned pool of BRAS’s (remote BRAS) could be co-located at each of these two transit sites in order to provide the required redundancy.

Following sections proposes architectural options to be assessed to support this new situation.

Page 30 of 99





3.2.5.2 Scenario A: evolutionary approachIn this first scenario (i.e. the reference scenario for the comparative analysis), the regional photonic mesh provides the optical transport capacity from the second aggregation level (i.e. access level) towards the remote BRAS (Figure 25). Therefore, these aggregation elements are distributed among the ROADMs, which compound the regional photonic mesh.

Figure 8: Scenario A – evolutionary approachThis architecture does not provide too many savings in terms of CapEx, since the first level of aggregation (i.e. MTU) is connected to the second level in the electronic domain, or in terms of bandwidth efficiency, since most lightpaths through the regional photonic mesh will require a capacity of 10Gb/s for the 50GHz-grid channels.

3.2.5.3 Scenario B: Fixed-based approach with adaptation layerIn order to cope with the problem of the evolutionary approach, we plan to investigate a second evolutionary scenario for the metro architecture that considers the substitution of the second level of aggregation (i.e. access level) by a named adaptation layer. In this case, the regional photonic mesh provides transport connections from the first aggregation level (i.e. MTUs) towards the remote BRAS, saving the electronic equipment expenses between the first and the second levels of aggregation in the reference scenario.

Figure 9: Scenario B1 – Fixed-based approach with adaptation layer without SBVT

Page 31 of 99





In order to improve the spectral efficiency (i.e. the channel bandwidth utilization through the regional photonic mesh), and assuming that each MTU (i.e. first level of aggregation) requests 10Gb/s traffic demands (as a result of the grooming of traffic from the OLTs) and can make use of Sliceable Bandwidth Variable Transponder (SBVT). An SBVT can transmit from one point to multiple destinations, changing the traffic rate to each destination and the number of destinations on demand [14]. We propose to aggregate N of these 10Gb/s-capacity channels, in the 12.5GHz grid each, into a 50GHz-grid channel by means of the adaptation level previously mentioned. This 50GHz channels are routed through the regional photonic mesh towards the ROADM associated to the remote BRAS. Before reaching the remote BRAS, another adaptation layer is needed to split/extract each of the 12.5GHz-grid channels. Figure 10 shows this scenario from the L2 point of view.

Figure 10: L2 view for the scenario B

At the remote BRAS site, each of these (12.5GHz) channels can be groomed optically and be sent in a much bigger capacity trunk towards the core network by means of a sliceable BVT (SBVT). This scenario is shown in Figure 11.

Figure 11: Scenario B2 – Fixed-based approach with adaptation layer with SBVT

3.2.5.4 Scenario C: Flexgrid-based approach with Sliceable BVTs at the remote BRAS

The last approach to be assessed is a flexgrid solution in which the WDM network is replaced by a flexgrid network and assuming the use of Sliceable Bandwidth Variable Transponders (SBVT) at the BRAS servers. This scenario is proposed in Figure 12

Page 32 of 99





Figure 12: Scenario C – Flexgrid-based approach with SBVTs at the remote BRAS

3.2.5.5 Evaluation of the Scenario C use-caseWe plan to develop algorithms to assign routes and wavelengths (in case of a WDM network) or spectrum slots (in case of a flexgrid network) to the connections. The algorithms to be developed will be extensions to the algorithms that we have developed to serve connections in traditional WDM and flexgrid networks [15], [16], taking into account the particularities of this scenario. In more details, we will extend our algorithms to optimize routing and resource allocation for traffic that is directed towards a few specific nodes of the network (BRAS nodes), and we will also account for the use of SBVT in a flexgrid network. Using these algorithms and the cost models developed in the framework of the Idealist project, we will analyze the feasibility of the different networking architectures and analyze their performance. Our goal is to draw conclusions on the targeted prices for the adaptation layer and the SBVT that would yield the flexgrid approach as the most viable solution for this case study.

3.2.6 ST-6: Scalable core networks with Architecture on Demand nodesAlthough research interest on elastic optical networks has been continuously increasing, there has been very little work focusing on elastic node architectures. This is perhaps due to the relative simplicity to provide elastic spectrum allocation using a broadcast-and- select arrangement with BV-WSS, as shown in Figure 13(a). Furthermore, the broadcast-and-select architecture is an optimum arrangement that provides maximum switching flexibility from its components [20]. So, why should we look into alternative elastic node architectures? It turns out that although the broadcast-and-select arrangement provides maximum switching flexibility, it is very restrictive in terms of functional flexibility, i.e. the capability to provide additional services such as spectrum defragmentation, wavelength conversion, regeneration, time multiplexing, etc. This limitation is even more problematic if the requirement for additional functionality is uncertain, e.g. it may be fluctuating or depend on the network or geographic regions considered.

Page 33 of 99





(a)

(b)

Figure 13. (a) Broadcast-and-select and (b) architecture on demand nodes.In addition, the broadcast-and-select architecture depicted in Figure 13(a) is not scalable to high-capacity networks. Considering that, due to traffic requirements, future optical core networks will likely need to support multiple fibres in any direction, the realisation of optical networks will require multi-degree nodes with very high switching scalability. A broadcast-and-select implementation is limited by the available number of ports in state-of-the-art WSS devices, around to 20 to 25. Similarly, power splitters with a large number of ports would introduce high loss, in addition to the loss of any add/drop network.

In view of this, the alternative solution of an adaptive optical node, shown in Figure 13(b), able to provide different architectures and services depending on traffic requirements seems very attractive. Such a node needs to be able to change its switching granularity in order to provide additional functionality and to improve scalability. For instance, assume that one of the 100 signals in a fibre/core requires wavelength conversion. It would first need to be isolated, then wavelength converted, and then recombined with other signals. Thus, signals that do not require wavelength conversion are not affected. The granularity required here is that of the channel for which wavelength conversion is carried out, i.e. a single channel is present in the switching path, which is not possible with broadcast-and-select. Alternatively, in cases where all signals in an input fibre/core need to be switched to

Page 34 of 99





a single output fibre/core it is not necessary to broadcast the contents of the input or to use a filtering device (e.g. BV-WSS) at the output. Instead, a single connection from input to output is able to satisfy the switching requirements of all these signals. The granularity required here is that of a fibre/core. Other fibre/cores may need to be demultiplexed into several bands or individual channels. Moreover, the node configuration needs to be dynamic and reconfigurable in order to cope with fluctuating requirements. This concept can be extended to a networking scenario with several adaptive nodes linked together to create very flexible optical networks. Such networks would be able to split the network infrastructure into several virtual networks, thereby providing physical and virtual network topologies on-demand, with support for different traffic granularities. The number of possible configurations is enormous and the flexibility provided is unprecedented in optical networking.

Page 35 of 99





4 Reference networks

4.1 IntroductionThis chapter provides a brief overview of the reference network infrastructures provided by the operators participating in the Idealist project: Telefonica (TID), British Telecom (BT), Deutsche Telekom (DTAG) and Telecom Italia (TI). Each operator has made available to the project a description of its national transport infrastructure with the aim of performing network evaluations in the realistic environment of a typical European country characterized by a network diameter from one to two thousand kilometres. Telecom Italia provides also a European backbone for studies over a longer distance context (longest paths are some thousands kilometres length) where performance of high spectrally efficient optical transmission without OEO regeneration is expected to be critical.

The networks defined are essentially the transport infrastructure in terms of topology (node, links), infrastructural specifics (span lengths, hierarchical organization) and physical characteristics (fibre parameters). Details about the topologies of the upper network layers (IP/MPLS in particular) are not specified according to the belief that architecture and topologies at upper layers should be the subject of the multi-layer optimization: that requires some additional hypothesis on the network and service architecture of the higher layer that is out of the scope of this deliverable.

The purpose to define these reference networks is to have a set of common scenarios for most of the network studies under development within the project. Such studies deal with networking and architectural issues (including control plane concerns), network planning and dimensioning, simulations for performance evaluation (in both the fields of pure transmission and networking), algorithms benchmarking (for spectrum allocation, defragmentation, single- and multi-layer resilience mechanisms and strategies) and techno-economic assessments.

In order to simplify the terminology, the network is divided into two main segments, the core and the metro. Both segments, core and metro, can be organized in more than one tier. The core is the national backbone, including regional networks in the case where the core is integrated in a two-tier structure (this is the case of Telefónica’s transport network), whereas the Italian, the German and British smallest networks are national flat networks; while the metro is the metropolitan network that can include the nearby regional network where regional networks are clearly separated from the national core. Metro segment can include the aggregation level as a differentiated tier. (This is the case of the biggest network provided by BT, which has 1113 nodes and includes the Aggregation level in addition to Core and Metro). Access is not considered due to the fact that Idealist does not deal with this network segment.

The reference networks are fully defined at least in terms of a list of nodes, each with its identifier and type (in terms of hierarchical rank, if the network is hierarchical), and a list of links, each with its fibre lengths and other optional characteristics (type of fibre, number of span and others). The data is included in the spreadsheets that accompany this deliverable. Traffic demands are only available at this stage of the project in a static mode. Dynamic traffic demand forecasting and modeling is under analysis by the operators participating in the project and results are expected in subsequent deliverables. Up to now dynamicity is created, where required, starting from static demands and creating the

Page 36 of 99





dynamics by means of models obtained from traffic engineering theory. For instance, the static demand (converted to Erlangs) can be used to create dynamic traffic with additional assumptions on the statistics of holding times and inter-arrival times of the connections.

4.2 Telecom Italia National Reference Core NetworkThe Italian optical backbone owned by Telecom Italia is a flat, non-hierarchical network composed by 44 nodes and 71 fibre links. In its current state, the nodes of the network are based on ROADMs, using fixed grid 1x9 WSS modules. OTN and other electrical switched layer 1 capabilities are currently not integrated into the optical layer.

A schematic topology of the Telecom Italia network and the main topological and fibre characteristics are shown in Figure 14.

Figure 14: Topology of the Italian national optical backbone and its main topological characteristicsThe topological nodal degree ranges from 2 to 5 with an average value of 3.2. In the upper table of Figure 7 the path lengths in terms of both number of hops and distance, and for both the working path and the protection path are shown. In the calculation of shorter

Page 37 of 99





paths, the protection path is imposed to be link disjoint to the working path (i.e. the two paths must not have any topological links in common but they can share one or more nodes in addition to the terminal ones). The network diameter is 2164 km for the working path and 2606 km for the protection path.

Three types of fibre are present in the network: G.652, G.653 and G.655 (Corning Leaf). The links are all of terrestrial type except three: two of them connecting Sardinia and the other connecting Sicily to the Italian peninsula. Concerning the occurrences of different type of fibre: 25 links are G.652, two links are G.653, 33 links are G.655, one link is mixed G652 and G.653, nine links are mixed G.652 and G.655. For the parameter values required in system design (attenuation, dispersion, non-linearity, Raman effect parameters) typical values can be used.

The traffic demand provided for the network originates from three sources: the optical connectivity required by the IP national backbone, an OTN network that requires connectivity for the switches and a set of demands required directly at the optical layer by customers, recognized as Lambda Wholesale. The first network installation, called “baseline” is foreseen for the year 2013 and is characterized by a total demand of 18 Tb/s. Up to 13 per cent of the traffic requires 1+1 protection. As a result the total bandwidth (working plus protection) to be carried by the optical network is about 20 Tb/s. For the future traffic evolution a number of traffic matrices have been generated, scaling up the total volume of traffic by a factor of two, and as expected, caused predominantly by an increasing relative percentage of IP traffic.

4.3 Deutsche Telekom national reference IP core networkDT’s generic IP backbone network is a 12 PoP network. It interconnects LSR routers (also known as P routers) and it serves internal traffic generated and consumed by residential subscribers exclusively. There is no inherent hierarchy. The value proposition of this generic reference network is its topological simplicity and full coverage of real end-to-end traffic volumes, which is especially helpful for further network studies and simulations.

4.3.1 Topology and Network ArchitectureThe network comprises 20 fibre edges with an average length of 243 km. The fibre is G.652 SSMF exclusively. More topological details like amplifier span lengths are expected in subsequent deliverables. All fibre edges are modelled to be mutually fully disjoint, i.e. they are not subject to any shared risk link group (SRLG). This simplifies all network studies (especially resilience mechanisms) considerably.

ODU (L1) or Ethernet (L2) switching in separate devices is not to be applied in this network. According to DT’s vision, it is the IP client layer that directly operates onto the physical-optical layer beneath. That said, DT’s operationally practical architectural preference can be ignored for scientific studies and for benchmarking various network concepts.

Page 38 of 99





Figure 15: Topology of DT’s national backbone comprising a meshed optical ROADM network (serving layer) and an IP router plane (client layer)

4.3.2 A and B network split One peculiarity exists for the DT backbone network, which is not often mentioned, but nevertheless is an important property and a practice often applied by European operators. In reality the network, as shown in Figure 15, is physically installed twice in order to guarantee robustness against failure.

That means for resilience purposes DT is about to build a so-called ‘A’ and a ‘B’ variant of the IP core network, which mutually protect each other. As a consequence, every given IP PoP consists of an A site (e.g. in case of Frankfurt there may be an A site in the South) and a B site (e.g. Frankfurt B in the North). The sites are several km apart. The average distance between A and B sites is about 6 km; the maximum distance is less than 12 km. This property is not included in the data set as this is confidential information. However, for some investigations this doubling of transport infrastructure needs to be taken into account.

For example, the doubling strategy has an impact on working and backup paths. In case of a link failure in the A network, there always exists a straightforward backup route in the B network. Both paths are mutually disjoint and follow similar geographical routes through the country. Therefore, latency is also usually similar.

Beneath the IP backbone there is a metro network, which aggregates the regional traffic. Physically this type of traffic is guided over a so-called horseshoe structure. This means that all regional nodes equipped with PE routers are interconnected to the P router through a ring-like topology. Multiple hundreds of horseshoes are installed for this purpose. Those rings are not fully closed but terminate at an A as well as at a B site of any single IP PoP location. From a transport layer perspective, the metro network is demarcated from the inner backbone network. All traffic flows coming from regional PE routers undergo an OE conversion at one of the 12 backbone PoPs and are subject to signal processing at the packet layer before being transported through the backbone network itself.

Page 39 of 99





4.3.3 Traffic matrix and forecastThe corresponding traffic model relates to the IP packet load at the peak hour, and represents the averaged traffic volume measured at every quarter of an hour, i.e. it is opaque to fast load fluctuations. The given traffic matrices refer to the years 2011 to 2016. The volume is assumed to follow the long-standing annual growth rate of 35 per cent flat for all traffic relations. The daily traffic profile shows the typical day-night shape and all traffic relations manifest very similar shapes. Besides these inherent dynamics, the traffic is considered as static.

In any case, each given traffic value refers to the total traffic between two associated IP PoPs. In the practically operated network, the traffic is split by the regional PE routers into two equal parts. Through this IP load-balancing, accomplished by ECMP (equal cost multi path), the loads on the A and the B networks are very similar. Thus for network studies, especially techno-economic assessments, the traffic matrix needs to be divided by 2 beforehand. Only then may the planning and dimensioning processing on this data base be done.

If one sorts the traffic values, the distribution turns out not to be evenly shaped, but follows a triangular shape. That means there are some nodes, which attract much more traffic than others. As an example, for our network the most important node is Frankfurt, while, e.g. Ulm or Dortmund are comparably small.

4.4 Telefónica National Transport NetworkTelefónica’s transport network is based on a hierarchical structure with 5 meshed regional domains and one “express” national domain. Intra-regional communications make use of resources within the specific region, while inter-regional communications may be established using some inter-regional links or through a dedicated (express) national domain, whose ROADMs are collocated with some selected ROADMs at each regional domain. As a starting point, we consider for the Telefónica network model a “Broadcast and Select architecture”, with LCoS based ROADMs with up to 9 degrees and cascaded Nx1 WSS blocks for the colourless tributary side at Rx. In addition as a transport technology, uncompensated schemes are generally deployed (regional and national domains), so that only coherent technologies are considered in the network.

Telefónica network topology is shown in Figure 16 and Figure 17.

Page 40 of 99





Figure 16: Telefónica reference network. Regional domains

.

Figure 17: Telefónica reference network. National domain

Page 41 of 99





SMF fibre is used on all the network segments and links. It is assumed that the oldest fibre layouts have been renewed and the fibre links are in a relatively good state. The fibre used is SMF G.652-D with the following physical characteristics:• Chromatic dispersion at 1550nm: 16.5 ps/nm.km• Chromatic dispersion slope at 1550 nm: 0.057ps/nm².km• Attenuation coefficient: following a normal distribution with mean 0.22 dB/km and standard deviation of 0.0068 dB/km. Minimum/maximum attenuation on a fibre segment: 0.1997/0.2371 dB/km.• PMD coefficient: the maximum value for a fibre link is 0.3995 ps/sqr(km), while the minimum is 0.0213 ps/sqr(km).

An Excel spreadsheet file with link and amplifier details for the regional domains and for the national domain is provided.

A preliminary static traffic matrix for current demand volumes over the Transit/Express domain of Telefónica network has been provided by means of a specific Excel file. It only considers aggregate flows at the higher hierarchy data network level (here named “transit”) whose nodes (“TRx”) are shown in Figure 17, as all the transit level routers are connected through the express domain. Detailed traffic demands are provided through a separate additional Excel file.

4.5 British Telecom Reference NetworksBritish Telecom has provided two reference optical networks, covering different network sizes to aid with different network studies. Both networks carry both residential and business traffic. All fibres in the British Telecom network are G.652 SSMF and typical values for parameters can be used for system design purposes.

The first reference network is a large, flat fibre network connecting 1113 nodes with 1956 fibres linking the core, metro and primary aggregation sites in the network. There is no inherent hierarchical structure in this network, apart from indicating the class of node. Fibre distances in this network range from 1 km up to 295 km with an average of 23.8 km and a nodal degree range between 2 and 17, with an average of 3.5. The majority of the aggregation network consists of chains, resulting in a large number of nodes with a nodal degree of 2. The network diameter of this network ranges from 1 km up to 1694 km, with an average diameter of 386 km. The full network is too large to be plotted in an understandable way, but is specified in terms of node and link lists.

A smaller reference network representing just the national core network connectivity is also provided. This network consists of 22 nodes connected by 35 fibre links. Fibre distances in this smaller network range between 2 km and 686 km with an average of 147km, and the nodal degree range is between 2 and 4 with an average of 3.2. The topology of this network is shown in Figure 18 along with a table showing the nodal degree distribution for both reference networks.

Page 42 of 99





5km59km

55km

115km

2km

7km

183km

105km

163k

m

73km

275km163km

686km

48km

240km

182km

120km

226km

20km

23km

127km

89km

27km

203km

75km

99km419km

87km109km

62km160km

439km 234k

m 1

2

17

19

614

5

13

7

12

20

8

4 93

10

11

71km

197k

m

18

1516

21

22

22 node1113 node

2 1 4983 16 2924 5 1225 656 307 238 149 1610 1811 612 1113 814 515 316 117 1

Nodal Degree

Number of Nodes

Figure 18: Topology of the British Telecom 22 node core reference network, including a table showing nodal degree distribution of both reference networks.

The static traffic matrices provided for these networks use the number of residential and business premises being served from each node to determine traffic volumes. The matrices provide peak traffic levels, therefore representing the maximum traffic load that the network has to support. The base matrices provide traffic levels for 2013 and these can be scaled by appropriate growth rates to obtain traffic levels for subsequent years.

4.6 Telecom Italia Sparkle European NetworkSparkle is the global telecommunication operator owned by Telecom Italia.

On the left side of Figure 19 is shown the worldwide structure of interconnection of the Sparkle network, while on the right side of the same figure the European portion of the worldwide Sparkle network (which is proposed as a reference network for Idealist studies) is depicted.

Page 43 of 99





Figure 19: Telecom Italia Sparkle worldwide network.The European Sparkle network is a transport infrastructure with 50 nodes and 69 fibre links. The type of fibre, as well as the number of sections on the link is known for the central part of the network; but these details are not available for the extensions in the East (Poland, Hungary, Romania, Greece) and West (Spain and Portugal). Topology data is provided, comprising a list of nodes identifiers and list of links. Link characteristics always include the distance, type of fibre, and number of amplifiers where known.

Currently most of the nodes in the central part of the network are hybrid OTN and optical nodes. Equipment comes from many vendors and so the optical network infrastructure has to be organized by interconnected islands, which impacts both optical transparency and network management. Optical nodes are ROADM based on 1:9 WSS modules. The central part of the network is designed for coherent transmission of circuits up to 100 Gb/s. Traffic grooming is done by an OTN switch, while the optical parts of the nodes perform the wavelength routing. Longest paths have a length of thousands km, and one or more regenerations are necessary.

Where known, the types of fibre are G.652 and G.655 (most of them are Truewave-RS, the remaining is Corning Leaf). For system design purposes, typical values can be used for the main fibre parameters (e.g. attenuation, dispersion, nonlinearity, parameters related to Raman Effect).

The traffic carried on the network are static circuits, which range from basic SDH and OTN circuits (from STM-1 or ODU0 respectively, grooming is handled by SDH or OTN part of the transport nodes) to 10 or 40 Gb/s circuits directly provided on DWDM equipment. Circuits often need very high requirements for protection (1+1 or more). For confidentiality reasons the traffic data cannot be disclosed.

4.7 SummaryThe general characteristics of Reference Networks collected in WP1 are summarized in Table 2. At present stage the IDEALIST project can relies on six networks of different type and geographic scope, from Nationwide to European Continental. All the networks are available with the topological details and often with the characteristics of the fibers and other valuable features (for instance optical amplifier positions and span length) that allows to perform an accurate network design. Topological features like node degree and link length (average and maximum values) are reported in Table 3. For most of the reference networks the traffic demand is also available but limited to the static version as this is the what it is possible to recover in the transport networks today. Dynamic traffic can be

Page 44 of 99





generated by traffic engineering modeling in waiting of real data that are expected to be collected during the progress of the project.

Table 2: Main Features of Reference NetworksOperator Location Segment covered Main features

TI Italy Core (National)Flat National 44 nodes network, mainly but not exclusively for carrying IP backbone traffic; mainly G655 and G652 and few G653.

DT Germany Core (National)

Flat 12 PoPs National Core, physically installed twice (12+12 nodes) to serve exclusively the IP core network, fiber is wholly G.652.

TID Spain Core (National part) Two levels National optical network, 30 nodes National Core, G.652 fiber.

TID Spain Core (Regional part) 5 Regional networks (30 nodes each); G.652 fiber everywhere.

BT Great Britain Core, Metro and Aggregation

1113 nodes network connected by a G652 fibre infrastructure. No inherent hierarchy but sites classified as Core, Metro or Aggregation.

BT Great Britain Core (National) 22 nodes flat core network with G652 fiber links.

TI Europe Core (Continental) Flat 49 nodes network. Fibers on the links are G.652 or G.655.

Table 3: Topological characteristics of Reference Networks

Operator Location Nodes Nodal degree Links Link length [km]average max average max

TI Italy 44 3.2 5 70 174 482DT Germany 12 3.3 5 20 243 485

TID Spain (National) 30 3.7 5 56 148 313

TID Spain(5 Regions)

30(each Reg.) 3.5 5 53 73 185

BT Great Britain 1113 3.5 17 1956 24 295BT Great Britain 22 3.2 4 35 147 686TI Europe 49 2.9 5 69 393 1212

Page 45 of 99





5 Techno-Economic Analysis

5.1 IntroductionThis section reviews and updates the CAPEX model first developed in the STRONGEST project, including updated information for Layer 3 components and a change of cost baseline from 10 Gb/s transponder to 100 Gb/s coherent transponder. There is a first study in the potential cost of a Sliceable Bandwidth Variable Transponder (SBVT). It also provides the basis for a range of OPEX related models that will be further developed in Idealist. Finally there is an in-depth analysis of the energy consumption in various kinds of fixed and flexgrid transponders – information that will be essential to allow fair comparison between them.

5.2 CAPEX Model

5.2.1 Summary of STRONGEST modelThe cost model to be applied in Idealist takes as a reference the cost model developed in the STRONGEST project that in its final version was published in [17].

As the STRONGEST project was focused on the whole core network, including all the network layers and technologies from layer 1 to layer 3, the cost model was developed for equipment in four technology areas: IP/MPLS, MPLS-TP, OTN and WDM, the last one in two versions: fixed and flexgrid. The multi-layer model, including possible interconnection schemes, is shown in Figure 20.

The STRONGEST model adopted as cost units relative to the price of one non-coherent transponder at 10 Gb/s with a reach of 750 km on a compensated SSMF fibre. This cost unit was recognized as the STRONGEST Cost Unit (SCU). All other devices or subparts of pieces of equipment are expressed with reference to that unit. This enables the decoupling of the cost values from any actual currency, and facilitates the collection of data because price data is considered highly confidential by most of the involved players. The model deals with price and not cost values. A practical reason for that is that price data is easier to collect than industrial cost (i.e. the pure cost without profit margins); fortunately the price values, however, determine how much an operator actually has to pay for deploying a network and this is exactly what it is required for techno-economic evaluations. Nevertheless, in general, the terms ‘price’ and ‘cost’ are used, and continue to be used in IDEALIST interchangeably.

The main characteristics of the STRONGEST CAPEX model are presented below:

IP/MPLS, MPLS-TP and OTN are modelled with blocks organized at three levels: chassis, cards, and interfaces (or transceivers). Chassis are characterized by their capacity in terms of slots, the cards in terms of capacity (throughput) and type and number of ports, and the interfaces in terms of capacity (client rate), framing (i.e. Ethernet or POS) and transmission characteristics (grey or coloured, for coloured cards reach). Cards of any type are supposed to always require one slot in the chassis.

The WDM layer is modelled in two versions: fixed grid and flexgrid.

The WDM fixed grid model includes transponders and equipment that work according to the standard 50 GHz grid. The model covers only current and short-term devices and the highest rate transponder is 400 Gb/s in 50 GHz and 150 km reach, suitable only in the

Page 46 of 99





metro context. Concerning ROADM, four models are considered: (i) a basic one with add drop on lines and use of AWG and interleavers; (ii) a colourless one; (iii) a colourless and directionless one; and (iv) a fully flexible one (colourless directionless and contentionless). The basic blocks are identical for all the models and are the following ones: amplifiers (pre, boost), AWG, interleaver, 1x9 WSS.

Flexgrid includes transponders and equipment that work according to flexgrid as defined by the ITU-T. In particular a set of reconfigurable transponders are modelled as a transponder that can accept a change in the rate of the client signal and as a consequence the transmission rate without changing the optical allocated bandwidth. This can be considered a first step in introducing flexibility into the optical network; in fact it allows one to allocate a bandwidth different from the one allowed by the fixed grid, but it doesn’t permit a change in bandwidth allocation for the connection.

For all the layers, components with different capacities and functionalities are considered to be available over different time periods. With some exceptions these periods are: the actual at the time of model finalization (year 2012), the short-term (2015) and the mid-term (2018).

Figure 20: Summary of basic building blocks of STRONGEST CAPEX model

5.2.2 Idealist CAPEX modelIDEALIST is focused on the optical layer and in particular on flexgrid devices and networks, so the area of interest is restricted to the optical transmission and switching in layer 1. Nevertheless, as the main client of the optical networks is currently the IP network and it is expected that it will continue to be true over the time duration of the Idealist project, in order to allow making significant techno-economic evaluations at the network level, the IP/MPLS part of the CAPEX model is retained and updated. According to this assumption,

Page 47 of 99





the Idealist CAPEX model covers three types of equipment: IP/MPLS, WDM fixed grid, and flexgrid for elastic optical networking (flexgrid and flexible rate).

Reference Cost UnitWith regard to the Cost Unit, i.e. the cost used as a reference for all the elements included in the model, this has been updated for the Idealist model. As nowadays (year 2013) the state-of-the-art in the transponder technology is now coherent 100 Gb/s (this being the highest rate device commercially supplied by many vendors with DP-DQPSK modulation, a baud rate of 32 Gbaud, with soft decision FEC, and a reach of about 2000 km on SSMF fibre), the Idealist Cost Unit (ICU) is defined as the cost of such a 100 Gb/s device (rather than non-coherent 10 Gb/s). All other devices and subparts of equipment are priced with reference to the ICU.

In the case where an Idealist case study requires OTN or MPLS-TP, the CAPEX model of STRONGEST is applied with the use of an appropriate cost conversion factor. Such a factor is calculated as follows: 12.5 SCU = 1 ICU.

Time ReferencesTwo time horizons in addition to the present starting period (year 2013) are chosen to define the commercial availability of future equipment and devices. They are a short-term period (year 2015) and a medium-term period (year 2018). In fact it is not important which year exactly in the future the aforementioned equipment will be placed on the market; rather, the important thing is to note that we assume that there will be two successive technology generations as from today. These two steps correspond to the availability of slots and interfaces at 400 Gb/s and 1 Tb/s respectively.

The current period is characterized by 200 Gb/s on line cards and up to 100 Gb/s electrical interfaces on electrical switching equipment, and by non-coherent 100 Gb/s in 50 GHz for transponders in the optical domain. In the short-term (year 2015) both line cards and interfaces at 400 Gb/s are expected to be available in electrical switching equipment, and fixed and flexgrid transponders at 400 Gb/s are estimated to be ready in the optical domain. SBVT at a 400 Gb/s line-rate will also assumed to be available, but the details of such devices is still under definition in the project. In medium-term (year 2018) the speed of line cards and interfaces on electrical switching equipment and the optical layer is expected to be 1 Tb/s. At the optical layer the 1-Tb/s line speed is assumed to be available for BCT, BVT and SBVT.

Idealist CAPEX model Excel file In the appendix of this document an Excel format file is included, composed of four sheets which include the three technologies (IP/MPLS, WDM fixed grid, EON flexgrid) plus a sheet dedicated to transceivers.

CAPEX model for IP/MPLSThe IP/MPLS model is organized (as in the STRONGEST Model) into three levels: the basic node, the line cards and the transceivers. The basic node includes the chassis (single- or, for core routers only, multi-chassis) the physical and mechanical assembly, the switch, power supplies, cooling, and control and management plane hardware and software. It also provides a specified number of bidirectional slots with a nominal (maximum) transmission speed (also named “slot capacity” in the model). Into each slot, a line card of the corresponding (or lower) speed can be installed. Each line card provides a specified number of ports at a specified speed and occupies one slot of the basic node. In each port, a transceiver can be plugged in. Depending on how it is configured, a router port

Page 48 of 99





can forward packets based on IP addresses, on MPLS labels, or (if it is an Ethernet card) on Ethernet media access control (MAC) addresses. Transceivers can be grey or coloured; in the case of coloured transceivers, if optical compatibility is satisfied, the output signal from the transceiver does not require an OEO to be transmitted on an optical infrastructure. Innovative devices, such as BVT / flexgrid transceivers or those implementing SBVT functionality (with client interfaces embedded in the line card) are also modelled.

Two categories of routers can be distinguished: a single-chassis router for metro nodes (in particular two sizes are considered: 10 and 20 slots) and a scalable multi-chassis router for large core nodes, with up to 72 chassis, with a single 16 slot capability shelf on each chassis.

This means that core routers are capable of a minimum of 16 (one shelf) to a maximum of 1152 (72 shelves) slots for hosting line cards. Table 4 reports the cost for the metro routers and for some core routers. The cost P of multi-chassis core routers (routers that require two or more shelves) can be computed according to the formula (1) selected from [17] and properly revised taking into account that the reference cost unit is different for the Idealist model. Equation (1) is derived according to the modular structure of equipment supplied by a specific vendor.

(1)

Where

C is the total switching capacity in Tb/s required to the router, and K the capacity of a fully equipped shelf, which depends on the reference year, where in particular K= 3.2 Tb/s in 2013, K= 6.4 Tb/s in 2015, and K= 16 Tb/s in 2018.

It is important to note that the cost of a core router has a significant increase from 1 chassis (P=4.3) to 2 chassis (P=22.9), but for more than two shelves the cost has an almost linear growth as Figure 21 shows for a number of shelves up to 20.

Figure 21: Basic router cost in function of the number of shelves.

Page 49 of 99





Table 4: IP/MPLS Basic Node costBasic Node for Metro Router

CapacitySlot @ 200 Gb/s

Year 2013


Year 2015

CapacitySlot @ 1 Tb/s

Year 2018Provided slots Cost (ICU)

2 Tb/s 4 Tb/s 10 Tb/s 10 Slot 0.424 Tb/s 8 Tb/s 20 Tb/s 20 Slot 0.83

Basic Node for Core RouterCapacity

Slot @ 200 Gb/sYear 2013


Year 2015

CapacitySlot @ 1 Tbps

Year 2018Provided slots Cost (ICU)

3.2 Tb/s 6.4 Tb/s 16 Tb/s 16 Slot 4.306.4 Tb/s 12.8 Tb/s 32 Tb/s 32 Slot 22.879.6 Tb/s 19.2 Tb/s 48 Tb/s 48 Slot 28.92

12.8 Tb/s 25.6 Tb/s 64 Tb/s 64 Slot 44.05

16 Tb/s 32 Tb/s 80 Tb/s 80 Slot 50.07

… … … … …230.4 Tb/s 460.8 Tb/s 1152 Tb/s 1152 Slot 8329.02

Table 5: Line cards costLine cards for 200 Gb/s slot

Interface type Available Cost Core (ICU) Cost metro (ICU)20 x 10 GE / MPLS-TP 2013 2.56 1.995 x 40 GE / MPLS-TP 2013 2.88 2.14

2 x 100 GE / MPLS-TP 2013 2.74 2.35Line cards for 400 Gb/s slot

Interface type Available Cost Core (ICU) Cost metro (ICU)10 x 40 GE / MPLS-TP 2015 2.56 1.994 x 100 GE / MPLS-TP 2015 2.88 2.141 x 400 GE / MPLS-TP 2015 2.74 2.35

Line cards for 1 Tb/s slotInterface type Available Cost Core (ICU) Cost metro (ICU)

10 x 100 GE / MPLS-TP 2018 2.88 2.142 x 400 GE / MPLS-TP 2018 2.74 2.35

1 x 1000 GE / MPLS-TP 2018 2.91 2.56

Table 6: Transceivers costTransceivers grey, short reach

Interface type Available Cost (ICU)10G 2013 0.00840G 2013 0.032

100G 2013 0.080400G 2015 0.256

1T 2018 0.512Transceivers, coloured in 50 GHz fix grid

Interface type Available Cost (ICU)

Page 50 of 99





40G, 2500 km 2013 0.48100G, 2000 km 2013 1.00400G, 150 km 2015 1.36

Bandwidth Variable Transceivers, coloured in flexgridInterface type Available Cost (ICU)

Transponder 1 (100G + AUA) 2013 1.44Transponder 2 (400G + AUA) 2015 1.76

Transponder 3 (1000G + AUA) 2018 2.00Transponder 400G 2015 1.20

Sliceable Bandwidth Variable Transceiver in flexgridInterface type Available Cost (ICU)

SBVT 400G type 1 2015 T.B.D. SBVT 400G type 2 2015 T.B.D.

SBVT 1T type 1 2018 T.B.D.SBVT 1T type 2 2018 T.B.D.

Concerning the line card capacity, the value chosen for the three time horizons are:

Current with 200 Gb/s of line card capacity and up to 100 Gb/s of port capacity (two ports at 100 Gb/s or 5 ports at 40 Gb/s are configurations available for exploiting the whole slot capacity).

Short term with 400 Gb/s for both line card and port capacity Medium term with 1 Tb/s for both line card and port capacity.

Assortments of transceivers per each period and their cost are as in Table 6, with three main categories: bandwidth constant in fixed grid, bandwidth variable in flexgrid, and sliceable bandwidth variable in flexgrid.

The cost coloured transceivers in flexgrid are assumed to be identical to the cost of WDM transponders with the same optical characteristics. Transceivers (or integrated functionality) that implement SBVT are also introduced, but at this stage of the project since the device is still under development, the price is not yet assessed (in the table it is marked - To Be Defined (TBD)).

CAPEX model for WDM fixed gridThe basic standard 50 GHz WDM Layer [18] is assumed to use coherent transmission exclusively and have SSMF (G652) fibre as the reference physical medium.

The WDM fixed grid model is composed of the following building blocks:

• Transponders with line rates from 40 Gb/s to 400 Gb/s (with different commercial time availability), and the maximum transparent reach for each type. The 10 Gb/s transponders are not included because they use incoherent on–off-keying modulation formats and need a dispersion compensation link - a scenario that is not compatible with the main WDM model assumptions. In the 50 GHz grid a transponder with a data rate of 1000 Gb/s has a very limited reach and this makes it unfeasible for long-haul transmission systems: the model does not include any 1000 Gb/s transponders.

• Muxponders with different configurations covering all the line rates. • Regenerators: for each transponder, a corresponding regenerator is available.• Optical amplifiers: in an optical transmission line the amplification span is assumed to

be 80 km.• WDM nodes:

Page 51 of 99





1. WDM terminal, which has a nodal degree of 1.2. Fixed OADM, which is not remotely configurable.3. OXC which is remotely configurable, commonly denoted as a ROADM.

For all the WDM nodes, solutions with 40 and 80 channels at 50 GHz in the C-band are modelled, as well as 160 channels in an extended C+L-band. Some vendors offer WDM systems that exploit the C bandwidth further than the basic 4 THz, and this allows the possibility of 88 or even 96 channel systems at 50 GHz; the cost of all the devices operating in the C band is assumed to be the same regardless of the number of channels at 50 GHz the WDM system can handle. Basic components for WDM terminal and fixed OADM nodes are arrayed waveguide gratings (AWGs), interleavers, and optical amplifiers. The different types of OXCs are characterized by the three basic following features: directionless (any client port can be connected to any node/line port); colourless, (the port of client interface can be tuned to any wavelength); and contentionless (the node architecture does not imply any wavelength switching contention).

Basic components for the OXC are: 1x9/9x1 WSS, 9x9 WSS, 1x20/20x1 WSS, optical amplifier, and passive optical splitter and combiner. The impact of the cost of splitters and combiners are considered negligible in such a WSS-based architecture. Details on how basic components can be combined together to build up an OXC, and the associated formulas to calculate the cost of the equipment are given in the Idealist CAPEX Excel file, where eight different OXC models are proposed: four for smaller sized nodes (node degree less than 10), and 4 four larger sized nodes (nodal degree less than 21). The four models for the smaller sized node implement: the basic (i.e. add and drop are on the line) colourless; the colourless and directionless; and the full flexible (colourless directionless and contentionless) ROADM, respectively. The same holds for the higher sized node. All the OXC architectures are realised with WSSs both at the input and output, implementing the so-called route and select architecture (see [19]). Cost of OXCs with architectures different from the ones included in the Idealist CAPEX Excel file, such as the broadcast and select ROADM architectures, can be also evaluated, as they can be built up relying on the building blocks defined by the model.

Figure 22 gives an example of a fully flexible colourless, directionless and contentionless route-and-select ROADM made with the components listed above; the shown architecture allows a node of up to degree 20 to be built up, if 1x20 (and 20x1) WSS modules are used. This fully flexible ROADM requires one add/drop chain for each network degree and this implies the use of 6 WSS modules for each node degree (2 modules for the input line, plus 2 modules for output lines, plus 2 modules on the add/drop chain - one on the add section and the other one on the drop section) plus a number of WSS modules on each add/drop chain, depending on the whole number of channels to be added and dropped in the node.

This leads to the following formula for the cost computation for the fully flexible route-and-select OXC:

(2)

where N is the node degree (up to 20 in case of 1x20 WSS), Amp_boost, Amp_pre, Add_amp, Drop_Amp are the cost of amplifiers (supposed equal in the model, see Table), WSS(1x20) is the cost of WSS modules (both types 1x20 and 20x1) and AD(%) is the Add Drop percentage (evaluated on the whole traffic crossing the node) with granularity of 20%. If the add drop percentage is 100% (all the traffic is terminated in the node, i.e. no pass

Page 52 of 99





through traffic) we have that 100(%)/20(%) = 5, such that the number of additional modules are 5x2xN; if the add/drop percentage is equal to or less than 20%, the number of modules are 2xN. (This approach is always valid for systems in the C-band with up to 100 channels as is assumed in the model. For systems with 80 channels or less, this leads to an extra dimensioning issue when adding the add/drop WSS modules; in particular for add/drop percentages greater than 80%, from the formula above we require five modules on each add and on each drop chain; but five modules are capable of 100 channels, and so they will exceed the capacity of the 80 channel system.

Figure 22: OXC c/d/cAn extract of the cost parameters in the CAPEX WDM fixed grid model is given in the following tables. Table 7 includes the basic fixed grid transponders and muxponders. Muxponders provide interfaces at 10 Gb/s, and this allows interconnection of WDM equipment to other equipment, for instance IP/MPLS routers, that require to be interconnected by 10G grey interfaces.

Table 8 includes all the basic components needed to build up optical line amplifiers (OLAs), WDM terminals, OADMs and ROADMs of different types. The cost value for the OLA is given for devices designed to span 80 km, and for both the versions operating in C-band only and in C+L-band. DGE (dynamic gain equalizer) functionality has to be inserted at least after every 4 spans and is assumed to be always present in terminals and nodes. Amplifiers located at nodes are all considered have the same price, even if they have different characteristics (e.g. preamplifier for line in, booster for line out, amplifiers for add or drop lines). Concerning the OXC basic components, WSS modules for both filtering directions (i.e. WSS 1x20 and WSS 20x1) are assumed to have the same price. The passive components that can be used as a splitter or as a combiner are considered to have a negligible impact on the whole cost of an optical node (which includes many costly WSSs and amplifiers) such that their cost is set to 0.00.

Page 53 of 99





In Table 7 and Table 8 the last column includes the required slots on a chassis of the corresponding component.

The STRONGEST model provided a framework to organize the node by the use of chassis only for IP/MPLS, MPLS-TP and OTN equipment, while for WDM the concept of chassis was not modelled. The costs of the common parts of basic node were spread and attributed to the subparts. The Idealist CAPEX model now also introduces for WDM equipment the chassis as an element to host the node subparts. In particular three types of chassis are introduced.

A small chassis called the Basic chassis, comprising one shelf of 10 slots that can host up to 6 slots for optical components and 4 slots for common services (power supply, communication and control). The Basic chassis can be used for OLAs and Terminals. A dual shelf chassis has to be used in the situation where a system requires more than 6 slots.

Two types of dual shelf chassis are defined: Master chassis and Slave chassis that can be combined together to build up multi-chassis nodes. A piece of equipment that needs more than 6 slots, always requires a Master and then a number of Slaves, depending on the capacity in terms of slots required by the complete system. A ROADM can require tens of chassis in the case of high degree nodes or a high percentage of terminated traffic.

The capacity of the Master chassis in terms of slots is lower than the Slave chassis, because on the Master the service cards require more space. The capacities are 14 for the Master and 16 for the Slave.

Cost of the chassis is given in Table 9. To take the chassis cost into account, it is necessary to evaluate the total requirements for the complete system in term of slots and then determine the chassis configuration required and the cost as a consequence. The resulting total cost is the cost of the components plus the cost of the chassis.

Table 7: WDM fixed grid transponders and muxponders

Table 8: Basic WDM line and node components

Page 54 of 99

Transponder for fixed grid 50 GHz Component Type Available Cost (ICU) Required slot

40G, 2500 km 2013 0.48 1100G, 2000 km 2013 1.00 2400G, 150 km 2015 1.36 3

Muxponder for fixed grid 50 GHz Component Type Available Cost (ICU) Required slot

40G Muxponder, 4 x 10G, 2500 km 2013 0.40 1100G Muxponder, 2 x 40G, 2000 km 2013 1.07 2

100G Muxponder, 10 x 10G, 2000 km 2013 0.87 2400G Muxponder, 10 x 40G, 150 km 2015 1.60 4400G Muxponder, 4 x 100G, 150 km 2015 1.44 4

Basic line system, terminal and node components for fixed grid 50 GHzComponent Type Available Cost (ICU) Required slot

OLA (bidirectional, C band, 80km span) 2013 0.15 2DGE functionality (one every 4 spans) 2013 0.16 2OLA (bid., 80km span for C + L Band) 2013 0.30 2Opt. Amplifier (unidirectional, any type, for nodes) 2013 0.06 1AWG (40 channel) 2013 0.07 1Interleaver (80 channel) 2013 0.04 1WSS 1x9/9x1 (unidirectional) 2013 0.32 1WSS 1x20/20x1 (unidirectional) 2013 0.48 2WSS 9x9 (unidirectional) 2013 3.84 2Splitter/combiner (any type) 2013 0.00 1





Table 9: WDM chassis

CAPEX model for flexgrid elastic optical networksThe CAPEX model for flexgrid systems is under definition as the work of developing the interfaces and node components are ongoing mainly within the WP2.

This first and incomplete version of the CAPEX model for flexgrid systems is based on the following assumptions.

The transponders that can be used in flexgrid EON belong to the three types:

1. Fixed grid transponders (the same as in the fixed grid model and included in Table 7)2. Bandwidth Variable flexgrid transponders (in line with the ones in the STRONGEST

model - they can change the client rate but not the allocated optical bandwidth)3. Sliceable Bandwidth Variable Transponders SBVT

Bandwidth Variable transponders and SBVT are reported in Table 10 with their characteristics known up to now. The acronym AUA means “Also Usable As” and refers to the feature of transponders classified as 1, 2 and 3 in the table to work with different client rates (by means of a change in the modulation format and/or in the baud rate) and resulting reach. SBVTs are devices that map N client signals into M optical signals with N≥M, where the optical signals can have different transmission characteristics to each other (carried rate, modulation format, bandwidth allocation) and in general a not contiguous allocated spectrum resulting from the M different optical modulated signals. On the line side all the signals (slices) of a SBVT are put on the the same physical output (a couple of fibers, one for ech transmission direction) and filtering to is demanded to the interconnected devices, for instance LCoS WSS for filtering and routing to differentiated destinations.

SBVTs as other components suitable for flexgrid are under specification in IDEALIST and their charcteristics are not known at the time this deliverable is due. It was decided by the project that almost a couple of devices should be developed in each timeframe scenery and this is the reason in Table 10 a couple of devices for both Year 2015 (400 G) and Year 2018 (1T) are inserted. The two alternatives for each timeframe will consent comparative techno-economic evalautions within flexgrid technologies in addition to the standard comparison with fixed grid.

In lack of specifications on SBVT some assumption on technical characteristics was done on hypothetical basic SBVTs expected available in 2015 and 2018. This assumption does not reflect any ongoing development in WP2 but constitute a first reference for cost parameters to be used in preliminary studies.

Concerning the nodes for flexgrid, the basic assumption is that they are of the same type as the OXC considered for fixed grid, but with flexgrid capability and this according to [18] implies the capability to handle the channel bandwidth with a central frequency resolution of 6.25 GHz, and to allocate bandwidth in slices of 12.5 GHz. WDM terminals can be obtained with WSS modules and are considered as an OXC of degree 1. The fixed OADM is not useful in a flexgrid context, as without flexibility in reconfiguration there is no benefit. The impact on the cost of flexgrid nodes implies an increased cost with reference to the same components or equipment that are appropriate for fixed grid. In particular the cost

Page 55 of 99

Chassis for hosting equipment partsComponent Type Available Cost (ICU) Slot capacity

Single shelf Basic Chassis (for OLA and Terminal) 2013 0.2 6Dual shelves Master Chassis (for ROADM) 2013 1 14Dual shelves Slave Chassis (for ROADM) 2013 0.8 16





increase is due to the complexity in the management of the bandwidth portions, and less due to the switching technology (LCoS for instance is already deployed for fixed grid in some networks which are flexgrid-ready). In the STRONGEST model, this cost increase is simply evaluated as increasing percentage and was estimated at 30%. Within IDEALIST the same approach is retained but the increasing percentage is considered too high and the suggested increasing percentage for CAPEX parameters is 20% with reference to the correspondent parameters for fixed grid component. IDEALIST are developing in WP2 node subparts that could not be modelled with the previously described framework. As for SBVT case the CAPEX model concerning flexgrid nodes will be updated during the project when novel node architectures and related components will be specified.

Table 10: Transponder for flexgridBandwidth Variable Transponders in flexgrid

Interface type Specification Available Cost (ICU) Required slotTransponder 1 100G, 50GHz, 2000km AUA 40G,

50GHz, 2500km 2013 1. 2Transponder 2 400G, 75GHz, 500km AUA 200G,

75GHz, 2000km AUA 100G, 75GHz, 2500km 2015 1.76 4

Transponder 3 1000G, 175GHz, 500km AUA 500G, 175GHz, 2000km 2018 2.00 6

Transponder 400G 400G, 100GHz, 1000km 2018 1.20 2100G Muxponder, 2 x 40G + Transponder1 2013 1,52 2100G Muxponder, 10 x 10G + Transponder1 2013 1,28 2400G Muxponder, 10 x 40G + Transponder2 2015 2,00 3400G Muxponder, 4 x 100G + Transponder2 2015 1,84 3400G Muxponder, 10 x 40G + Transponder 400G 2018 1,44 3400G Muxponder, 4 x 100G + Transponder 400G 2018 1,28 31T Muxponder, 10 x 100G + Transponder3 2018 2,24 61T Muxponder, 2 x 400G + Transponder3 2018 2,08 6

Sliceable Bandwidth Variable Transponders in flexgridInterface type Specification Available Cost (ICU) Required slot

SBVT 400G basicInput: up to 10 client interfaces (10400 Gb/s)

Output: up to 4 slices on C band (100400 Gb/s)Max overall rate on output slices: 400 Tb/s

Perfor. on 400G slice: 400G, 150 GHz, 2000km

2015 3.00

SBVT 400G type 1 T.B.D. 2015 T.B.D. T.B.D. SBVT 400G type 2 T.B.D. 2015 T.B.D. T.B.D.

SBVT 1T basicInput: up to 10 client interfaces (100 - 400G) Output: up to 10 slices on C band (100G-1T)

Max overall rate on output slices: 1TPerfor. on 1 T slice: 1T, 375 GHz, 2000km

2018 4.00

SBVT 1T type 1 T.B.D. 2018 T.B.D. T.B.D.SBVT 1T type 2 T.B.D. 2018 T.B.D. T.B.D.

IDEALIST CAPEX model details are included in this Excel file:

Page 56 of 99





5.3 Target cost for Sliceable Bandwidth Variable TranspondersThe aim of this study is to identify the target cost of 400 Gb/s and 1 Tb/s SBVTs to reduce, by at least 30%, transponder costs in a core network scenario. This target cost is calculated in relation to estimations for non-sliceable transponders of 400 Gb/s and 1 Tb/s. In light of our results, cost savings of 30% are feasible for 1 Tb/s transponders in the next nine years with a higher cost than non-sliceable transponders. Savings of 30% for 400 Gb/s case are possible in the short-term before 1 Tb/s SBVTs can appear in the market. Feasibility of such savings with a target cost higher than current non-sliceable transponder shows that SBVT can be a reality.

5.3.1 Case study definitionThe objective of this study is to calculate the cost that a SBVT should have in order to achieve a percentage cost reduction in transponders for a backbone network. To this end, two node models are compared: (a) current model without sliceable transponders and (b) node with SBVT (Figure 23). Previous architectures to interconnect client equipment and SBVT are possible choices for implementation, but they are out of the scope of the paper. The main difference between both models is that the non-sliceable transponder model requires least one interface for each destination, while the SBVT transponder reuses hardware and optical spectrum to transmit to multiple destinations.

We consider coherent modulation formats in the model without sliceable transponders (40Gb/s, 100Gb/s, 400Gb/s and 1Tb/s), while 400Gb/s and 1 Tb/s SBVTs are used for the second model. Let us remark that the maximum traffic rate is the same in both models. Therefore, when studying 400 Gb/s SBVT model, there are no 1 Tb/s transponders in the non-sliceable transponder model.

(a) Non-sliceable transponders model (b) SBVT model

Figure 23: Models for the study with and without SBVTA network model based on the Spanish backbone is used for this study (Figure 24) [20]. The network is made up of 20 edge nodes that aggregate traffic from transit and access routers and forwards it over wavelength channels requested to the photonic mesh. In this analysis OXCs are assumed to switch the optical channels in an elastic manner, but they do not have any influence on transponder requirements, as they do not implement transponders themselves. We assume that there is enough optical resources in both situations with and without SBVTs and the objective of this work is to reduce the investment in transponders.

Page 57 of 99





The initial traffic matrix is created based on information for the Telefonica backbone network in 2012. The link dimensioning is done using an over-dimensioning factor of 30%. For instance, given a traffic demand of 35 Gb/s, links are dimensioned for 45.5Gb/s (35Gb/s x 1.3). Traffic is incremented yearly by a 50% factor in order to compare the cost performance of the different architectures proposed over the next 10 years. In order to assess the cost of non-sliceable model, the STRONGEST cost model is used in this study [17]. Table 11 contains information about the relative unit cost used for the non-sliceable transponders of different bit rates.

TxP parameters Cost40Gb/s, 2500km, 50 GHz 6100Gb/s, 2000km, 50 GHz 15400Gb/s, 75GHz, 500km 221000Gb/s, 175GHz, 500km 25

Figure 24: Reference network based on Spanish national backbone [11]

Table 11: Non-sliceable transponders cost [12]

5.3.2 Case study resultsFigure 25 shows the target cost of the SBVT (bars) to achieve 30% overall savings in transponder costs for 400 Gb/s and 1 Tb/s SBVTs across the whole network. The cost for non-sliceable transponders is also shown in Figure 25, to illustrate that in some instances this 30% overall savings are possible even when unit cost of SBVT are higher than the cost of non-sliceable transponders.

Page 58 of 99





Figure 25: Target cost for SBVT (400Gb/s and 1Tb/s) and reference costs for non-sliceable transpondersFigure 26 shows the possible cost increment of SBVT in comparison with non-sliceable transponder cost. The target cost of 1-Tb/s SBVT steadily increases for the next three years to reach a peak of 140% the cost of a non-sliceable transceiver in 2015 (a total traffic demand of 12.5 Tb/s). For 400 Gb/s, the peak target cost is reached in 2013 (a total traffic demand of 5.5 Tb/s) and then it steadily drops so that in 5-6 years the cost of the 400 Gb/s SBVT should be similar to the cost of a fixed grid transponder to achieve overall 30% savings in the network. Therefore, when the price of 400 Gb/s SBVTs approaches the price of non-sliceable transponders, it would make sense to migrate some nodes to 1 Tb/s SBVT. Finally, we should note that we have not applied any discount or price degradation in our model, so costs remain the same along time. It is reasonable to expect decreasing non-sliceable 400 Gb/s and 1Tb/s transponder costs.

Page 59 of 99





Figure 26: Cost increment of the SBVT (400Gb/s and 1Tb/s) in comparison with non-sliceable transpondersIn addition to the target costs for 30% overall savings, we have also studied the maximum overall savings that can be achieved when SBVT costs are the same as those of non-sliceable transponders. To this end, we have computed transponders costs across all the network using non-sliceable transponders in three scenarios: (1) no savings, (2) 30% savings and (3) 50% savings. On the other hand, we have computed the required number of SBVTs to handle the same capacity and multiplied this number by the cost of non-sliceable transponders. Figure 27(a) and (b) show the results for 400 Gb/s and 1 Tb/s cases, respectively.

In light of these results, we can say that 50% savings is not possible in the case of 400 Gb/s, as shown in Figure 27(a). In the case of 1 Tb/s, more than 50% savings could be possible if the technology were available before 2018 (or when 42 Tb/s traffic is reached).

(a) SBVT capacity 400Gb/s

Page 60 of 99





(b) SBVT capacity 1Tb/sFigure 27: Comparison of transponders cost for Non-Sliceable model without savings and with 30% and 50% and minimum cost of SBVT

5.4 OPEX modelThe operational expenditure (OPEX) is the amount of money that network operators spend on an ongoing, day-to-day basis in order to run their business. According to [21] the OPEX for an operator can be divided in to seven general categories: network operation, interconnection and roaming, marketing and sales, customer service, charging and billing, IT and general support, and service development. In particular, the OPEX category where Idealist technologies can play a role is the network operation, which includes OSS operation, maintenance and repair of the network elements, equipment and software licenses, rental of network resources, costs for site rental and electricity.

First, in order to have an insight to give an order of magnitude estimate for the network operation related to OPEX, [Yankee09] estimates that for a fixed line operator, the network operation takes a significant part of expenses, accounting for 39% of the total OPEX. Thus, it is key to be able to drive down network related OPEX.

In this section of deliverable D1.1, the main components of network operation related OPEX in which Idealist solutions can play a role, are modelled. The Idealist OPEX model will take into account:

Cost of floor space Cost of field operations and repair Spare parts (stock) maintenance. Energy (including cooling)

Page 61 of 99

Lord,A,Andrew,DES8 R, 06/21/13,

Missing reference from Oscar





5.4.1 Cost of floor spaceEvery network component occupies a certain amount of space and needs to be placed in a central office, or even in a field location (e.g. cabinet). The extent to which this impacts different operators varies enormously, depending on whether they are an incumbent (with existing buildings) or a new entrant. However, in order to be able to compare different network configurations and equipment, a simplified model needs to be applied. The Idealist OPEX model will consider a fixed price per square metre. The suggested value is the mean value of rental per square metre in the country of the network. The Idealist CAPEX model includes the size of the components.

5.4.2 Field operations and repair modelThe field operations and repair are a big part of a network operator’s expenses. Nowadays, there is a trend in carriers outsourcing the operations and maintenance to third-party vendors that specialize in specific network technologies. In this way, the third-party operation and repair provider is able to leverage its field force across multiple carriers. In this subsection, an operation and repair model is explained, together with a methodology to estimate, on the one hand the workforce needed to achieve a given service guarantee, and on the other hand a methodology to calculate the amount of spare parts (stock) that needs to be maintained.

The operations and repair model can be used to measure the OPEX savings that can be achieved by using the Idealist control plane technologies developed in WP3, which aim at enhancing the automation of the network and improve the network resilience.

Operators and service providers typically consider two main models:

Flat Fee Model: There is a fixed monthly cost during the year, which includes any number of events that may exist in the month.

Event Base Model: The cost is computed by the number of events multiplied by the price per event. An event can be the failure of a network component (e.g. transponder, router card) or an installation.

Idealist will consider a flat fee model based on the number of operation and repair teams that are need to be maintained to achieve a desired repair time.

Repair Time (Service Level) AvailabilityThere are two main availabilities considered:

8x5 - the operation and repair teams are available 8 hours a day, 5 days a week. 24x7 - the operation and repair teams are available 24 hours, 7 days a week.

Within the availability period, the Idealist operations model will consider a repair time of 5 hours.

Number of repair teams ModelThe goal of this model is to obtain the minimum number of repair teams needed to be maintained, in order to achieve a given Minimum Mean Time To Repair (MMTTR). Each repair team has an associated monthly cost (salaries), and can give an estimate of the labour cost of repair of the communication network. The model assumes that upon each failure detected in the network, the failure should be repaired within the desired MMTTR. Note that the MMTTR will depend on the resiliency mechanism of the network. In this

Page 62 of 99





sense, Idealist control plane solutions aim to maximize this value, in order to reduce the size of the repair teams.

The model takes as input the number of components subject to failure, the mean time between failures (MTBF), as well as the mentioned MMTTR.

Centralized reparation modelThis model is based on a centralized location of the repair teams. This means that the failure location is not taken into account. In this model, independent of the location, a fixed Mean Time To Repair (MTTR) has been assumed. The MTTR is measured from the moment that the repair team starts working on the failure (not the moment of the failure).

Optionally, location and travel time can be considered. In this case, the MTTR can be calculated as a travelling time (depending on the location) plus actual repair time. The travelling time is the time that the repair team needs for going from their base location to the failure location. The actual repair time is the time that the repair team needs to work on the failure.

Related to the MTTR is another important parameter - the Minimum MTTR. This is the actual time by which the failure must be solved. Based on this value the repair team could delay the start of the repair process. This additional delay can help to minimize the number of repair teams, as the same team could work on several nearby faults or close in time failures (or operations).

With the intention of understanding better the relationship between MTTR, MMTTR and number of repair teams, Figure 28 below shows the different cases that can be defined:

F1 F2 R1

MTTRmin(F2)

MTTR(F2)2

MTTR(F1)

R2

Figure 28: Relationship between MTTR, MMTTR and number of repair teams

In this case it is observed that the second failure is completed with the minimum time to repair, although its repair starts when the repair of the first failure finishes, therefore just one repair team would be necessary to repair the two failures, saving thus one repair team.

Page 63 of 99





F1 F2 R1

MTTRmin(F2)

MTTR(F2)2

MTTR(F1)

R2

Figure 29: Impact of second failure

Also, each repair team has a certain availability. Previously we mentioned the two available models, 8x5 and 24x7. In the 8x5 it is assumed that the repair teams are only available from Monday to Friday, eight hours a day. Thus, in the best case, one repair team is needed. On the other hand, in the 24x7 case, in the best possible scenario, 4 repair teams are needed, each of them working 40 hours a week.

The Mean Time Between Failures (MTBF) is another essential variable of the model. This variable provides information about the frequency that failures occur.

Modelling of the number of repair teamsBased on the number of failures, MTBF, MTTR, MMTTR and availability model (8x5 or 24x7) a random generation of failures needs to be done. The random generation of failures is based on an exponential distribution with mean MTBF. This distribution describes the time between events as a Poisson process, i.e. a process in which events occur continuously and independently at a constant average rate.

5.5 OPEX reduction by stock improvement with Sliceable Bandwdith Variable Trannsponders.

5.5.1 Rationale of the studyPart of the cost of maintenance and repair of the network is related to keeping a stock of spare parts. Whenever there is a failure in a network element, there may be damaged parts that need to be replaced. In order to be able to repair the network elements, a stock of spare parts for replacement needs to be maintained. Such stock can be maintained either by the network operator or a third party supplier. In any case, maintaining such stock is mandatory and translates into a yearly cost.

In the case of optical transport networks, spare transponders need to be stocked. Following current model, transponders of different rates are needed, and thus a number of each rate needs to be stored. However, with the newly proposed Sliceable Bandwidth Variable Transponders, it is possible to reduce the number of transponders and reduce the variety of transponders. This study aims to analyze whether equipping a network with Sliceable Bandwidth Variable Transponders instead of fixed rate transponders of multiple rates reduces the needs in terms of stock maintenance.

5.5.2 Methodology This study is based on a centralized stock model. It is assumed that a central warehouse has stored all the transponders and that, in the event of a damaged part event, the transponder is shipped from such central location.

Page 64 of 99





The study has been performed for the Spain Reference Network shown in section 4.4. The number of transponders needed has been first obtained for two cases, one in which different kinds of transponders are used (e.g. 40 Gbps and 100 Gbps), and a second case with 400 Gbps sliceable BVT. The, a MTBF (Mean Time Between Failures) has been selected. This is the mean time between two failures of a transponder. It has been assumed that all transponders have the same MTBF. Then, a set of simulations have been made where the transponders are set to fail according to an exponential distribution with mean MTBF. Every time an element fails, it is replaced and a new element is requested to factory. The MDT (mean delivery time) is the time to receive a new transponder from the vendor. In the study 3 months has been assumed. Then, it has been obtained which is the number of stock that needs to be maintained to guarantee a certain availability.

The steps of the study are summarized bellow:

1. Number of transponders needed is obtained

2. Failures in every transponder are distributed randomly in time based on an exponential distribution with mean MTBF.

3. At the same moment that failure happens, a spare transponder is taken from the stock and a replacement request to factory is made.

4. Stock accounting:

a. One is added to the stock counter of the given transponder if a failure happens.

b. One is subtracted to the stock counter when the replacement happens (a new transponder arrives from factory to the stock).

5. From the previous steps, the peak value of the minimum stock number to be maintained at the warehouse is obtained.

6. Once the maximum stock is known, steps 2-4 are repeated, reducing the maximum stock in one each time, with one remark:

a. When a failure happens, if the stock is zero, one is added to the number of failed cases. Otherwise, one is added to the successful case.

7. The percentage of the successful cases is obtained for each stock value, until the stock is one.

Finally, the study has been repeated with several sets of random variables, in order to achieve a higher accuracy.

5.5.3 ResultsThe starting point of the study is the set of results obtained in 5.3.2 for 40G and 100G fixed rate transponders and 400G SBVT transponders in the Spain Reference Network. In this first study, the traffic of the first year has been used to obtain the total number of transponders needed, accounting for 144 40G transponders and 16 100G transponder for the fixed case, and 26 400G SVBT for the other case.

When there is a mix of fixed rate transponders, in this case studies, 40G and 100G, there is a need to keep two types of transponders in stock. In particular, the results, presented in Figure 30, show that, to achieve a 99% availability, 10 40G transponders (7% regarding the

Page 65 of 99





total 40G transponders) and 2 100G transponders (25% regarding the 100G transponders) are needed. In contrast, just a total stock of 3 SVBTs is needed.

If the requirement is more restrictive, for example a 99’99% availability, the number of 40G transponders to be stocked raises to 17, and the number of 100G transponders in the warehouse is 5. In the sliceable case, a total stock of 5 SVBTs is needed.

Figure 30 Stock of 40G transponders

Figure 31 Stock of 100G transponders

Figure 32 Stock of 400G SVBT transpondersThe results are summarized in table

Page 66 of 99





99% Avail 99’99%avail

Fixed case 10 40G, 2 100G 17 40G, 5 100G

SBVT case 3 400G SBVT 5 SBVT

Table 12 Stock for 99% and 99,99% availabilityIn section in 5.3.2, the cost model estimates a relative price of 6 for the 40G transponders, of 15 for the 100G transponders and between 22 and 35 for the 400G SVBT. Based on those numbers, in the best case, for a 99,99% availability, the relative cost of the stock for the fixed case is 177, while for the SBVT case is 110, that is a 37% cost reduction. In the worst case, the price would be similar. The results shown in this preliminary study are obtained with the traffic for year 1 in the SVBT study. Next steps will consider different traffic mixes, as well as more transponder types.

5.6 Energy model for high data rate transpondersEnergy efficiency and carbon management are becoming increasingly important for ICT companies as they focus on driving down environmental footprint. It is expected that the next generation of optical networks will bring about greater energy efficiency than legacy networks. With this intent in mind, within the Idealist project we want to investigate the energy impact of the next-generation of high data-rate networks and study the possible improvement of the energy consumption due to the introduction of both elastic and sliceable transponders.

The present deliverable only presents the energy model associated to rate-adaptive transponders (working on a fixed 50 GHz grid), because we suppose that the capability of reducing the spectral occupancy of the optical signal does not have an impact on its power consumption. Concerning the power models for sliceable transponders, these will be the subject of further investigation during the project.

The architectures of the transponders, which are taken into account here, are only based on coherent transmission because it has been demonstrated to be the technology of choice, satisfying both the long reach and high data rate requirements [23]. All considered transponders exploit the two polarization modes for transporting the signal; for this reason the signal is said to be dual polarisation (DP), which prefix is used before the declaration of the modulation format used for encoding information.

5.6.1 Energy model for fixed data rate devicesIn this section we present the energy model relative to four high-rate transponders. Two among them are already commercially available for current optical networks and transports up to 40 and 100 Gb/s, called in the following TSP40 and TSP100. The two other transponders are possible solutions proposed by vendors to take their place in the next generation of optical networks for coping with the increase in the traffic, and will transport up to 400 and 1000 Gb/s, called hereafter TSP400 and TSP1000. In this project, transmission is supposed be bidirectional, such that all transponders are composed of an emitter and receiver supporting exactly the same capacity.

Figure 33 and Figure 34 depict two types of possible transponder realization that can be considered in the Idealist project and are used for building up the power model in the following sections. In Figure 33 we have presented the muxponder, where various client cards are connected to a transponder by black-and-white connections. The capacity of such client cards is lower than that of the transponder, and client cards with different

Page 67 of 99





capacities can be connected to it. For simplicity purposes, in the description presented we consider that all the line-cards have the same capacity: equal to 10 Gb/s when 40 and 100 Gb/s transponders are considered; and 100 Gb/s for 400 Gb/s and 1 Tb/s transponders. Inside the architecture of these transponders a framer/deframer has to be included so as to distribute the incoming data to the modulators to be coded and sent.

In Figure 34 the transponder is assumed to be directly plugged into the router. This architecture is called uplink and a backplane (that can be an OTN or IP router) guarantees the framer/deframer operations. In this structure the transponder does not include the client side part, comprising the client cards and framer/deframer (cf. Table 13).

In the rest of this document, we only describe a transponder based on muxponder, remembering that the uplink is easily deduced by maintaining unchanged the transceiver/receiver parts and by eliminating the client side section by the consumption assessment.

Transeiver/Receiver

Muxponder

XGb/s

XGb/s

XGb/s

…ClientSide

WDMTransmission

Uplink

Transeiver/Receiver

BackPlane

WDMTransmission

ClientSide

Figure 33 Architecture of a transponder realized with a muxponder, where N client cards are plugged on the transponder.

Figure 34 Architecture of a transponder directly connected to an electric backplane. This configuration is called uplink.

40Gb/s transpondersThe 40 Gb/s transponder is connected to N line cards (XFP) working at any capacity going from 2.5 to 40 Gb/s. For simplicity we gas assumed to have all the line cards equal and working at 10 Gb/s, hence N = 4. On the emitter side, we have a framer that aggregates data from the line cards and adds to them the appropriate forward error correction (FEC) overhead, providing two electrical signals at 20 Gb/s, which drive the modulators to create a BPSK optical signal per polarization. The baud rate is called the speed at which each symbol is transmitted; a DP-BPSK signal transmits one bit per symbol per polarization, giving for a 40 Gb/s signal a baud rate (R) of 20 Gbaud.

Figure 35 shows the scheme of a possible DP-BPSK transponder, with the emitter (upper part of the figure) and receiver (bottom part). During propagation, the signal is affected by phase noise and the symbols shift along the unitary ring. For this reason, four photodiodes are required at the receiver so as to detect the BPSK points in the bidimensional space (I and Q coordinates) on each polarization. A local oscillator is required for enabling the coherent reception at the desired wavelength. After the photodiodes, the electrically converted signals are sampled and digitalized by an Analog to Digital Converter (ADC).

Page 68 of 99





Digital signal processing (DSP) then allows the reconstruction of the transported information (by eliminating the effects of chromatic dispersion, CD, polarization mode dispersion, PMD, and non-linear effects). A FEC decoder is used for correcting the incurred errors. Finally, the signal is deframed and sent to the respective client cards.

In Idealist we have assumed a reach of 2500 km for a 40 Gb/s signal. This is possible by considering a simple hard-FEC, whose power consumption values are assumed in Table13 (left side).

Laser

I1

I2

Clie

nt

FEC/

Fram

er

RF Amp

RF Amp

20Gbit/ s

20Gbit/ s

Local oscillator

Clie

nt

Dec

isio

n

Car

rier F

requ

ency

es

timat

ion

Car

rier P

hase

est

imat

ion

Pola

rizat

ion

dem

ultip

lexi

ngEq

ualia

zatio

n

Chr

omat

ic d

ispe

rsio

n co

mpe

nsat

ion

AD

C S

ampl

ing

Pola

rizat

ion

dive

rse

90

°

hybr

id

Def

ram

er/F

EC

(a) Emitter

(b) Receiver

T.E.

T.M. Laserp/2

I1

Q1

I2

Q2Cl

ient

T.E.

T.M.

FEC/

Fram

er

RF Amp

RF AmpRF Amp

RF Amp

28Gbit/s

28Gbit/s

28Gbit/s

28Gbit/s p /2

Local oscillator

Clie

nt

Decis

ion

Carri

er Fr

eque

ncy

estim

atio

n

Carri

er Ph

ase e

stim

atio

n

Polar

izatio

n de

mult

iplex

ing

Equa

liaza

tion

Chro

mat

ic di

sper

sion

com

pens

atio

nAD

C Sam

pling

Polar

izatio

ndi

verse

90hy

brid

Defra

mer

/FEC

(a) Emitter

(b) Receiver

PBC

Figure 35: Architecture of a 40 Gb/s DP-BPSK transponder.

Figure 36: Architecture of a 100 Gb/s DP-QPSK transponder.

100 Gb/s transpondersThe architecture of a 100 Gb/s transponder is very similar to that of a 40 Gb/s one. In this case, the number of client cards N = 10 since we assume we have 10 Gb/s line cards. On the emitter side, the framer now has to provide four electrical signals at R = 25 Gb/s (each symbol is made by two bits per polarization), which will drive the modulators creating a QPSK optical signal per polarization. For detecting the bidimensional constellation on the two polarizations, four photodiodes are required on the receiver side (I and Q coordinates). The same as for the 40 Gb/s transponder, a local oscillator, ADC and DSP, and FEC are present in the receiver. The deframer sends data onto the 10 respective line cards. Figure36 shows a schematic of the 100 Gb/s DP-QPSK transponder.

In Idealist we have assumed a 2000 km reach for a 100 Gb/s signal. If the same FEC is used for the 40 and 100 Gb/s transponders, the reach ratio between 40 and 100 Gb/s is 2.5 (linear factor) [24]. The improvement of the 100 Gb/s reach is made possible by introducing a soft decision FEC, allowing us to cover distances up to 2000 km [25]. The soft decision FEC is more energy greedy than soft FEC, as shown Table 13, where the power consumption for the 100 Gb/s transponder is estimated.

The power difference of the Framer/Deframer for the 40 and 100 Gb/s transponders is due to their different size: more capacity is supported, the device has to guarantee more operations resulting in a higher power consumption. The power consumption of various devices is obtained by considering the scheme presented in [26].

Page 69 of 99





Table 13: Consumption values of 40 Gb/s and 100 Gb/s transponders based on coherent technologies and able to reach distances aimed by the IDEALIST project

Component

40 Gb/s transponder 100 Gb/s transponder

UnitPower

consumption (W)

UnitPower

consumption (W)

Client sideClient card (@10 Gb/s) 4 3.5 10 3.5


FEC FEC 1 4 1 27

E/O modulation

Drivers 2 2 4 2

Laser 1 6.6 1 6.6

O/E receiver

Local oscillator 1 6.6 1 6.6

Photodiode +TIA 4 0.4 4 0.4

ADC 4 2 4 2

DSP 1 60 1 60


Total power (W) 173.8 243.4

It is possible to provide fewer power greedy transponders if:

a) It is possible to use only one laser instead of having two different lasers for the emitter and the local oscillator. By doing this, we are obliged to have the same wavelength at the emission and the reception side. This means that if the transponder is used for regeneration purposes, signal re-colouring will not be possible. This reduction of functionality makes 40 Gb/s and 100 Gb/s transponders 3.7% and 2.7%, respectively, less power greedy.

b) No soft FEC is considered. In this case the FEC part of the 100 Gb/s transponders will only consume 7 W [26] but the optical reach will decrease from 2000 km to 1000 km.

c) It is possible to reduce the amount of operations ensured by ADC, DSP and FEC. With the model presented in Table 13 the ADC+DSP+FEC blocks consume together 95 W; simpler blocks at 30 W can be provided (reducing the chromatic dispersion and polarization mode polarization compensation blocks and using just hard-FEC), but the maximum reach ensured then will only be up to 500 km.

Considering the points b) and c) into account, we have deduced intermediary power consumptions associated with intermediary reach values as shown in Figure 37.

Page 70 of 99





175

200

225

250

0 500 1000 1500 2000 2500Pow

er co

nsum

ption

(W)

Optical reach (km)

Power consumption as a function of the target reach

Figure 37: Power consumption values for short reach 100 Gb/s transponders.

Higher modulation formatsTo cope with the increase of data-rates, transponders using more complex modulation formats are required. For such formats, symbols are phase and amplitude modulated; this is possible thanks to the use of digital-to-analogue converters (DACs) placed before the modulators. In this manner it is possible to switch from QPSK (also named 4QAM) to 16, 32 and 64QAM. The choice of the symbol rate and of the modulation format will determine the physical performance of the optical signal and its total capacity. In the Idealist project we have supposed that 400 Gb/s optical signals have the following features: 75 GHz of spectral occupancy and can cover around 500 km without regeneration. To satisfy such requirements, a possible choice coping with such features will use two subcarriers, with each signal transporting 200 Gb/s with a 16QAM modulation format. The introduction of soft-FEC is mandatory for high-rate transponders, requiring a signal overhead of around 27%. Hence the signal baud-rate is now 32 GBd and 37.5 GHz of channel spacing is the minimum channel slot associated to this channel. With such hypotheses and considering as reference the 100 Gb/s DP-QPSK signal, the transmission reach reduces by a linear factor of 5, hence 400 Gb/s will cover 400 km without regeneration [24]. This reach is slightly shorter than the one aimed at in the project, but it is also optimistic because it does not take into account the increased filtering penalties due to the reduction of the channel bandwidth (from 50 to 37.5 GHz) while using the same baud-rate.

With respect to the 1 Tb/s transponders, in Idealist the target channel occupancy is 175 GHz and the optical reach has to be of the same order as for 400 Gb/s. To increase the channel data-rate two options are possible: the first one relies on the use of more complex modulation formats whilst keeping a constant symbol-rate; conversely the second option consists in increasing the symbol-rate whilst keeping the same modulation format. With the first option, since the optical reach is highly dependent on the modulation format, the reach of 1 Tb/s transmissions will be highly impacted and the 400 km reach will not be achieved. Conversely, by increasing the symbol-rate the reach is not impacted in the same manner if the filtering functions are larger than the signal width (a demonstration of the reach

Page 71 of 99





sensitivity with the modulation format and symbol-rate has been shown in [28] for signals up to 100 Gb/s).

Keeping this in mind, the transponder architecture realizing 1 Tb/s transmission is based on four subcarriers working at 250 Gb/s, obtained with the use of a 16QAM modulated signal with a baud-rate increased to 40 GBd and the channel spacing increased to 43GHz; the larger spectral slot will ensure the same filtering penalties as those observed for 400 Gb/s channels.

For spatial capacity (size) issues in high capacity transponders (due to the integration of a card with multiple carriers), some devices have been integrated together, as is the case for the FEC, ADC, DSP and DAC. The device obtained by their integration is called ‘ADD’ in Table 14. To reduce the number of line cards connected to the transponder, 100 Gb/s client cards have been considered, whose power consumption is estimated be 24 W [27]. To estimate the power consumption of the framer/deframer, we have supposed that its power consumption scales linearly with the amount of processed data; the power consumption values for 400 Gb/s and 1 Tb/s transponders are deduced from the ones already available for the 40 Gb/s and 100 Gb/s transponders. We assumed that the ADD power consumption will depend on the amount of data transported by each subcarrier, which is assumed to scale with the amount of data in the same manner as that of the framer.

In Table 14, the number of devices required for each transponder is indicated by a multiplication product: the first factor denotes the number of subcarriers used to carry the required data-rate, while the second factor refers to the number of devices necessary per subcarrier. With all these assumptions in mind, we have estimated the power consumption of high rate transponders and depicted them in Table 14.

Table 14: Consumption values of 400 Gb/s and 1000 Gb/s transponders based on coherent technologies and able to reach distances aimed by the Idealist project

Component

400 Gb/s transponder 1 Tb/s transponder

UnitPower

consumption (W)

UnitPower

consumption (W)

Client sideClient card (@10 Gb/s) 4 24 10 24


E/O modulation

Drivers 2x4 2 4x4 2

Laser 2x1 6.6 4x1 6.6

O/E receiver

Local oscillator 2x1 6.6 4x1 6.6

Photodiode +TIA 2x4 0.4 4x4 0.4

ADD 2 80 4 90


Total power (W) 481.9 1061.4

Page 72 of 99





In Figure 38 we have depicted the power consumption and efficiency associated with the different data-rates (the values of 200 Gb/s transponders have been extracted from Table14 when only one subcarrier has been considered). From Figure 38 we notice that the energy efficiency (energy per bit) weakly improves with the increase of the rate for bitrates higher than 200 Gb/s. This is due to the fact that higher capacity transponders are obtained by increasing the number of subcarriers and the energy improvement per subcarrier is only due to the integration of some component and little functionality.

0

1

2

3

4

5

0

200

400

600

800

1000

1200

0 200 400 600 800 1000 1200

W/(Gb/s)

Tran

spon

der p

ower

(W)

Transponder capacity

Power consumption as a function of transported capacity

Figure 38: Power consumption and power efficiency relative to the different datarate transponders.

Table 15 summarizes the power consumption relative to the muxponder and uplink (no client side consumption is considered) for the different data-rates.

Table 15: Power consumption relative to transponders realized by a muxponder or an uplink architecture.

40Gb/s 100Gb/s 400Gb/s 1Tb/s

Muxponder 173.5 243.4 481.9 1061.4

Uplink 119.8 149.4 285.9 621.4

5.6.2 Energy model for elastic transpondersThe increase of the traffic transported in the optical network will produce an increase of the energy consumption; an increase that will not be sustainable either in terms of costs or in terms of energy availability. In contrast to traffic that often exhibits fluctuations, optical networks have tended to be quasi-static, as all network elements are fully powered for the peak traffic (including over-provisioning), without considering the actual transported capacity. It appears evident that the first step for improving the energy efficiency in such a

Page 73 of 99





network is to consider the possibility of adapting the number of, or the power consumption of, optoelectronic devices as a function of the traffic to be actually carried. Rate adaptive devices allow a dynamic reconfiguration of the modulation format and/or symbol rate of the optical signal, and this property can be used for enhancing the energy efficiency of future optical networks.

It has been demonstrated in [28] that considering format-flexible transponders have the same power consumption independent of the chosen modulation format. The main advantage of format-flexible transponders relies on the large scale of transmission distances, hence on the possibility of skipping intermediary regenerators. In contrast, because in Idealist we consider only uncompensated transmission links, the symbol-rate adaptation does not offer many opportunities to trade data-rate for reach [28]. However, symbol-rate adaptation offers much better energy saving opportunities than format-adaptive transponders, because it has been demonstrated that the power consumption of optoelectronic devices strongly depends on the symbol-rate [29].

In the next sections we present the energy model for symbol-rate adaptive transponders. The energy models presented have been aligned with the modulation format schemes of the 40, 100, 400 and 1000 Gb/s transponders presented in Table 13 and Table 14. Lower symbol rates are obtained by scaling down the clock reference of the electronic devices (which are required to have tunable clock references). The proposed model can be applied for any data-rate. In order to obtain any payload once the modulation format is fixed, we assume that suppose the symbol-rate of each subcarrier is scaled down so as to obtain the desired transponder capacity.

Power scaling as a function of the symbol-rateIn a transponder, not all devices exhibit a power consumption that depends on the rate of transported data; namely only the ones providing electronic processing do so. Moreover, their power consumption is composed of two parts: one that is static (Pstatic) and another that is dynamic (Pdynamic).

In many systems, the amount of static power has been shown be between 30-50% of the total power of the device. This static power allows the electronic maintenance of logic levels in the device, but also the various leakage currents (more and more important with the decrease of feature sizes and the lowering of the threshold levels). The dynamic power consumption of the device depends on the frequency at which the device operates. This power is directly proportional to the square of the voltage and to the clock frequency.

If only the frequency of the device is tuned, from equation (3) we observe that the dynamic power (and the total power) scales linearly [29]. If it is possible to adapt together with the frequency also the voltage of the device, the relationship between the power and the frequency becomes cubic [30].

In coherent-based transponders, a large part of the power consumption is due to the data processing by ASICs and/or FPGAs, noticeably the framer, FEC codec and DSP blocks. In 40 Gb/s transceivers, such devices consume 73% of the whole power consumption, while for 100 Gb/s and higher capacity transponders this ratio will be about 90%.

Linear power scaling modelAs said before, the dynamic power scales linearly with the clock frequency, as demonstrated in [29], where the authors presented a prototype of a real-time bandwidth-variable coherent muxponder aggregating multiple 10GigE clients onto a symbol-rate-variable DP-QPSK optical signal. Thanks to this technology, a linear power consumption of the device with respect to the actual carried traffic has been demonstrated. In particular, it

Page 74 of 99





has been demonstrated that the power consumption of the DSP unit versus the total bit-rate has the following relationship:

With R being the baud-rate of the device. We assume, in a first approximation, that this linear dependency is the same whatever the electronic device. Hence we rewrite Table 13 and Table 14 and deduce the power dependency of the rate-adaptive transponders when a linear scaling power is assumed, and by considering that the static power is around 50% of the whole transponder power.

Table 16: Linear power dependency for the different high rate transponders (muxponder and uplink) considered by the Idealist project

Linear power consumption as a function of the data-rate (W)

40 Gb/s 100 Gb/s 400 Gb/s 1 Tb/s

Muxponder 102.96+2.53*R 140.16+3.69*R 268.32+6.70*R 565.44+11.20*R

Uplink 70.56+1.37*R 89.16+1.35*R 150+3.00*R 325.44+2.51*R

Cubic power scaling modelFrom equation (3) we notice that a better power reduction can be obtained if, jointly with the frequency, also the voltage is scaled down. At low frequencies, the driving voltage can be lowered; such adaptive schemes, known as Dynamic Frequency and Voltage Scaling (DVFS), yield a steeper dependence on the clock rate, typically proportional to f 3 [30].

We assume that the power savings computed on a single component will be proportional with the savings observed on the whole device. Following this hypothesis, we proceed in computing the savings for the component and then translate it to the entire block.

Different to the frequency variations where no limitation on the minimal frequency values is assumed, this is not the case for the driving voltage. Indeed voltage values are restricted to a range of values ensuring correct device functioning: V [Vmin; Vmax], with Vmin ranging from Vmax/3 and 2/3·Vmax. Indeed too low voltage values mean that correct biasing of the electronic component is no longer ensured.

In the device data-sheets it is possible to find the power consumption of a given component. Such power, called Pdevice, is obtained for a specific clock rate (FclockMax) and driving voltage (VdrMax). The power savings are computed with respect to Pdyn = Xdyn Pdevice.

(3)

With

We assume that for the minimal (maximal) possible frequency will correspond the minimal (resp. maximal) voltage, hence we compute the linear relationship between fclock and V(fclock). Figure 39 represents the linear scaling between the clock frequency and the driving

Page 75 of 99





voltage, while equation (4) expresses such relationship and equation (5) provides the scaling factor.

(4)

(5)

m

Clock frequency (Hz)FclickMIN FclockMAX

VdrMIN

VdrMAX

Drivi

ngvo

ltage

(V)

Figure 39: Linear relationship between the driving voltage and clock frequency.

The total dynamic power consumed by a device as a function of the used clock frequency is obtained by equation (6):

(6)

During the Idealist project we will define the technologies that will be used for the elastic transponders and define the range of voltages that allow for the correct operation of the devices, and define the power consumption as a function of the data-rate.

Page 76 of 99





6 Network planning

6.1 State of the ArtThis section will give a brief overview of several algorithms applicable to network planning in Flexgrid networks. Since those algorithms need to be implemented in planning tools, we review some commercially available tools.

6.1.1 Clustering of nodes for hierarchical traffic groomingTraffic grooming facilitates to minimize the overall cost and justify the (long term) reservation of a lightpath to transport traffic aggregates of smaller granularity packet flows through a wavelength switched core network. Among other techniques, node clustering has been proposed as means for efficient traffic grooming (e.g. [31] - [33] ) since it can provide an intuitive node grouping exploiting topological conditions to achieve reduction of long paths and sharing of resources improving utilization and overall efficiency. Recently [34] a clustering methodology has been implemented to maximize the efficiency of flexgrid technology over core networks aggregating traffic from a large number of aggregation nodes. According to [34] the network is partitioned into metro areas, i.e. locations are grouped into a number of metro areas aiming at the maximization of the aggregation traffic that will be conveyed by the optical core network utilizing flexgrid technology.

In other works (e.g. [35], [36]) partitioning of network nodes into different groups or clusters has been assumed as an inherent feature of multi-domain networks. Thus, in both [35], [36] the network topology is fixed determined by administrative domain criteria and optimization is achieved through appropriate grooming, wavelength allocation and routing of traffic over the existing network infrastructure. In [32], [33] however, the selection of network partitions/clusters is part of the problem in order to improve overall network efficiency. In [33] like in [36] the authors aim to exploit multi-granular optical cross-connects (MGOXC) capable of grooming wavelengths at the waveband level. Clustering in this case addresses the problem of heterogeneous networks, where node clustering is based on node functionality (waveband grooming capabilities or not) and number of hops required to reach an MGOXC node. No optimization in cluster formation is performed in their work. Finally, the work in [32] proposes a method to partition the network into clusters with a single gateway node (named hub), where connection termination and traffic grooming is performed. However, the work in [32] assumes a star topology of clusters, a connection oriented traffic model and traffic aggregation performed at Layer 2.

In [31], CANON has been proposed, an architecture that decomposes a Core/Metro network into a number of clusters which can be viewed as a superposition of ring topology networks. Two node classes are introduced in CANON: the Metro-Core Edge Nodes (MENs) and the Core Transit Node (CTN), the latter serving as a gateway between clusters. All MENs with traffic towards the same destination cluster are sharing the same wavelength(s) in a TDMA fashion. The collision-free launching of slots in the CANON ring topology network is scheduled under the centralized arbitration of the CTN, by means of a MAC protocol. This operation results in statistical multiplexing directly at the optical layer leading to traffic profile smooth-out [37]. In [31], unlike [32], no electronic processing and frame (de)aggregation is performed at CTNs.

Page 77 of 99





6.1.2 Off-line RSA modelsTo properly analyse, design, plan, and operate Flexgrid networks, efficient methods are required for the Routing and Spectrum Allocation (RSA) problem. Specifically, the allocated spectral resources must be, in absence of spectrum converters, the same along the links in the route (the continuity constraint) and contiguous in the spectrum (the contiguity constraint).

Due to the spectrum contiguity constraint, RWA problem formulations developed for Wavelength Division Multiplexing (WDM) networks are not applicable for RSA in Flexgrid optical networks and they need to be adapted to include that constraint. Several works can be found in the literature presenting Integer Linear Programming (ILP) formulations of RSA [38]-[40]. In [38] the authors address the planning problem of a flexgrid optical network, where a traffic matrix with requested bandwidth demands is given. To solve the problem, an ILP formulation that aims at minimizing the spectrum used to serve the traffic matrix is proposed. The RSA formulation cannot be solved in practical times and, therefore, the authors present a decomposition method that breaks the previous formulation into two sub-problems: a demand routing sub-problem and a spectrum allocate sub-problem. Since both sub-problems are solved sequentially, global optimality cannot be guaranteed. In [39] the authors study the RSA problem by providing a different ILP formulation but with the same objective as in [38]. Finally, in [40] the authors formulate the RSA problem and propose an effective heuristic algorithm to obtain near optimal solutions.

In light that the contiguity constraint adds huge complexity to the RSA problem, the concept of channels were introduced in [41] for the representation of contiguous spectral resources. The use of channels allows removing the spectrum contiguity problem from mathematical formulations. Channels can be grouped as a function of the number of slots, e.g., the set of channels C(2) = {{1,1,0,0,0,0,0,0}, {0,1,1,0,0,0,0,0}, {0,0,1,1,0,0,0,0}, … {0,0,0,0,0,0,1,1}} includes every channel using 2 contiguous slots, where each position is 1 if a given channel uses that slot. The size of the complete set of channels C that need to be defined is ∑|C(·)|≤|S|·n, where S represents the set of frequency slots and n is the number of different amounts of contiguous slots that connections can request, e.g. if connections can request for either 1, or 2, or 4, or 16 contiguous slots, n=4.

Using pre-computed set of channels authors in [41] addressed the off-line RSA problem in which enough spectrum needs to be allocated for each demand of a given traffic matrix. To this end, they presented novel ILP formulations of RSA based on the assignment of channels. The evaluation results revealed that the proposed approach allows solving the RSA problem much more efficiently than previously proposed ILP-based methods and it can be applied even for realistic problem instances, on the contrary to previous ILP formulations.

6.1.3 Dynamic RSARSA algorithms are also used both to dynamically provision connections at arrivals of requests (dynamic RSA algorithms). A novel approach for the RSA problem was proposed in [42] facing the routing and the spectrum allocation problems separately. They included the spectrum availability in an adapted Dijkstra’s shortest path routing algorithm. Regarding the spectrum allocation, any heuristics similar to that used for wavelength assignment (first fit, best fit, etc.) can be applied.

The benefits of the above RSA algorithm are: 1) only one graph G (N, E) containing the availability of every frequency slot needs to be stored and maintained, 2) the routing algorithm runs once for each connection request, 3) the complexity of the proposed

Page 78 of 99





extension for spectrum availability to the routing algorithm is negligible, and 4) the spectrum assignment is performed after the shortest route is found, adding flexibility to use any heuristic.

6.1.4 Spectrum ReallocationThe problem of fragmentation has been studied in the literature in the context of WSON and two main strategies have been proposed to reallocate (including rerouting and wavelength reassignment) already established paths: periodic defragmentation and path-triggered defragmentation. The former strategy focuses on minimizing fragmentation itself all over the network at given period of time, whereas the latter focuses on making enough room for a given connection request if it cannot be established with current resources allocation. Periodic defragmentation, requiring long computation times as a result of the amount of data to be processed, is essentially performed during low activity periods, e.g. during nights. Conversely, path-triggered defragmentation, involving only a limited set of already established connections, might provide solutions in shorter times and can be run in real time. It is worth noting that incoming connection requests or tear-downs arriving while network re-optimization or defragmentation operations are running must be kept apart until they terminate.

A novel approach, called SPectrum REaLLOcation (SPRE (LLO)->(SSO)) [42], was developed in the context of STRONGEST; it follows the path-triggered spectrum defragmentation in Flexgrid networks whenever not enough resources have been found for a connection request. Every link in the routes from a set of k-shortest routes connecting source and destination nodes is checked to know whether the amount of available frequency slots is equal to or higher than the required for the incoming connection request. If enough frequency slots are available in one of the shortest routes, the SPRESSO mechanism is triggered to find a set of already established path reallocations so to make enough room for the connection request in the selected route (newP). Otherwise, the connection request is blocked.

It is worth highlighting that a path can be hitlessly reallocated by shifting its Central Frequency (CF) using the push-pull technique described in [43].

6.1.5 Elastic Spectrum Allocation for Variable TrafficThe introduction of flexgrid opens new functionalities to be developed at the optical layer, such as the adaptation of lightpaths through appropriate Spectrum Allocation (SA) schemes in a response to bandwidth variations, in particular, expansion/reduction of the spectrum when the required bit rate of a demand increases/decreases, respectively.

Authors in [44] defined two schemes for variable SA in flexgrid, namely: Semi-elastic SA and Elastic SA, which put some restrictions on the assigned CF and SA. In the semi-elastic SA the assigned CF is fixed but the spectrum width may vary, whereas in the elastic SA both the assigned CF and the allocated spectra are flexibly selected at each time interval. Note that the elastic SA can be implemented by performing sequentially first CF shifting (e.g. using the push-pull technique [43]) and then the semi-elastic SA. They formulated an off-line Multi-Hour Routing and Spectrum Allocation (MH-RSA) optimization problem and showed that the elastic SA scheme with Expansion/Reduction minimizes the amount of un-served bit-rate, since it provides gains double than that of the semi-elastic SA scheme.

Page 79 of 99





6.1.6 Available Planning toolsToday a number of companies offer network planning and operation tools for fixed-grid WDM and IP networks. These tools are being used by network operators and equipment vendors to meet service level agreements, achieve capital expenditure savings, maximize network lifetime, and gain insight into the capabilities of their network.

Some of the most important functionalities these products provide are the following: Optimize routing and equipment placement to efficiently meet traffic demands Produce equipment configuration and equipment requirements Perform capacity planning, determining how to expand the network (e.g., purchase

links) to handle traffic growth Analyse the impact of failures and plan protection strategies to maximize resiliency Analyse and minimize equipment costs Evaluate various what-if networking scenarios, validating network changes or situations

before deploying the production network Perform traffic analysis and engineering Visualize the above information

Some commercial such tools are presented in Figure 40.

Cariden MATE portfolio consists of a tightly integrated set of products (Design, Live, Collector) that support planning, engineering, and operational tasks for IP/MPLS networks. OPNET’s SP Guru Network Planner enables planning and design of multi-technology, multi-vendor IP/MPLS networks.

Opnet’s SP Guru Transport Planner is an advanced network planning solution that enables service providers and network equipment manufacturers to design resilient, cost-effective DWDM, OTN, and SONET/SDH optical networks. It contains advanced network design algorithms that minimize investment costs and optimize operational efficiencies.

IP/MPLSView is WANDL's multi-vendor, multi-protocol, and multi-layer solution for IP and/or MPLS networks, for design & planning, management & monitoring, and service creation & provisioning. NPAT (Network Planning & Analysis Tools) is WANDL's solution for ATM, Frame Relay, TDM, Voice, and Optical Transport networks providing cross-vendor support for all stages of network planning, design, and analysis.

Aria Networks IP/MPLS-TE Operational Planning solution provides planning operation that analyze even the largest IP/MPLS networks.

VPIsystems Multi-layer Transport Optimization solution allows to visualize, understand and optimize transport networks, providing capacity analysis and planning, re-optimization, survivability analysis and greenfield planning.

The Infinera Network Planning System (NPS) provides users with offline graphical modelling, planning, and configuration capabilities for designing optical network solutions. Other companies (e.g., like Huawei, BTI, Alcatel-Lucent Bell, Nokia Siemens, Cisco) also offer similar solutions which are however more integrated with their products.

Nokia Siemens Networks provides network providers with SURPASS TransNet and SURPASS TransConnect, in order to building efficient transport systems.

Page 80 of 99





The Cisco Transport Planner is a fully comprehensive DWDM network design and design management tool. Cisco Transport Planner uses the latest in optical transport technologies from the Cisco Optical portfolio.

These tools offer functionality for the IP and optical WDM layers, most of the times separately. This is because changes in the two network layers occur at different time scales, as the WDM layer is very static and does not really need to be operated at real time.

The introduction of flexgrid technologies will force tools to consider the specifications of this new technology. However, the adaptability of this new architectural paradigm brings the optical layer closer to the IP layer. The IP layer can request and control the bandwidth that the BV- transponders use, meaning that operators no longer need to massively over-provision to accommodate possible fluctuations in the IP layer. This implies that future tools will have to more closely integrate these two layers in order to achieve more efficient resource utilization, including resources used for protection/restoration purposes, and save in capital and operational expenditures.

Note that the tools in Figure 40 are standalone applications and there is no provisioning for them running on the Cloud.

Tools IP/MPLS planning

DWDM planning

Flex-grid planning

Multi-vendor

Cisco – Cariden www.cariden.com/products/mate-product-family/

✓ ✓

Opnet - NetOne Network Planner www.opnet.com/solutions/network_management/spguru_network_planner/

✓ ✓

Opnet - NetOne Transport Planner www.opnet.com/solutions/network_management/transport_planner.html

✓ ✓

Wandl - IP/MPLS www.wandl.com/html/mplsview/ipmplsview.php

✓ ✓

Wandl NPAT www.wandl.com/html/npat/npat.php

✓ ✓

Aria - IP/MPLS-TE Operational Planning www.aria-networks.com/solutions/ip-mpls-te-operational-planning

✓ ✓

VPI - Multi-layer Transport Optimization www.vpisystems.com/solutions/multi-layer-transport-planning---optimization/

✓ ✓

Infinera - Network Planning System www.infinera.com/products/mgmtsuite.html

✓ ✓

Nokia Siemens Networks SURPASS www.nokiasiemensnetworks.com/portfolio/products/transport-solutions/network-management-planning

✓

Cisco - Transport Planner www.cisco.com/en/US/prod/collateral/optical/ps5726/ps11348/data_sheet_c78-658849.html

✓

Even

tual

ly to

ols

will

incl

ude

flex-

grid

ne

twor

ks a

nd in

tegr

ate

the

IP a

nd th

e op

tical

la

yers

Figure 40: Network planning tools overview.

6.2 Algorithms for network planning

6.2.1 Algorithms for off-line network planningNational IP/MPLS networks have been designed using a multilayer approach to take advantage of the optical longer reach. In that approach, the IP/MPLS layer performs routing and flow aggregation whereas the optical layer, based on the wavelength division

Page 81 of 99





multiplexing technology, transports those aggregated flows into optical connections. However, the Flexgrid technology featuring a finer granularity, allows performing grooming also at the optical layer and hence, the aggregation level of the incoming flows can be reduced.

Taking advantage of the above fact, in [45] we propose a new network architecture consisting of a number of IP/MPLS areas performing routing and aggregating flows to the desired level, and a flexgrid-based core network connecting the areas among them.

The results presented next were obtained from solving a close-to-real problem instance consisting of 1113 locations, based on the BT network. Those locations (323) with a connectivity degree of 4 or above were selected as potential core locations. A 3.22 Pb/s traffic matrix was obtained by considering the number of residential and business premises in the proximity of each location. Locations could only be parented to a potential area if they were within a 100 km radius. Finally, four slot widths (50, 25, 12.5 and 6.25 GHz) were considered.

We used the cost model produced in the STRONGEST project [17]: Ten IP/MPLS router types were considered, their capacity ranging from 4 to 57.6 Tb/s and the number of slots where cards are plugged in ranging from 10 to 144. Different cards for access (48x1, 14x10, 3x40 and 1x100 Gb/s), internal (14x10, 3x40 and 1x100 Gb/s), and 400 Gb/s MF-TP ports were considered. Two types of WSSs (1x9 and 1x20) were used to build BV-OXCs. Finally, four types of 3Rs (up to 10, 40, 100, 400 Gb/s) were taken into consideration.

Figure 41 presents CAPEX results as a function of the number of opened IP/MPLS areas. Figure 41a reports aggregated costs of the IP/MPLS areas, including IP/MPLS routers, cards, and ports (excluding MF-TPs which have been considered part of core networks) costs. Briefly, costs decrease as much as 23.5% with the number of opened IP/MPLS as a consequence of cards and ports costs decrement.

Figure 41b shows disaggregated costs for the flexgrid core network as a function of the slot width considered. We found that MF-TPs and 3Rs costs are almost the same irrespective of the slot width. Notwithstanding, MF-TPs costs remain constant regardless the number of opened IP/MPLS areas whereas 3R costs first increase exponentially with the number of areas up to a point where they decline significantly. The sharp increment is as a result of that in the number of aggregated flows in the core network to connect an increasing number of areas while their on-average capacity is still high enough so to need the highest capacity 3Rs (and shortest reach). Once the on-average capacity of the aggregated flows decreases optical signals reach increases significantly and 3Rs are seldom needed. In contrast, BV-OXC costs depend remarkably from the used slot width; starting from the same values, all costs show an upwards trend being those for the 50GHz slot steeper.

When all costs are aggregated (Figure 41c) we observe 31.3% savings when all areas were opened and the finer slot width was used, comparing to the case where only 50 areas were opened and 50GHz slots were used. Note that the latter represents the case where just super-channels are added to fixed-grid DWDM-based networks. Saving climb to 43.8% compared to the case where all areas are opened and 50GHz slots are used.

Figure 42 gives insight on the solutions for the IP/MPLS areas. Figure 42a illustrates the switching capacity and the number of IP/MPLS routers installed on all the areas of each network as a function of the number of IP/MPLS areas. A significant decrement, being as high as 19.2% in switching capacity and 12.0% in terms of number of routers, is shown. Figure 42b focuses on the number and on-average capacity of internal ports while Figure42c provides disaggregated values for 10, 40, and 100 Gb/s internal ports. A noticeable

Page 82 of 99





reduction in the on-average port capacity with the number of opened areas is shown, reaching 52.3% when all areas were opened. In terms of number of internal ports, the reduction is as high as 23.6% as a result of a reduction in the number of 100 Gb/s ports which are gradually substituted by 10 and 40 Gb/s ports as the number of opened areas increases (Figure 42c).

Figure 41: CAPEX vs. amount of opened IP/MPLS areas. a) Aggregated and breakdown IP/MPLS CAPEX. b) Flexgrid core network CAPEX for each considered slot width. c) Core network costs breakdown.

Figure 42: Details of the solutions against the amount of opened IP/MPLS areas. a) Installed capacity and number of IP/MPLS Routers. b) On-average capacity and aggregated number of area internal ports. c) Number of area internal ports.

However, having a larger number of relatively small metro areas implies a reduction in the traffic aggregation at the IP/MPLS layer, thus resulting in higher variability of the traffic flows data-rate offered to the optical layer along the day. We study the impact of aggregation level and traffic variability on the efficiency of adaptive spectrum allocation in supporting time-varying traffic demands in [46]. In particular, we investigate the relation between aggregation and the performance of the elastic SA policies proposed in [44]. To this end, we evaluate the performance of elastic SA assuming different dimensions of the Flexgrid-based core network, which in turn translates into the level of aggregation of the traffic to be carried by lightpaths in the core network.

6.2.2 Algorithms for network clusteringTaking into account the impact of node clustering on traffic grooming and its potential performance gains in the framework of Idealist we will develop methodologies for the optimal segmentation of a mesh connectivity network into a set of interconnected “closed-loop” sub-networks. In contrast to other approaches, we address a different problem, where network clusters exploit a ring interconnection, simplifying routing and resiliency,

Page 83 of 99





without assuming any restrictions on the installed fibre capacity. Additionally, multi-layer, traffic grooming at sub-wavelength granularities is targeted exploiting traffic multiplexing directly at the optical layer. It is worth noting that this optimization problem differs significantly from other clustering techniques [32], [47] since i) clusters are decomposed to closed-loop topology networks that are length limited due to physical layer constraints, ii) clusters are interconnected through gateway nodes with no further grooming at sub-wavelength level, and iii) a set of optimization parameters are taken into account.

The global optimization problem of partitioning a mesh network into clusters is NP-hard, as shown for the relaxed problem in [32]. Thus, for networks with more than a few nodes, it is important to develop heuristics that return solutions in polynomial time. Such heuristics should compute a number of different solutions (clustered topologies respecting the restrictions discussed above), which should be evaluated against a fitness function in order to obtain the optimal result. It is worth noting that the selection of an optimal solution should ultimately be subject to the minimization of the actual resulting network cost. However, an exact actual cost value cannot be computed and used at this stage, since this cost depends on the final planning and dimensioning of the overall network including the inter-cluster network, which will be determined at a second stage. Additionally, the computation of a single value expressing network cost addressing in a general way all network cases is hard to achieve, since this cost may take different parameters into account under different scenarios. For example link distances as well as node dimensioning may affect the overall fibre deployment cost as well as construction works and operation expenses, which may differ for different operators facing different requirements. Since general analytical models for determining an exact actual cost value do not exist, actual cost could is related to the required overall number of transceivers, which is an objective cost factor demonstrating the resulting network complexity and the amount of required resources to serve the specific traffic demand ([32], [36]).

Adopting the ring topology the RWA and resiliency problems are inherently resolved simplifying the clustering problem. Thus, the methodology could be exploited by problems that address i) transformation of random mesh graphs into closed-loop clusters and ii) implementation of node grouping heuristics to optimize cluster selection not based on topological conditions alone (total distance, mean distance from a central point etc.) but on a set of conditions that optimize traffic aggregation and overall network cost and resource utilization. As a case study we will investigate the migration of mesh IP/WDM networks to CANON. The improved efficiency of CANON, due to its inherent grooming features, compared to other network architectures that employ different switching modes over general mesh network topologies has been shown in past publications [31], [37] and references therein). However, an open issue that remains is how to exploit CANON in an evolutionary manner, migrating from a legacy network infrastructure, which has been designed implementing a partial-mesh topology. In this context the problem that remains to be addressed is to take as input a given network topology and a known distribution of average traffic demand between pairs of source destination nodes (traffic matrix) and design a clustered network, where the existing nodes should operate either as MENs or CTNs implementing the CANON architecture. Intra-cluster interconnection of MENs should re-use existing links with appropriate re-organization when required e.g. new point-to-point links between two consecutive MENs of a cluster may be constructed by concatenating/splicing existing fibres or additional (e.g. dark) fibre links re-using parts of the existing infrastructure. Similarly, CTN interconnection could be achieved implementing a mesh inter-cluster network over the existing infrastructure. Finally, the new links should be dimensioned so that they can transport the aggregated traffic between clusters. The main objective of this process should be to maximize the traffic aggregation gains, i.e. to

Page 84 of 99





minimize the amount of resources required to serve the traffic demand, hence the overall network cost.

6.2.3 Transmission configurations selection and RSA under physical layer impairments

Optical connections in flexgrid networks span over many and long links, physical layer impairments (PLIs), accumulate and affect the quality of transmission (QoT). Accounting for PLIs is a challenge for algorithm designers, especially with respect to their exact modelling and the interdependencies introduced. Flexgrid networks are expected to use coherent detection and DSP, implying that impairments, particularly those related to dispersion, will be substantially reduced or fully compensated. However, the additional degrees of flexibility available in flexgrid networks make the minimization of these effects complicated, since they also depend on the transmission parameters that are tuneable.

To formulate the physical layer effects and the transponders’ tuneability in a flexgrid network we assume that each connection has a specific optical reach, defined as the length it can transmit to with acceptable QoT (e.g., BER). The optical reach depends not only on the connection’s transmission configuration, but also on the presence of adjacent interfering connections, their transmission configurations and guard bands used. The number of combinations of possible configurations can be huge; also, PLIs analytical models may not capture all effects or experimental measurements may be limited for some of the options. So, it seems that the only viable solution is to resort to some sort of simplification that captures PLIs in a coarser but safe manner, reducing the parameters and the solution space without eliminating good solutions.

We propose such a simplification that fits well with the above described requirements. We calculate the transmission reach assuming that a connection suffers worst case interference (four-wave-mixing, cross-phase modulation, cross-talk) by the adjacent connections for a given transmission configuration and guard band distance. To be more specific, assume that a flexgrid transponder of cost c can be tuned to transmit r Gb/s using bandwidth of b spectrum slots and a guard band of g spectrum slots from its adjacent spectrum connections to reach l km distance with acceptable QoT. This defines a physical feasibility function l=fc(r,b,g) that captures PLIs and can be obtained experimentally or using analytical models [48]. Note that defining the rate r and spectrum b incorporates the choice of the modulation format used. Using the functions fc of the available transponders (BVT) we define (reach-rate-spectrum-guard band-cost) transmission tuples, corresponding to feasible configurations of the BVT. The term “feasible” is used to signify that the tuple definition incorporates PLI limitations, while the cost parameter is used when there are BVT of different capabilities and costs. The above definition is very general and can be used to describe any type of flexible or even fixed-grid optical network. Using the above methodology the feasible transmission options of the BVT can be enumerated so as to incorporate physical layer effects. The planning algorithm takes these as input when examining the options for serving the demands.

In the context of Idealist, we have developed ILP and heuristic algorithms [15][16] that can be used for planning both transparent (without regenerators) and translucent (with regenerators) flexgrid networks. The developed IA-RSA (IA stands for Impairment Aware) algorithms take as input the traffic matrix and the feasible transmission options of the BVT transponders and serve the demands for their requested rates by choosing the route, breaking the transmission in multiple connections, placing regenerators if needed, and allocating spectrum to them. The objective is to minimize both the spectrum and the cost of transponders used in a multi-objective optimization formulation.

Page 85 of 99





Given: the network topology represented by a graph G(N, E), and the number S of spectrum

slots that the network supports

the traffic matrix Λ

Specification of BVT transponders: transmission reach as a function of the tuneable parameters (rate, spectrum) and guard band left from adjacent connections.

Output: Routes, spectrum allocation, placement of BVT transponders and regenerators, selection of transmission parameters

Objective: Minimize a weighted combination of the spectrum that is used and the cost of the transpondersThe developed heuristic algorithm of [49][16] has been incorporated in the off-line network planning tool described next.

6.2.4 Specifically-designed recovery for Flexgrid Flexgrid technology allows allocating the spectral bandwidth needed to convey heterogeneous client demand bitrates in a flexible manner so that the optical spectrum can be managed much more efficiently. In addition to provisioning, new strategies for recovery can be devised that take full advantage from the spectrum allocation flexibility. In [51] we propose a new recovery scheme, called single-path provisioning multi-path recovery (SPP-MPR), which provisions single-paths to serve the bitrate requested by client demands and combines protection and restoration schemes to jointly recover, in part or totally, that bitrate in case of failure. We defined the bitrate squeezed recovery optimization (BRASERO) problem to maximize the bitrate which is recovered in case of failure of any single fibre link. Exhaustive numerical experiments carried out over two network topologies and realistic traffic scenarios show the efficiency of the proposed SPP-MPR scheme approach while providing recovery times as short as protection schemes.

6.2.5 Large-scale optimization techniquesFinding optimal routes and spectrum allocation in Flexgrid optical networks, known as the RSA problem, is an important design problem in transport communication networks. The problem is NP-hard and its intractability becomes profound when network instances with several tens of nodes and several hundreds of demands are to be solved to optimum. In order to deal with such instances, large scale optimization methods need to be considered. We have developed a column (more precisely, path) generation-based method for the RSA problem. The method is capable of finding reasonable sets of lightpaths, avoiding large sets of pre-computed paths, and leading to high quality solutions. Numerical results illustrate effectiveness of the proposed method for obtaining solutions for large RSA problem instances.

6.2.6 Algorithms for in-operation network planningWDM transport networks are typically initiated with an offline/planning algorithm assuming an oversubscribed traffic matrix to absorb short term fluctuations (e.g. daily cycles) and avoid frequent network upgrades (long term traffic increases eventually require upgrades, of course). Thus, the network is typically operated in an incremental manner, with new connections added sporadically, when utilization exceeds a certain percentage, and existing connections rarely (if ever) are terminated. Flexgrid networks using adjustable transponders (BVT) require a different approach as their operation will be more dynamic,

Page 86 of 99





having time scales at which optical connection rate changes occur probably 1-2 orders of magnitude smaller than in fixed grid networks. Flexgrid can bring the optical layer closer to the IP layer, making the IP layer able to “dial”/control the bandwidth that it uses.

Dynamic traffic variation in flexgrid networks can be accommodated at two different levels. We consider the first level to be the establishment of new connections, as in fixed grid networks. Given the high capacity that bandwidth variable transponders (BVT) are expected to transmit (designs of 400 Gb/s or higher are considered in Idealist project), relatively long periods of time will pass until a new connection is established, probably longer than in WDM networks. A second level is to absorb changes in the requested rate that are short- or medium-term by adapting the BVT, e.g., tuning the modulation format and/or the number of spectrum slots they use, a feature not available in WDM systems.

In the framework of IDEALIST project we are developing algorithms to establish and adapt connections following the above described framework. In particular we are developing algorithms to establish new connections but also to adapt existing connections to serve dynamic traffic variations in a holistic manner.In the control plane of flexgrid-based networks, the Path Computation Element (PCE) is commonly used to perform such path computations each time a connection request arrives to the network. PCE is evolving from a pure state-less condition to an active stateful architecture [52], [53]. In the latter, Label Switched Path (LSP) state information is stored at the PCE and used to directly trigger network planning operations, e.g., virtual topology reconfiguration or defragmentation. In this section, we briefly present some works presenting algorithms developed in the IDEALIST project to allow performing network planning while the network is in operation.

Flexgrid networks and BVTs enable transmission of different channel’s capacities and different signal’s format. This characteristic allows going a step further introducing the adaptation of the modulation format of the signal to the distance of the path to traverse. In a network scenario where the routes can vary between longer and shorter reach paths, the latter can be transmitted through a more efficient modulation format such as DP-16-QAM.

To this purpose, a novel heuristic algorithm called: KSP Distance Adaptive Multifiber Spectrum Assignment (KSP-DA-MSA) has been implemented to compare the network evolution results with a more efficient use of the spectral resources by adapting the modulation format to the path. Table 17 shows the different modulation formats considered and the related reach.

Table 17 Signal’s formats available by the distance adaptive algorithm

Modulation Format Capacity SE FEC Guard Band Reach (km)OOK 10 1 0.12 7 2200

DP-QPSK 40 4 0.12 7 2800

DP-16-QAM 40 8 0.12 7 800

DP-QPSK 100 4 0.12 7 2800

DP-16-QAM 100 8 0.12 7 800

OFDM-DP-QPSK 400 4 0.12 10 3560

OFDM-DP-16-QAM 400 8 0.12 10 800

OFDM-DP-QPSK 1000 4 0.12 10 3560

OFDM-DP-16-QAM 1000 8 0.12 10 800

Page 87 of 99





Dynamic restoration in multi-layer IP/MPLS-over-Flexgrid (DYNAMO) [54].

Although the flexgrid technology favours more efficient spectrum utilization, multilayer IP/MPLS-over-flexgrid networks would still be needed. To be operated, a centralized PCE could be used. In the event of a failure, tens or hundreds of client flows could become disconnected and thus, restoration routes need to be found by the PCE for these flows. In standard restoration, path computation for each client flow is performed which derives into resource contention as a result of several connections trying to use some common resources. We thus propose to group client flows’ restoration requests into a single bulk in the PCE. Next, a Global Concurrent Optimization (GCO) module focuses on reconfiguring the virtual topology and finds routes for all the flows in the bulk. Exhaustive simulation results performed on two national core topologies show that a PCE with a GCO module solving DYNAMO highly improves restorability and reduces remarkably the number and capacity of transponders, at the expense of some increment in restoration times.

Algorithms for dynamic lightpath adaptation under time-varying traffic.

Work in [46] focuses on lightpath adaptation under time-varying traffic in a dynamic Flexgrid optical network; it explores the elastic spectrum allocation (SA) capability of Flexgrid and, in this context, we study the effectiveness of three alternative SA policies, namely Fixed, Semi-Elastic and Elastic. For each elastic SA policy, we develop a dedicated algorithm which is responsible for adaptation of spectrum allocated to lightpath connections in response to traffic changes. The evaluation is performed for a set of network scenarios with different traffic variability. In our experiments up to 21% more traffic is served with the proposed elastic SA than with the fixed SA.

In [55] we proposed a framework to serve dynamic traffic in a flexgrid network. The offline algorithm used to initialize the network, or the dynamic online algorithm that subsequently adds connections, assigns to each connection a path and a reference frequency. A connection occupies a certain amount of spectrum slots around that reference frequency, and traffic variations can be absorbed by the BVT by tuning the modulation format or expanding/contracting the spectrum they use. Slots that are freed by a connection can be assigned to different connections at different time instants, obtaining statistical multiplexing gains. To enable the dynamic sharing of spectrum, we need Spectrum Expansion/ Contraction (SEC) policies to regulate how this is performed. After establishing and tearing down multiple connections in a flexgrid network the spectrum slowly becomes fragmented, reducing its ability to accommodate new connections.

An algorithm for elastic operations and hitless defragmentation [56].

The objective of this algorithm is to perform elastic operations, i.e., increase/decrease the bitrate of already established lightpaths. An optimization strategy can be triggered, if required, for hitless defragmentation of the optical spectrum targeted to make enough room to perform the elastic operation. Defragmentation is achieved shifting already established lightpaths in the spectrum.

6.3 Architecture of network planning toolIn this section we describe the architecture of the planning tools currently being developed in IDEALIST.

Page 88 of 99





6.3.1 Off-line network planning tool (MANTIS)Mantis is the IDEALIST network planning and operation tool for designing the next generation flexgrid optical networks. It includes novel flexgrid optical network algorithms for planning and operation functions. Mantis architecture permits fast execution of the included mechanisms, efficient usage of the computational resources’ utilized and enables the deployment of the tool both as a desktop application and as a cloud service (SaaS).

In Mantis, users can define various parameters (e.g., network topology, traffic demands, equipment, devices monetary and energy cost) and select among a set of algorithms for routing and wavelength assignment for WDM networks, routing and spectrum allocation for flexgrid networks. Algorithms evaluate future network plans and demands and report detailed solutions including the required bandwidth to serve the demands, the number and configurations for transponders and regenerators, total monetary cost and total required energy, and report on connections that could not be established either due to physical layer impairments or due to bandwidth unavailability.

ArchitectureMantis components are organized in three layers: the access layer, the application layer and the execution layer. Furthermore, there are two common interfaces whose primary purpose is to provide loose coupling between the application layer and the other two layers. In this way, the same access and execution layers can be used whether Mantis is deployed as a desktop application or as a cloud service. Figure 43 shows Mantis architecture main components.

Figure 43: Mantis architecture main components.

The access layer handles the interaction with the users through a web-based interface and exposes a RESTful API. The execution layer consists of the execution engine and the library of available network planning and operation algorithms. The execution engine receives requests, for starting or terminating algorithms’ executions, through the common interface from the application layer and is responsible for performing all the required actions, including the preparation of the execution environment, the monitoring of the execution progress and the handling of the final results or possible failures. In the current

Page 89 of 99





version, the algorithms are written either in Cython or in C++ language and are accessed from the execution engine through a custom plug-in mechanism. This mechanism enables new algorithms to be added in the tool without any modification of the application layer and the execution engine.

The application layer implements the application logic and orchestrates the execution of user requests. It is the only layer that differs between desktop application and cloud service deployment as there are different requirements and operations that should be performed. In the first case, there is a server that contains the desktop application engine and the execution layer implementations. The desktop application’s engine receives requests from the access layer and stores them in a local queue and a disk file that provides a simple fault tolerance mechanism, eliminating the possibility of requests getting lost or not served due to server problems. Also, desktop application’s engine limits the number of concurrent executions based on the capabilities of the hosting machine in order to avoid resources saturation since the algorithms are executed only in the machine where the server is deployed.

Software as a Service (SaaS) OperationWhen MANTIS is deployed as cloud service, the application layer implements the cloud application’s engine that handles the interaction with the cloud infrastructure. In this deployment, there is one execution engine in every virtual computing node, which runs a particular algorithmic instance. Cloud engine has been designed to be modular in order to support multiple cloud service providers with minimum effort and changes. In the current version, Mantis supports Amazon Web Services and ~okeanos, GRNET’s cloud service for the Greek Academic Community.

Cloud engine consists of the following components: request and response queues, information provider and dispatcher. Cloud engine receives requests from the access layer and stores them in the request queue and a disk file. The dispatcher reads the request queue and checks if the user’s request relate to a new execution or the termination of an old one, still in progress. In both cases, the dispatcher uses the available information from the information provider in order to handle the requested operation. The information provider keeps useful details about the available cloud resources, their capabilities, their current load and the tasks that are assigned to each one for execution. Using this information the dispatcher can decide the execution node each request should be forwarded to, while it can automatically adjust the available cloud resources so that they are better aligned with the total demand. Finally, the response queue is used by the execution nodes in order to inform the cloud engine about the usage of their resources and the status of the executed jobs.

User InterfaceMantis comes with a simple web-based user interface through which users access all its functionalities: network topology and traffic demands creation, algorithms selection and configuration, execution and results presentation. Mantis user interface enables users to easily and graphically design network topologies, store, edit, and use them later. Similarly, users can define their own traffic matrices, either graphically or by importing them from comma-separated values (CSV) files. Figure 44a and Figure 44b show the available interface for network topologies and traffic demands respectively. New configurations can be created for each algorithm by selecting a network topology, traffic demands and specifying all the other required parameters (Figure 44c).

Page 90 of 99





(a) (b)

(c) (d)

Figure 44: Creation of (a) network topology, (b) traffic matrix, (c) configuration, and (d) display of information for different instances.

Users can always check the status of the running instances and have access to useful details (Figure 44d). Furthermore, charts can be created by combining the results of completed instances. More details on Mantis can be found in

Using Mantis it is possible to create a common benchmarking environment with social characteristics where researchers share topologies, traffic matrices and CAPEX/OPEX parameters, and evaluate their algorithms under common conditions. In this way, Mantis could also evolve as an online collaboration platform for optical network researches, improving the comparability, quality and reliability of the results presented in various research articles and projects.

Algorithms included in MantisThe current Mantis version includes network planning and operation algorithms for fixed-and flexgrid optical networks that can be used for both transparent (without regenerators) and translucent (with regenerators) networks. The IA-RSA (IA stands for Impairment Aware) algorithm [15] considers the planning problem of a flexgrid optical network under PLIs, allocates BVT transponders, selects their transmission configurations and assigns routes and spectrum to the connections. Also implemented are two algorithms for planning mixed and single line rate WDM systems. Dynamic versions of these algorithms are also available that take as input the output of the offline case and serve the new demands one-by-one.

6.3.2 In-Operation network planning tool (PLATON)The requirement of executing network re-optimization operations to efficiently manage and deploy new generation flexgrid-based optical networks has brought to light the need of some specialized PCEs capable of performing such high time-consuming computations. Just to mention, some network re-optimization operations are optical spectrum defragmentation, re-optimization after a failure has been repaired, etc.

Page 91 of 99





The objective of such re-optimizations is to compute network reconfigurations to be done based on the current state of network resources to achieve near-optimal resources utilization. Since these operations need High Performance Computing (HPC) hardware to produce a solution in practical times, specialized PCEs can be deployed in the back-end while having a PCE capable of solving some common tasks, such as plain path computations in the front-end. Front-end and back-end PCEs can communicate by using the PCE Protocol (PCEP) as Inter-PCE communication protocol.

Back-end PCEs require high performance computing equipment to process the huge amount of data in both the Traffic Engineering (TED) and the Label Switched Path (LSP-DB) databases in a unified computation step. To deal with this problem, a HPC Graphics Process Unit (GPU) -based cluster architecture is proposed. This architecture is capable of attending to PCE requests demanding execution of network re-optimization tasks, perform such computations and report a near-optimal solution in practical times.

When a request received at the front-end PCE requires computing a specific optimization algorithm which is not among the local algorithms, front-end PCE looks for into its algorithm lookup table to find a back-end PCE able to run such algorithm ID. An inter-PCE request is then sent via the PCEP protocol to that back-end PCE. Optimization algorithms need not negligible time to produce a solution, typically several minutes or even hours depending on its complexity and the size of the TED and LSP-DB databases. Hence, the front-end PCE cannot be stopped waiting for a computation request to terminate; instead, back-end PCEs send a notification after the computation finishes.

The architecture of the IDEALIST specialized back-end PCE is shown in Figure 45. It consists of a cluster manager module and some HPC agents which run algorithms on highly parallel GPU-based hardware. The cluster manager module contains a PCE server responsible for attending remote PCE requests and storing them into the requests database (RqDB), which stores pending computation requests. Another module, named manager, is in charge of managing computation queues assigning computation requests to HPC agents which in turn perform the requested computation. A request-response UDP-based protocol is used between manager and HPC agents.

Each HPC agent consists of two different parts. Running in the host computer CPU a communications module is in charge of the UDP-based protocol, while a set of optimization algorithms are available. Each optimization algorithm is separated into two blocks, the one running in the host’s CPU and the set of kernels running in the GPU device. The cluster might contain a number of these HPC agents, each in charge of a single GPU device, to create a highly scalable system with hundreds or even thousands of computers, each running several HPC agents.

The execution sequence used in PLATON is illustrated in Figure 46. When the PCE server receives a PCEP request (1) it creates a new entry in the RqDB (2). This database encodes a priority queue of requests used to decide the next request to be computed; the oldest request is not necessarily the first one to be computed. After committing the insertion, RqDB triggers a notification to the manager (3) so that the latter may know that a new request has been added to the queue. When manager receives the notification, it queries RqDB to get the new request (4) synchronizing the copy of the priority queue that it maintains in memory (5).

Page 92 of 99





Figure 45: IDEALIST HPC-based Planning Tool Architecture (PLATON)

When a HPC agent becomes idle after computing some request (6), manager selects a request from the priority queue and assigns that request to the HPC agent, transfers the data and parameters required for the computation and informs the agent about the optimization algorithm ID to be used (7). An HPC agent should inform manager about the evolution of the algorithm it is running (9). These intermediate results are stored in the RqDB and are available to be accessed from outside PLATON (10). When the agent finalizes the computation, it transfers the final results to the manager (11) who stores them into the RqDB (12). Then, the RqDB triggers a notification to PCE server informing that the request has been computed and results are available (13). PCE server queries the RqDB (14) to get the results (15) and sends them using a PCEP response to the front-end PCE that originated the request (16).

Figure 46: Sequence Diagram

The main loop of HPC agents consists in: a) receiving the computation input data, parameters, and optimization algorithm ID (7); b) executing computation kernels, i.e. GPU computation functions, corresponding to that algorithm (8); c) periodically notifying

Page 93 of 99





manager computation statistics (9); and 4) sending the final results to manager upon a request computation is completed (11).

A specific planning tool architecture based on HPC GPU devices, PLATON, capable of solving large optimization problems related on optical network planning and re-optimization has been presented. Although at the time of writing this deliverable PLATON is under development we noticed that especially GPU designed shortest paths algorithms should be investigated so as to take full advantage from GPU massive parallelism. For this very reason, we are also currently focused on devising and implementing efficient parallel SP algorithms that eventually reduce algorithms’ execution time.Table 18 presents the GANTT diagram for the development of PLATON, which includes the current implementation status.

Table 18 PLATON development status

Task %Done May

201

3

Jun

2013

Jul 2

013

Aug

2013

Sep

2013

Oct

201

3

Nov

201

3

Year

#2

Year

#3

PLATON 18% Cluster Manager 60% Requests Database 100% Manager 100% Management Web Server 100% Web-services Server 0% PCE Server 0% HPC Agent 13% Communications 0% Optimization Framework 25% Deliverable D1.2 0% Algorithms year #2 17% Single Layer Flexgrid Network Design Problem 50% After Failure Repair Optimization (AFRO) 0% Spectrum Defragmentation (SPRESSO) 0% Algorithms year #3 0% Define Algorithm Set 0% Implement Algorithms 0%

6.4 Problems to be implemented in PLATONWe have selected three problems that will be implemented in PLATON. They cover the three stages of the networks’ life cycle:

Single Layer Flexgrid Network Design Problem

After Failure Repair Optimization (AFRO)

Spectrum defragmentation (SPRESSO)

6.4.1 Single Layer Flexgrid Network Design ProblemThe objective of this problem is to design a single-layer Flexgrid network to serve a given set of demands. The problem statement is as follows.

Given: a network topology represented by a graph G(N, E), being N the set of core locations

and E the set of fiber links connecting two locations,

a set S of available slots of a given spectral width for each link e ∈ E,

a set D of demands to be transported,

Page 94 of 99





BV-OXC cost, which includes a fixed cost for common hardware and a variable cost which depends on the nodal degree and the number of local ports,

an installation cost for each fibre link actually installed and a cost for every optical amplifier (OA) to be equipped in the used fibre links.

Output: The optical network, including BV-OXCs and its configuration, OAs and fibres.

Objective: Minimize the expected CAPEX for the core network designed for the given set of demands.

6.4.2 After Failure Repair Optimization (AFRO)The AFRO problem is a re-optimization problem for dynamic traffic scenarios, which is triggered after a fibre link that had failed has now been repaired and is ready to be used again.

In the event of a link failure, we assume that some recovery mechanism is activated to recover those optical connections affected by the failure. While the failed link remains unrepaired in the network, such link is not available for incoming connection requests. Once the failed link is repaired and active again, a sub-optimal traffic routing exists in the network due to, at least, two reasons: a) recovered optical connections that used the repaired link might follow long routes after recovery, and b) connections arrived during the failure time-to-repair might follow longer routes due to the temporary unavailability of the failed link.

Therefore, the presence of optical connections whose current routing was affected directly (restored connections) or indirectly (new connections) by the link failure, and the fact that new repaired link is currently unused, justify the application of the AFRO problem as a mechanism to minimize the use of optical resources in the network by rerouting some of the established optical connections from current to shorter routes that use the repaired link.

The AFRO problem can be formally stated as follows:

Given: an optical network, represented by a graph G (N, E), being N the set of core locations

and E the set of fibre links connecting two locations,

The set of working links E’ and the repaired and empty link e*, so that E = E’ U e*

a set S of frequency slots available in each link e ∈ E,

a set P of already established lightpaths, only using links in E’

Output: a set of re-allocated lightpaths P*, so that each lightpath must contain e*

Objective: Minimize the use of optical resources (e.g. the total amount of used slots)

6.4.3 Spectrum defragmentation (SPRESSO)The SPRESSO problem defined in [42] can be formally stated as follows:

Given: an optical network, represented by a graph G (N, E), being N the set of core locations

and E the set of fibre links connecting two locations,

a set S of frequency slots available in each link e ∈ E,

a set P of already established paths,

Page 95 of 99





a new path (newP) to be established in the network. A route for the path has been already selected but there is no feasible spectrum allocation,

the threshold number of paths to be reallocated.

Output: for each path to be reallocated, its new spectrum allocation,

the spectrum allocation for newP.

Objective: Minimize the amount of paths to be reallocated so to fit newP in.

Page 96 of 99





7 Conclusions

The first deliverable has provided all of the information needed to carry out a thorough investigation into the benefits of flexgrid and more generally Elastic Optical Networking. Reference networks, applications or Use Cases, modelling tools and algorithms and techno-economic data are all provided here.

Of course these project inputs will, to some extent, require refinement. Although the reference networks will remain stable now for the remainder of the project, there will be an ongoing adjustment to the Use Cases. It is likely that further Use Cases will be added to the list, as the project digests the results of the simulations and modelling.

Regarding CAPEX, the status is mature, at least regarding existing network technologies. But it will be required to describe and to assign the cost parameters to new devices when they will be more clearly defined (SBVTs and flexgrid node parts, especially the ones on A/D side allowing the connection with SBVTs). In addition it could be useful add to the model an integrated OTN with fixed grid WDM as a competitor to the all-optical flexgrid option.

Regarding OPEX, there is clearly a great deal more that can be done on all of the OPEX related aspects, from automation to energy consumption and space saving. Energy consumption has an impact both from a network architecture perspective but also at the level of individual components: energy savings should come directly from the additional flexibility provided. There has already been strong input on energy savings to be expected from adaptive data-rate transponders, but further work is expected here.

Finally, the scene has been set for a great deal of intensive algorithm-based work to provide the key tools to carry out the comparisons required throughout Idealist. One area of particular interest is the research on large scale optimisation, which stands to shed light on the largest problems (e.g. modelling BT’s huge network of over 1000 nodes, and with many other simultaneous variables to be optimised).

EON and flexgrid offer the potential for significant bandwidth efficiency increases and this will enable operators to squeeze significantly more life out of their network infrastructures – one estimate presented here suggests up to 5 years. Although this is a large benefit, and although all carriers are experiencing continuing large traffic volume increases, the non-elastic fixed grid solutions can meet these capacity needs for several years to come. In the mean-time, there is a lot of discussion about when carriers should make their networks EON-ready and how they should migrate.

Additionally, there is increasing thought being given to other benefits that might emerge, as well as the basic capacity boost. Already in Idealist, we are seeing interest crystallising in the nearer term towards multi-layer protection applications, and in the longer term towards the use of technologies such as the sliceable bit rate variable transponder. The current perception is that elastic and flexible concepts have much to offer, but we need to continue the creative and analysis processes to find the application sweet spots.

For example, with SBVTs, a comprehensive techno-economic study is required, both for the longer term dynamic traffic scenario, but also for a nearer term more pragmatic case in which the traffic grows but in a predictable, relatively static way. Are BVTs useful – and if so, what are the additional advantages of sliceability? The SBVT architecture (including the

Page 97 of 99





structure of the device and the architecture of the application in which it will be used) is still in the process of being properly defined, and part of that exercise involves interworking with WP2.

Overall, EON has strong potential to provide capacity increase, flexibility to future traffic dynamics and a reduction in equipment costs and energy consumption. Which version of EON could achieve these advantages, and which carrier applications are most applicable in the medium and long terms, will require the detailed study open which WP1 is now embarked.

Page 98 of 99





8 Annex 1 - Detailed reference network information

Note – this annex should not be made public

The embedded file contains all the necessary reference network topology and other information needed to do simulations on Idealist networks.

END OF DOCUMENT

Page 99 of 99

Date post:	31-Mar-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times