1

An Operator's view on introduction of White

Boxes in Optical Networks
Emilio Riccardi, Paul Gunning, Óscar González de Dios, Marco Quagliotti,
Víctor López, Member, IEEE and Andrew Lord, Member, IEEE
 manufacturers (ODM) of hardware (HW); and selectively
Abstract—Hardware and software disaggregation is a exploited and adapted free and open source software (SW).
recognized strategy for achieving efficiency and cost reduction The most prominent activity is Facebooks Open Compute
within datacentre warehouse. More recently this approach has Project (OCP) in 2009 [2]. Some hyperscale operators have
been applied to high-bandwidth inter-datacentre connectivity at
adopted the 'bare metal' model - where the ODMs physical
transport layer. Telecom Operators look with great interest at
this approach which promises savings that could make the HW is sourced separately from the SW operating system
difference in years of ever decreasing margins on revenues. This (open or commercial). This presumes some in-house
paper presents and analyses the disaggregation models in the competence, or external expertise, for HW-SW integration and
WDM transport layer to replace the established aggregated ongoing lifecycle support. Other hyperscale operators prefer a
model based on single vendor systems. Three optical 'white box' approach where the ODM pre-installs an operating
disaggregation models are considered implying different levels of
system of choice and provides some level of support.
involvement of the Telecom Operator in WDM system design,
assembly and integration. The impact on network lifecycle of The 'hyperscalers' have now started to extend these models
each model is then analysed with particular reference to the roles to support high-bandwidth interconnectivity between their
of the Operator, the equipment Vendors and the System datacentres including the transmission and switching of optical
Integrator. The issue of organizational changes and heavy frequencies. The most notable initiative is Facebooks Telecom
redefinition of processes is addressed and a comparative techno Infrastructure Project (TIP) that commenced in 2016 [3].
economic analysis is also proposed.
Telecommunications operators (TELCOs) appreciate these
Index Terms— Partially disaggregated network, Optical white trends and now seek to adapt them to fit both their own
boxes, Software Defined Networking, Open Line System, Optical datacentre estate and their decentralised physical
Transport. infrastructure. The latter spans across both fixed/ wireless
access networks through metro aggregation to national and
international core networks. One notable initiative from
I. INTRODUCTION AND CONTEXT AT&T - central office re-architected as a datacenter (CORD) -
envisions an overlay of higher-layer Virtual Network
O ver the last decade western hyperscale operators
(Facebook, Amazon, Microsoft, Google and Apple) have
consolidated their storage, computation and networking assets
Functions (VNFs), as modular microservice applications
within containers hosted atop virtualised compute servers
within warehouse scale datacentres [1] across their global within their central offices/ telephone exchanges [4].
estate to exploit the economies of scale occasioned by the Underpinning this is a white box, packet-switched HW fabric
confluence of several trends most notably: the continued forming part of an access network underlay. Attention is now
influence of Moore's Law to underpin increases in the density turning to understand how white box optical HW might be
of logic cells in semiconductor hardware elements (CPUs, utilised in TELCO’s networks to address their business
memory, switching ASICs, etc.); the disaggregation of the ambitions, mainly the ability to cope with the margin squeeze;
operating system software that controls and manages the the ability to diagnose and pre-empt service disruption; the
storage, compute and networking elements; and the emergence prospect of being able to turn-up tailored services to customers
of open Application Programming Interfaces (OpenAPIs) that in an automated manner. This might also introduce
abstract the complexity of the underlying hardware. opportunities for customers to self-provision their
They have displayed a relentless focus on reducing capital requirements directly via a network service API.
and operation expenditure (CapEx and OpEx) by sourcing But the execution of these ambitions is not straightforward.
directly, and in sizeable volumes, from original design TELCOs have traditionally maintained close relationships
with vertically integrated network equipment vendors who
The research leading to these results has received funding from the provide support during the lifecycle of their product. This
European Commission for the H2020-ICT-2016-2 METRO-HAUL project begins with the initial tendering of equipment to match a
(G.A. 761727).
E. Riccardi and M. Quagliotti are with Telecom Italia, Torino, Italy (e-
TELCOs requirements; the ongoing maintenance and upgrade
mail: [email protected]). support during the operational life of the equipment; and
A. Lord and P. Gunning are with British Telecom, Adastral Park, Ipswich, culminates in eventual decommissioning at the end of life.
United Kingdom (e-mail: [email protected]). This is the traditional vertically-integrated vendor black box
O. González de Dios and V. Lopez are with Telefonica GCTIO, Madrid,
Spain (e-mail: [email protected]). model where both the HW and SW are closely aggregated; the
 2

oversight of HW components is also assumed by the vendor. disaggregated equipment from some WDM equipment
The vertically-integrated vendors actually source most of their vendors (e.g. among others [11][12][13]).
branded HW from ODMs and are now starting to offer This is paving the way for the ability to 'disaggregate'
branded white box HW - often termed 'brite boxes'. Brite individual optical domains, enabling also the use of HW from
boxes, like white boxes, disaggregate the operating system and different vendors interchangeably, opening new opportunities
application software from the hardware. They come pre- for equipment vendors traditionally less strongly focused on
installed with a third-party operating system of the operators transport including new actors: pure white boxes on blades
choice with provision for a tailored level of lifecycle support vendors, control and management SW developers, integrators.
from the vertically integrated vendor. Opportunities exists also for TELCOs in the form of
The four models: bare metal; white box; brite box and black claimed cost reduction, vendor lock-in elimination and
box are schematically depicted in Fig. 1 together with the expected enhanced flexibility and modularity, but it is still to
graded 'spectrum' of lifecycle support that a TELCO must be fully demonstrated and may depend on the specific use case
sustain. The bare metal and white box models requires the considered.
most commitment from the TELCO and pre-suppose some in-
This paper is organized as follows: in section II we
house resource and lifecycle commitment. This may prove
introduce the concept of optical disaggregation from the
challenging to some operators in the near-term as the software
perspective of a TELCO: the focus is on the metro-regional
engineering talent and skills, in particular, are a scarce
network segment. Section III gives a brief overview of the
resource. Notwithstanding, some TELCOs are already
most relevant HW, and SW, specifications, and automation
introducing these new paradigms especially to address L2/L3
tools supporting or needed for the introduction of full or
applications and services, facing the skills issue with
partial disaggregation in the optical domain. Section IV
initiatives to reskill their workforce [5]; others are working
discusses the process of designing, developing and testing of
closely with systems integrators (SI) to the same ends.
WDM transmission systems, a TELCO have to face if
The extension of these lifecycle models to include also the
undertaking a disaggregation approach in the optical network.
WDM transport layer is an exciting new possibility recently
Section V analyses the impact of optical disaggregation in the
gaining momentum and interest among TELCOs [6], with an
processes of network creation and operations. Section VI
open and lively debate between supporters and sceptical [7].
complements the discussion with a techno-economic
In fact traditional system vendors are adding SDN
comparison of full optical disaggregation and the traditional
management solutions to their WDM transport network
black box approach. Section VII concludes the paper.
portfolio, to abstract and expose resources at a North Bound
Interface (NBI) enabling enhanced network programmability
II. DISAGGREGATION IN THE OPTICAL DOMAIN
and flexibility. But, often, these solutions are still dedicated to
mono-vendor optical domains and thus imply a black box The term 'disaggregation' in the context of WDM transport
lifecycle approach: direct access to the control and monitoring network is often used to collectively designate all the
of single Optical Network Elements (O-NE) is precluded and operational models in which TELCOs are actively involved in
is fully mediated by the system vendor SW. These solutions the design, assembly, testing and lifecycle management of the
are certainly suitable for large optical transport networks due WDM transport Systems (WDM-Sys) deployed in their
to the complexity of managing physical layer impairments in a networks (spanning across all the operational models of Fig. 1
vendor agnostic way; but vendor 'lock-in' is still present. bar the Black Box). This involvement is conceivable mainly
On the other hand, especially targeting the metropolitan or for the metro regional networks, as already discussed in the
regional network segments, where distance between nodes is introduction, and only if a mature ecosystem of O-NEs, optical
reduced, and degradation effects due to fibre transmission are subsystem blades, control, management, design and planning
more manageable (the algorithms to recover transmission software exists, together with standards or multi source
impairments are potentially less sophisticated, and can be agreements (MSA) for multi-vendor vertical (between O-NEs
more easily implemented at the SDN control level), a new and management and control SW) and horizontal (among O-
ecosystem of optical white/brite boxes is quickly arising NEs) compatibility.
triggered by initiatives like OpenROADM, OpenConfig, For a metro/regional WDM-Sys, the relevant O-NEs are
OOPT [8][9][10] and the introduction to the market of new pieces of equipment housing homogenous network functions,
possibly made by several shelves or blades, but seen by
management and control systems as a single management
entity through a suitable OpenAPI, often termed South Bound
Interface (SBI).
With reference to Fig. 2, O-NEs can be broadly classified
into the following categories:
• Client to WDM adapter (TP): including the network
functions of Transponders (1-1 mapping of clients to line side
interfaces); Muxponders (N-1 mapping and multiplexing);
Fig. 1. Schematic graphical representation of relative level of responsibility
in network equipment lifecycle of TELCO and equipment vendors for Switchponders (N-M mapping, switching and multiplexing).
different operational models.
 3

Within this paper we will designate this type of O-NE simply

as 'Transponder' (TP).
• M-ROADM: Multi-degree Reconfigurable Optical
Multiplexer, including Add&Drop, switching, amplification
and equalization optical functions. In some implementations
the node is assembled from several separate modular
subsystem blades, one for each line degree or Add&Drop
chain.
• Line Terminal (LT): a single line side optical multiplexer
often fitted with colourless functionalities. Often several LT
blades may by interconnected to form a M-ROADM.
• In Line Amplifiers (ILA): inserted in a long transmission
line between LTs or M-ROADMs to recover optical
attenuation.
A suitable interconnection of these O-NEs with the addition
of a WDM transport controller/management SW makes a
complete WDM-Sys (fig. 2): TPs constitute the 'Digital to
WDM adaption layer' (DtoWDM), being in charge of the
adaption of digital client signals to analogical 'media
channels'; while M-ROADMs, ILAs and LTs constitutes the
actual 'WDM Analog transport layer' (A-WDM).
Disaggregation implies that TELCOs are, directly or
indirectly (through SIs), involved in the process of design,
assembly, integration and testing of a whole WDM-Sys
starting from a Control SW together with already assembled Fig. 3. A) An Open Line System as part of a partial disaggregated WDM
DtoWDM and A-WDM layers; or from all or some O-NE transport system: OLS and controller are from a single vendor (1-2); TPs
may be in pair form the same supplier (3) or mixed (4); the WDM Transport
category as elemental bricks; or even from subsystem blades Controller interfaces directly with TPs (5) and through a NBI (7) to the OLS.
on bare metal HW. Single Wavelength Interface (SWI) need to be standardized (6);
Several degrees of TELCO involvement could therefore be B) Alternative partial disaggregated WDM transport system: OLS and
WDM controller are proprietary from a single vendor (1-2); TPs may be in
envisioned as discussed more deeply in next paragraphs. pair form the same supplier (3) or mixed (4); the proprietary WDM Transport
Controller interfaces directly with TPs with standard SBI (5). Single
A. Fully Aggregated Optical Domains Wavelength Interface (SWI) (6) and SBI (5) need to be standardized.
This is the current evolution of mono-vendor WDM-Sys,
simplified abstract model to be used by higher order
with the introduction of an open NBI for control and
controllers or orchestrators. Only this NBI needs to be
managing the whole network island in a more flexible way
standardized.
(Fig. 2). The optical system lifecycle management is
responsibility of the system vendor in the pure black box B. Partial Disaggregation: Open Line System and Multi-
approach. System vendor provides both a proprietary WDM Vendor Transponders
transport controller and all the O-NEs. SBI to O-NEs may run In this approach (Fig. 3A and B), the disaggregation applies
proprietary protocols with tailored equalization algorithms. to the DtoWDM layer (i.e. to TPs) whose lifecycle is
Typically these networks support digital transport services decoupled from that of a mono-vendor and proprietary A-
(e.g. Ethernet or OTN from client side ports of TPs): mapping WDM layer. The A-WDM layer remains a proprietary black
of digital clients and activation of network media channels is
under the control of the proprietary domain. NBI translates
sufficient information of the underlaying optical network in a

Fig. 4. Fully disaggregated WDM transport system: O-NEs can be from the
same (1-2) or from different suppliers. No separation between DtoWDM and
A-WDM layers exist. A standard SBI (5) is needed to simplify the direct
Fig. 2. Fully aggregated WDM transport System: Controller and O-NEs are control of the whole WDM-Sys by the controller (4). Both Single
form the same vendor. A black box lifecycle model is implied. Only the NBI Wavelength (6) and Multi Wavelength Interfaces (7) need standardization.
need to be specified/standardized. The inset classifies O-NEs (see text).
 4

box analogue transport system (boxes 1-2 in Fig. 3) supporting SW). The small involvement of the TELCO in the lifecycle of
Optical Channels from external TPs as client signals. Thus, to the OLS itself is conceivable, for example in the form of a
this Open Line System (OLS) applies all the considerations process of joint customization with the vendor of some
made in the previous paragraph; the term 'Open' refers to the specific features of the system (a weak form of brite box).
fact that it is open to be used by any signal which follows a
C. Full Disaggregation: Multi-Vendor Optical Network
given behaviour, specified by the Single Wavelength
elements
Interfaces (SWI). An OLS-NBI API (6 in Fig. 3A) is needed
to configure and report events from the OLS. The In this approach (Fig. 4) the involvement of the TELCO in
standardization of this OLS-NBI is of great help in the process the WDM-Sys lifecycle is strong, certainly not limited to
of vertical integration with the Open WDM Transport vertical integration of control and management SW. Actually
Controller of the whole WDM-Sys. O-NEs from both the A-WDM and DtoWDM layers are
The rationale behind this approach is that the operational potentially purchased from different vendors, leaving
life of an A-WDM is much longer than that of TSs, the latters interworking at the control and data plane to the system
useful life goverened by the continuous increase in capacity integrator. Therefore most of the control intelligence is moved
needed, requiring a very strong pace of innovation and to the WDM controller (necessarily vendor agnostic) which
therefore obsolescence. Furthermore, leaving the analogue becomes the most critical element of the whole chain, having
domain (including M-ROADMs, LTs and ILAs) under the to face also all the analogue transmission issues (equalization,
responsibility of a single vendor means leaving the transient suppression, etc.).
development, testing and management of complex control, Furthermore detailed specification for both SWI and Multi
equalization loops and analogue heurestics solely the Wavelength Interface (MWI) is needed to support horizontal
responsibility of the vendor. Also horizontal interoperability integration; likewise a standardization of the SBI is
issues among analog O-NEs, again implying analog optical paramount.
design (including linear and non-linear transmission Planning and design of such a multi-vendor network
impairment control), are left to the system vendor requires specialized technical skills that often only vendors
responsibility. have. Alternatively vendor agnostic automatic planning and
Furthermore the multi-vendor environment in the DtoWDM design tools could be employed if they were available to the
layer leaves to TELCO the freedom to choose the best supplier market (an initiative in TIP is ongoing [14]), and sufficiently
for each specific application favouring form time to time comprehensive to be used by a skilled user and not necessarily
performance, cost or other metrics. an optical design expert. A TELCO could profitably utilise
Transponders or pluggable modules on L2 switches are them to automatically provide bill of material, equipment
more easily integrated and controlled directly by a WDM configuration and interconnection schemes for a multi-vendor
transport controller, even without a SBI standard (dedicated disaggregated environment.
drivers/adapters for SDN controller, may be developed by the D. Full disaggregation : Multi-vendor Optical subsystems
transponder vendor itself). This extreme case addresses the situation, albeit
However with a standard SBI it is possible for the OLS hypothetical, of an ecosystem of exclusively optical, low level
Controller itself to take charge of TPs, thus assuming the role functional subsystems (EDFA, WSS, Attenuator, etc…), on
of the controller of the whole system, and eliminating the need separate compatible standardized blades (with control and
for a OLS-NBI and strongly simplifying the integration management interfaces, power supply, cooling etc.) for rack
process and network operations (Fig. 3 B). mounting (Fig. 5); O-NEs like M-ROADM are an assembly of
A standard for interoperability among different vendor TPs several of these subsystem blades, possibly from different HW
is not strictly needed (TPs might be matched pairs from the suppliers. Potentially each subsystem could be directly
same vendor), but is welcomed to relax, as much as possible
constrains in the purchasing process and simplify the process
of horizontal integration. However TP interoperability, may
limit the transmission performance and hamper the
introduction of innovations in the TP. On the contrary a
standard at the Single Wavelength Interfaces (SWI) is
mandatory and suitable monitoring and equalization functions
should be added at the interconnection points to clearly
separate the DtoWDM and A-WDM domains, ensuring stable
operation and easy troubleshooting.
Concerning the DtoWDM layer, all lifecycle disaggregation
models may be applied including the brite box one, for
example in the integration of matched pairs of TPs with
proprietary features (TP supplier involvement in the Fig. 5. Full disaggregation: Multi-vendor optical subsystems assembled in O-
integration process may include the development of the NEs. (1) shows a M-ROADM assembled from three subsystem blades from
different vendors; each blade has its own SBI (2) implying multiple parallel
needed SDN driver/adapters and some dedicated application control communication sections (3) with the network controller (4).
 5

controlled and managed by the WDM transport controller of interest and direct involvement in initiatives supporting a
the whole WDM-Sys, but this would be at the expense of disaggregated optical ecosystem and the introduction by them
more complexity in the controller, due to the lower level of of disaggregated pieces of equipment.
abstraction, and a greater number of concurrent Disaggregation is contingent on the availability of open
communication sessions. APIs that are, in turn, dependent on the development of open
If a single control and management SBI to the network YANG service and data models [15] for managing optical and
controller is desired at the O-NEs level (e.g. only one for each packet network services. There are two levels of management
M-ROADM), several subsystem blades should be 'virtually granularity or scope to consider: a) coarse-grained Network-
assembled' and connected to a local NE controller (e.g. a wide services, where the O-NEs are considered as nodes that
separate blade) which will implement an adaptation and are interconnected by links; and b) fine-grained O-NE devices
control software for the whole 'virtual' O-NE. The result is a proper. A Layer 0 service path is a directed graph formed from
simplification of the network controller with the introduction a subset of nodes and links subject to network resource
and a hierarchy of controllers and more complexity and SW in constraints and service demands. An inventory of discovered
the O-NE. nodes and links; and the associated resources and constraints
A more realistic scenario is a hybrid approach, with multi- is stored in a traffic engineering database (TED). The
vendor optical subsystem disaggregation limited to M- analogue nature of impairments particular to optical
ROADM (Fig. 5); other O-NEs are still multi-vendor, but not propagation with path length i.e. attenuation, dispersion, non-
disaggregated at the subsystem level. This is motivated by the linearities, amplifier and receiver noise are additional
appearance on the market of the so called ROADMs on a constraints unique to Layer 0 because all-optical 3R
blade (line degrees and Add&Drop complete subsystems on regeneration is lacking. In the case of multi-hop or mesh
separate blades) (e.g. [11][12][13]). It seems a reasonable optical networks the lack of wavelength conversion adds an
compromise enabling a multi-vendor approach inside a M- additional constraint to maintain non-blocking end-to-end
ROADM, but standardization of several low level physical, wavelength continuity. Optimisation is performed by a
optical and management/control interfaces remains an issue to dedicated path computation element (PCE).
be solved; furthermore the implications that the bare metal or Open specifications and abstractions for the management
white box approach is critical for a TELCO, CapEx and OpEx and control layers of end-to-end network services and
advantages in particular, compared to other form of individual O-NE devices via the NBI and SBI, respectively of
disaggregation is not proven and currently difficult to the WDM Network Controller are progressing. At the heart of
quantify. these endeavours are the definition and development of
service data models and device data models. Preferably these
III. TOWARD AN OPEN DISAGGREGATED OPTICAL are open, vendor-independent, and specified in a modelling
ECOSYSTEM language such as YANG [15]. For SBI application, in the
Three of the four disaggregation alternatives outlined in context of optical systems, the protocol of election is
Section II are disruptive to the status quo of the mono-vendors NETCONF [16] A very useful recent overview and
that support Fully Aggregated Optical Domain. The vertically comparison of optical network YANG service and data
integrated 'Tier 1' optical mono-vendors distinguish models is provided in [17] to which we refer for details on the
themselves by apportioning appreciable investment and subject.
resource to the in-house design and development of In parallel to the standardization bodies several 'open'
proprietary hardware components i.e. ASICs, photonic initiatives are gaining importance in specifying all the relevant
integrated circuits etc. that are at the cutting-edge of aspects of an open disaggregated optical ecosystem. In the
performance. Of course, when appropriate, this may be following the most relevant ones are briefly introduced.
complemented by commodity, off-the-shelf components too. OpenConfig [9] is focused on open data models and is
Access to the API of the proprietary hardware components is driven by hyperscale operators and TELCOs. It is less of a
tightly controlled by the 'Tier 1' mono-vendor via a closed standard and more of an informal grouping of actors with
element management systems (EMS) and network likeminded interests. Code is posted directly on github
management system (NMS). Considerable resource and effort including a base YANG device data model. It encourages
is dedicated to development, integration, procurement, vendors to provide open vendor-specific YANG model
standardization, customer engagement and lifecycle support of extensions that can augment the base model. Although
their integrated product portfolio. OpenConfig is indifferent to the RPC API, Google’s gRPC is
In contrast, vertically integrated 'Tier 2' optical mono- well-represented. The YANG device data models provide a
vendors have smaller market share and so are more dependent layer of indirection to ‘hide’ low level details of the O-NEs.
on commodity, off-the-shelf hardware components which they From this a variety of REST APIs NBIs can be generated.
control by a similarly closed EMS. This can explain their
 6

Fig. 6. Alternative processes for assembly a WDM transport systems starting from components (Bare metal blades, Subsystems or Network Elements). Each
making process relies on different external inputs (see text), and is in charge of a specific player depending on the disaggregation model adopted: i) Full
subsystem disaggregation: all processes are in charge of TELCO and SI; ii) Partial disaggregation or full multi-vendor O-NE disaggregation. Vendor is
responsible of processes up to subsystem blades making, TELCO and SI are responsible for WDM system integration; iii) Fully aggregated optical domain:
Vendor is responsible of all processes, including WDM system making. Hybrid scenarios are obviously also possible.

The Open ROADM Multi-Source Agreement (MSA) [8] is The Voyager and Cassini essentially reframe the TP as a
created to define interoperability specifications for ‘hardware-modified’ commodity, off-the-shelf merchant
Reconfigurable Optical Add&Drop Multiplexers (ROADM). compute server with a commercial CPU running an open
Founded in 2015 by AT&T, Ciena, Fujitsu and Nokia, linux-based operating systems. This allows a software
currently it counts fifteen members including world leading development ecosystem to support client- and line-side, fixed
vendors and Continental scale TELCOs. The objective of this or pluggable optical hardware modules.
group is the specification of optical interoperability and Open merchant-based LTs, ILAs and 'on a blade' ROADMs
YANG data models, for ROADM switch as well as are also commercially available [11][12][13]. It must be
transponders and pluggable optics. emphasized that these offerings are simple to integrate in
Telecommunication operators allocate appreciable capital- point-to-point, single-hop optical line systems. Consequently
and operational-expenditure (OpEx/CaPex) to TPs in their deployment of the technologies is most likely to first occur for
optical line systems. It is the key O-NE for driving the datacentre interconnect in metropolitan areas. This explains
development of the optical white box ecosystem associated the interest of the hyperscale operators - most notably
with Partial disaggregation: OLSs and Multi-vendor TPs (Fig. Facebook - through their support of the open compute project
3). The availability of commodity, off-the-shelf optical and and TIP activities.
packet switching hardware components outfit TPs with open Also the Open Networking Foundation (ONF) has recently
northbound APIs, modulation formats and forward error commenced the Open Disaggregated Transport Network
correction schemes with interoperability through multisource (ODTN) activity [21] to extend and complement existing work
agreements (MSA) or standards bodies. to disaggregate telephone exchanges/central offices through
The Open Optical Packet Transport (OOPT) project group the CORD activity [4] centered on the vendor agnostic ONOS
[10] of TIP allows vendors to enter the market with controller [22].
competitive TPs for bookending point-to-point dark optical
fibre transmission links of modest span (~100km.) The IV. DESIGN AND DEVELOPMENT PROCESS FOR WDM
Facebook Voyager [18] is a variant of the Facebook Wedge TRANSPORT SYSTEM
100 data centre Ethernet switch [19], the variation being Disaggregation implies some level of TELCO involvement
commodity, merchant silicon modules to drive the fixed in the processes of designing, developing and testing WDM
lineside ports supporting up to 200Gb/s and a switching chip transmission systems. These processes are described
re-purposed in point-to-point, rather than default bridging schematically in Fig. 6 with the aim of emphasizing the role of
mode, to accommodate pluggable transceivers. the TELCO. It identifies: the main inputs and outputs, the
The EdgeCore Cassini is another example which actors involved, and the existing alternatives.
conveniently supports pluggable CFP2 analog coherent optical
and digital coherent optical lineside modules offered by A. WDM Transport System assembly from Optical Network
competing merchant optical component vendors. Just like the Elements
Facebook’s Voyager it has leveraged the development of an In the black box approach the entire lifecycle of the WDM-
existing merchant silicon-outfitted packet switch [20] and it Sys (Fig. 6, box 1) is controlled by the system vendor (box 2)
too, has been submitted to TIP [3]. including SW and HW upgrades and bug fixes. The pace at
which new releases are introduced is driven by the vendor,
 7

with modest TELCO involvement to steer developments to of O-NE white boxes from equipment vendors (box 4).
their needs. Likewise the process of subsystem blades assembly starting
The alternative is the direct assembly of the WDM-Sys by from bare metal blades (box 8) is shown in box 7; the
the TELCO (box 3) acting now as a SI or relying on a trusted alternative process of purchasing fully equipped white box
external SI. In this case the main HW input is a collection of subsystem blades from component vendors is identified by
complete O-NEs possibly from several suppliers. They come box 6. Obviously both approaches are compatible only with a
as standalone devices completely equipped and with an open full disaggregation approach for the optical domain.
SBI (HW white boxes). Depending on the model adopted by Such a deep involvement in the design and assembly chain,
the TELCO, either all the O-NEs of the WDM-Sys are from blade to systems, is an even greater challenge for a
involved in the process, or only the TPs devices. In both cases TELCO and, at present, difficult to quantify economically.
vertical integration is needed. It is conceivable that the role of 'open consortia' could be of
From the SW perspective the main input is the WDM great importance in this respect, as they are managing to
controller/management subsystem. The SI must work closely seriously address the problem of a white box ecosystem in all
with the supplier of the SW for vertical integration (adaptation of its facets. A further stimulus that could lead TELCOs to
layers; drivers for devices; suitable application software; start a direct involvement in blades/O-NEs assembly may be
integration with TELCO’s automation BSS/OSS; etc..). the availability on the market of a set of competitively priced
Other essential input ingredients are a collection of standard equipment with pre-certified on-board SW compatible and
specifications for vertical and horizontal interoperability and a ready to be integrated with modest effort, but, this scenario
clear transmission design criteria. seems unrealistic in the short term.
The system assembly involves mainly the SI in all the
phases of design and integration, leaving the TELCO directly V. DEPLOYMENT AND OPERATIONAL ISSUES
responsible for validation and testing. Engineering and Fig. 7 shows a general network planning and design
innovation departments are responsible for adding and deployment, operation and maintenance process in an
continuously updating/replacing the equipment from a evolving perspective during the entire life of the network.
'catalog', from which other departments can participate in the Process depicted in Fig. 7 holds for any kind of disaggregation
process of network creation and upgrade. model adopted by the TELCO, but differences exists as regard
The system vendor role becomes that of a supplier of the actor who is responsible each singular activity. The model
certified sortware and hardware, including the specification proposed in Fig. 7 is a simplified vision which assumes
rules for configuration, interconnection and operations. The network planning and design (P&D) as a single activity (box
assembly and integration then becomes the responsbility of the 4). This is motivated by an expected future highly dynamic
TELCO, or its chosen SI. and uncertain environment in terms of changes on both service
The system development phase is continuously running in demand and technologies. The new challenging environment
background with feedbacks from the field, the client needs, requires quick changes and adaptations of the network, not
new HW and SW releases or even new technology and the only in terms of reconfigurations, assured by the control plane,
evolving of standards. It is exactly the continuous interaction but also in terms of new HW and SW installations and
between developers (TELCO engineering and innovation updates. The traditional approach, which assumes a phase of
departments/SI) and operators (e.g. network operation long term planning followed by design and engineering phases
departments) in a synergistic way (in the information as separate steps, seems unsuited to react effectively in such a
technology world this approach if often termed DevOps [23]) fast changing context.
that can add value to all the process and can justify TELCO Referring to numbered boxes in Fig. 7, inputs of the whole
involvement in system design and assembly. process are the service demand, i.e. the list of client circuits to
However the 'in-sourcing' of these activities alone may be be allocated with the required QoS (box 1), and the WDM
of small benefit or even add unwanted overhead and transmission system (box 2). WDM transmission system (i. e.
constrains, while its synergistically incorporation in a leaner the updated equipment catalog available for the network
and autonomous service and network lifecycle process, should creation) is as results from the scheme of Fig. 6 and depends
give TELCO the readiness, flexibility and freedom from on the disaggregation model chosen. The process to obtain the
vendor lock-in often wished. WDM-Sys is analysed in deep in Section IV and is
Of course the key and critical role is played by the SI in summarized in Fig. 7 by the box 3.
strict combination with testing and innovation departments, The first step of the process is a new P&D (box 4) activity:
which are virtually the 'system vendor' for the TELCO. In in the first period of the network life (T=1) P&D gives the
order that this intermediary not become the bottleneck, a guidelines for the early deployment of the network with
careful selection of personnel and skills is paramount, and we equipment installation and service activation (provisioning) of
believe this is the fundamental challenge for the TELCO. the initial set of demands (box 5).
B. Optical Network element and Subsystem blades assembly In following periods (T>1), triggered by the change of the
demand or the availability of new functionalities on equipment
The process of O-NE assembly under TELCO responsibility
(trigger point, diamond 8), network P&D gives the update on
is identified by box 5 in Fig. 6, as opposed to the direct supply
equipment installation (new equipment), the upgrade on
 8

healing capabilities, able to reconfigure the O-NEs

automatically upon failures. Network operators, with the
deployment of control plane intelligence have witnessed a
significant decrease in the number of service outages upon
failure events, such as fiber cuts. The self-healing capabilities
need to be maintained when disaggregation is applied.
Optical restoration is the ability of the optical network
control plane to react to catastrophic events affecting services
configured with the capability of being restored into a
different network path. In the partial disaggregated option, a
cooperation between OLS and transponders is needed to
complete the restoration. Performance information and alarms
need to be exchanged in order to take the appropriate
decisions.
Fig. 7. Planning and Design, Deployment, Operation and Maintenance The P&D is required to be applied in a new cycle when
process flow during the Network Lifecycle. something relevant happens. Trigger point (box 8) models
functionalities (new HW or SW on already deployed three types of events that can require modifications on the
equipment), provides equipment to be uninstalled and services network. The first is the change in the service demand (here
to be activated or released (box 6). In doing so, the Network intended as an important change in the demand which imply
design takes into account the current state of the network possible interventions in the network equipment to be installed
before the required upgrade. or upgraded). The second triggering event is the availability of
Network P&D (box 4) is under the responsibility of the an upgrade in the WDM transport system and in this case a
TELCO for any type of disaggregation model. Nevertheless, replacement of equipment or an upgrade of functionalities is
while in case of full aggregated option the design tools are required. Finally, when the network ends its lifecycle, the
usually provided by the vendor, in case of disaggregated corresponding trigger event starts the final decommissioning
options the operator has to develop ad hoc design tools taking of the network (box 9).
into account the heterogeneous WDM transport system which
is made of pieces of equipment assembled with parts from VI. TECHNO ECONOMIC ANALYSIS
many sub-systems vendors. As regards this point, the It is yet to be demonstrated, that the introduction of
disaggregated options could be a critical issue for the operator. disaggregation in optical networks can lead to economic
Installation, upgrades and decommissioning in an initial and savings on both CapEx and OpEx, and then in Total Cost of
incremental deployment phases (box 5 and 6) is under the Ownership (TCO) of the network, when the reference is the
responsibility of the vendor for the whole network in case of consolidated fully aggregated model [24]. The main argument
full aggregated optical domain option, or for the only OLS in brought for this potential expenditure saving is the removal of
case of partial disaggregated option. The same tasks are under the vendor lock-in. Freeing itself from the single vendor and
the responsibility of the operator or the SI in case of putting in competition many suppliers, a TELCO would
disaggregated options (all network for full disaggregation and reduce the price of equipment, especially for the HW
TP part only for partial disaggregation). components. On the other side adopting a disaggregated
When the network is ready after the implementation of a option introduces the cost for the system integration and for
given cycle of planning and design, deployment and service the internally development, or the buying from third parties, of
provisioning, it is operated and maintained by the dedicated SW for network control. Such costs are very hard to predict
function of operation and maintenance (O&M) that, for because they depend on the Operator’s strategy on which
simplicity, is grouped in a single task (box 7). activities are carried out internally, and which to outsource.
O&M is another crucial and potentially critical task. Where the operator chooses to do it internally the development
Normally under the responsibility of the operator, the way in of control SW, for instance following the DevOPs [23]
which O&M is implemented changes noticeably for the methodology, the costs required for re-skilling personnel and
different disaggregated options. In case of the fully aggregated for company organizational changes have to be taken into
option O&M can relies on specific services that the vendor account. A dispute that involves TELCO and ICT service
deliver to the operator and which is tailored on the providers, when they have to decide their plans for the
homogeneous deployed WDM transport system. In case of network development, concerns the most efficient way to
disaggregated options the operator have to develop its own achieve CapEx and OpEx reduction through the introduction
O&M applications and services which, as it is for P&D tools, of SDN and, possibly, white boxes. It is important to underline
have to take into account the heterogeneous environment of a that a reduction of OpEx can be achieved by the introduction
disaggregated solution. Operator have to handle this task with of SDN, regardless of the aggregation model adopted. Indeed,
the SI and complexity and cost could be important issues in it is very hard to assess how disaggregation and softwarization
doing that. During the O&M cycle, maintaining a Service would concur in the OpEx reduction.
Level Agreement (SLA) in the TELCO networks requires self- A paradox, also named 'cognitive dissonance' in [25], is raised
 9

because on one side the introduction of disaggregation and assure, on all the paths, to route optical circuits in the WDM
softwarization is recognized as the way to extract significant layer without regeneration. An OTN layer for the grooming of
OpEx reduction in the mid- and long- term. But, conversely, it lower rate circuits is also present. The Unit of Cost (UC) for
requires significant up-front investment and so a high CapEx this evaluation is the cost of a coherent 100Gb/s TP. All other
to be accomodated in the short term. Under the pressure of costs of HW and SW components are referred to this UC. The
increasing capacity to be provided with limited budget, the cost model and cost parameters for WDM and OTN
investment for the network paradigm transformation (i.e., equipment have been taken from [27].
softwarization and disaggregation) are delayed, and the After an approximate network dimensioning, which
benefits for OpEx shifted into the future. provides the bill of materials, the total network cost is
In this section we present an evaluation limited to a calculated for the two scenarios under analysis. The bill of
comparison of CapEx for the two options: the Fully materials is the same, in terms of WDM and OTN equipment,
aggregated (Fig. 2) and the Fully disaggregated WDM system and costs differ between the network scenarios analysed only
(Fig. 4). Both the options assume a centralized SDN controller for the prices of HW and basic SW components, and for the
and the difference are on the controller and its interfaces: presence in the disaggregated scenario of the cost of the
proprietary control system or WDM open control system, development of control SW and of the system integration.
possibly developed by the operator itself. The approach Fig. 8 gives the cost in UC of the compared configurations.
followed is inspired to the method applied in [26] to estimate The bar on the left shows the cost components for the fully
the saving of a disaggregated packet transport network based aggregated model. Control plane SW cost (two slices on the
on SDN when it is compared to a conventional IP/MPLS top, for common SW and for licenses of SW on equipment
network. While in [26] the comparison is between the respectively) is about 15% of the total cost. The second bar
presence and the absence of SDN in a packet transport from the left, labelled as HW Equicost, is for the
network, in this section a CapEx analysis is performed on a disaggregated solution characterized by the same HW cost
SDN WDM network in the two options of fully aggregated (HW discount = 0%) as the Fully aggregated case (highlighted
and full disaggregation as regard optical equipment. by the horizontal arrow in Fig 8) and integration and SW
Two types of network models with its own WDM transport development nominal costs. HW Equicost disaggregated case
equipment and related control system are considered for the shows a total cost 10% higher than the fully aggregated
two selected scenarios under comparison. In the first, the fully solution, essentially for the presence of integration cost which
aggregated scenario, pieces of equipment (ROADM, TP, ILA, is not compensated, in this specific case, by a cheaper HW.
OLT) are provided by a single vendor together with the Nominal cost case (third bar from the left of Fig. 8) reaches a
control SW, which is accounted separately. In the second, the reduction of 7% taking advantage of a discount of the HW of
fully disaggregated scenario, pieces of equipment of the same 20%. The other three configurations on the right of Fig. 8
type can be provided by more than one vendor, and their cost show how combinations of cost parameters can change from a
is assumed to be lower of a certain percentage (named HW situation of significant disadvantage (+20%) to a situation of
discount, a parameter) of the piece of equipment which big benefit (-32%) as regard the total cost for a network
performs the same functionality in the fully aggregated solution based on disaggregation. In general, according with
scenario. the assumptions made, CapEx reduction for the fully
The cost model is applied to a network scenario of disaggregated model can be reached if a moderate cost
metro/regional WDM network of 50 nodes linked by 90 fibre reduction of the order of 20% or more is available for HW, the
links and carrying a mix of 1, 10 and 100 Gb/ s client circuits control SW cost is the same, and the Integration cost does not
(about 1,000 in total). The diameter of the network is of the exceed the 15% of the total CapEx.
order of 300 km and assumptions on TP and line systems To complete the analysis on TCO an evaluation on OpEx
should be also considered. A reduction of OpEx can be
achieved thanks to the introduction of SDN and, assuming that
SDN is present in all the solutions including the fully
aggregated one, this reduction will be enjoyed regardless of
the disaggregation model adopted for the network
implementation. Impact of disaggregation on OpEx, which
could make the difference between models, is difficult to
evaluate and it is highly dependent to the specific context
(e.g., the process and organizational models of the TELCO
company). Such topic is left for a further dedicated analysis.

VII. CONCLUSION
In an endless era of continuously growing traffic, but with
margin on revenues that shows the opposite trend, TELCOs
Fig. 8. Cost of Fully Aggregated optical domain solution (bar on the left)
are looking with high interest in 'disaggregated' models for the
compared with costs of Fully Disaggregated solutions (five bars in the box). cost-reduction opportunities they seem to promise. Setting
 10

aside L2/L3 applications and services, this article focused on [5] J. Donovan & C. Benko, "AT&T’s Talent Overhaul", Harvard Business
Review, Oct. 2016. Available: https://hbr.org/2016/10/atts-talent-
disaggregation at the WDM transport layer. Four models with overhaul
different degrees of disaggregation at the WDM layer were [6] H. Adams, “Optical Network Strategies Service Provider Survey –
identified and discussed, namely: full disaggregation based on 2017”, IHS Markit, Dec. 2017. Available:
https://technology.ihs.com/588823/optical-network-strategies-service-
bare metal blades; full disaggregation based on white boxes;
provider-survey-2017
partial disaggregation (OLSs and multi-vendor transponders); [7] S. Hardy, “Open optical line system pros and cons at the Open Optical
and full aggregation (this last one is the common mono vendor Conference”, Lightwave Online, Nov. 2017. Available:
scenario based on a black box approach, characterizing the http://www.lightwaveonline.com/articles/2017/11/open-optical-line-
system-pros-and-cons-at-the-open-optical-conference.html
legacy). In all these operational models (except the last) [8] OpenROADM web site: http://www.openroadm.org/home.html
TELCOs are actively involved in the design, assembly, testing [9] OpenConfig web site: http://www.openconfig.net/
and lifecycle management of the WDM transport Systems [10] Open Optical & Packet Transport (OOPT), TIP project web site:
https://telecominfraproject.com/project/backhaul-projects/open-optical-
deployed in their transport networks: each disaggregation packet-transport/
model has a different 'spectrum' of lifecycle support that a [11] Lumentum Inc. ,"Transport ROADM Whitebox/Graybox". Available:
TELCO must sustain. https://www.lumentum.com/en/optical-communications/products/sdn-
whiteboxes-and-grayboxes
The roles of the Operator, the system Vendor and the [12] Fujitsu, “Network Evolution with 1FINITY Series”. Available:
Integrator (the last is an important player supporting TELCO, http://www.fujitsu.com/global/products/network/products/1finity/
which is required to ensure that different parts of a system are [13] Juniper Networks Inc.,”TCX1000 Programmable ROADM”. Available:
https://www.juniper.net/us/en/products-services/packet-optical/tcx-
designed and properly assembled) have been analysed with the series/tcx1000/
support of a model for the processes of system integration and [14] H.-J. Schmidtke and L. M. Garcia, “Driving Openness in Optical
network creation, operation and maintenance. Networks: An Update from the OOPT Project”, Nov. 2017. Available:
http://telecominfraproject.com/riving-openness-in-optical-networks-an-
The key question is if the advantages envisaged by the
update-from-the-oopt-project-group/
disaggregation are justifiable from technical, organizational, [15] M. Bjorklund, "RFC6020 - YANG - A Data Modeling Language for the
and economic viewpoints. Actually, embracing the Network Configuration Protocol (NETCONF)", IETF, Oct. 2010.
disaggregation implies for a TELCO a radical change of the Available: https://tools.ietf.org/pdf/rfc6020.pdf
[16] R. Enns, M. Bjorklund, J. Schoenwaelder, A. Bierman," RFC6241 -
consolidated paradigm in planning, engineering, deploying, Network Configuration Protocol (NETCONF)", IETF, June 2011.
operating and troubleshooting the network. Well established Available: https://tools.ietf.org/pdf/rfc6241.pdf
processes have to be abandoned while new skills, especially in [17] T. Szyrkowiec, A. Autenrieth, and W. Kellerer, “Optical Network
Models and Their Application to Software-Defined Network
SW development, require development, and new mode of Management”, International Journal of Optics, Vol. 2017, Article ID
operation, for instance adopting the DevOps methodology, 5150219, 2017. Available: https://doi.org/10.1155/2017/5150219
must be introduced. TELCOs that embrace disaggregation will [18] I. Lyubomirsky, B. Taylor, and H.-J. W. Schmidtke,"An open approach
for switching, routing, and transport", Nov. 2016. Available:
need to rely less on the vendors and re-build their knowledge https://code.facebook.com/posts/1977308282496021/an-open-approach-
of the optical network and devices. This transition, if chosen, for-switching-routing-and-transport/
is a big challenge. [19] A. Eckert, L. Martin Garcia, R. Niazmand, and X. Wang, "Wedge 100:
More open and versatile than ever", 18 October 2016.
As a further element, a preliminary CapEx analysis was https://code.facebook.com/posts/1802489260027439/wedge-100-more-
performed comparing a disaggregated solution to an open-and-versatile-than-ever/ )
aggregated legacy one. The results critically depended on a [20] Edgecore Networks "AS7712-32X-EC Datsheet". Available:
https://www.edge-core.com/_upload/images/AS7712-32X-
reliable estimate of the SW development and integration costs.
EC_DS_R01_20170608.pdf
In fact, the expected cost saving achievable with the [21] Open Disaggregated Transport Network (ODTN) web site:
disaggregation, can be totally absorbed and replaced by the https://www.opennetworking.org/solutions/odtn/
high costs of integration and ad hoc SW development for the [22] Open Networking Foundation (ONF) "Open Network Operating System
(ONOS)" web site: https://onosproject.org/
control plane. An OpEx, analysis was not carried out in this [23] DevOps community web site: https://devops.com/
article but, in a first approximation, the opportunity to benefit [24] Ž. Bulut, “Low Footprint Open Optical Networks”, Open Optical
from the introduction of SDN for operational cost reductions Conference, Dallas, TX, Nov. 2017. Available:
http://www.lightwaveonline.com/content/dam/lw/Marketing/Coriant.Res
applies to all models, no matter the level of disaggregation. ponsible%20Networking_Low%20FootprintOpenOpticalNetworks.v07.
Specific drivers for OpEx reduction due to disaggregation pdf
require a further investigation. [25] “Cognitive dissonance: CapEx, OpeX, SDN, DevOps and White Box”,
article on Plexxi web site. Available:
http://www.plexxi.com/2013/10/cognitive-dissonance-capex-opex-sdn-
REFERENCES devops-and-white-box/
[26] B. Naudts, M. Kind, S. Verbrugge, D. Colle, M. Pickavet, “How can a
[1] L.A. Barroso, J. Clideras, U. Hölzle, "The Datacenter as a Computer: An mobile service provider reduce costs with software-defined
Introduction to the Design of Warehouse-Scale Machines, Second networking?”, Network Management, Vol- 26, N. 1, Jan./Febr. 2016
Edition", USA, Morgan & Claypool 2013. Available: [27] F. Rambach, B. Konrad, L. Dembeck, U. Gebhard, M. Gunkel, M.
https://research.google.com/pubs/pub41606.html Quagliotti, L. Serra, V. Lopez, “A multilayer cost model for metro/core
[2] Open Compute Project (OCP) web site: http://www.opencompute.org/ networks”, Jou. of Opt. Comm. and Net.,Vol. 5, No 3, March 1, 2013.
[3] Telecom Infrastructure Project (TIP) web site:
http://telecominfraproject.com/
[4] Central Office Re-architected as a Data Center (CORD) web site:
https://opencord.org/