Sustaining a Vertically Disintegrated Network through a Bearer Service Market

Sustaining a Vertically Disintegrated
Network through a Bearer Service Market
Petros Kavassalis1, Thomas Y. Lee2, Joseph P. Bailey3
1
Ecole polytechnique, Centre de Recherche en Gestion
<petros@rpcp.mit.edu>
2
Massachusetts Institute of Technology, Context Interchange
<tlee@mit.edu>
3
Massachusetts Institute of Technology, Technology, Management and Policy
<bailey@rpcp.mit.edu>
Abstract
Based upon the Internet perspective, this chapter will attempt to clarify and revise
several ideas about the separation between infrastructure facilities and service
offerings in digital communications networks. The key notions that we will focus on
in this paper are: i) the bearer service as a technology-independent interface which
exports blind network functionality to applications development; ii) the sustainability
of an independent market for bearer service and the organizational consequences
associated with such a market.
1. Introduction
During the past two years, applications like email and the World Wide Web have
combined with evolving network protocols to propel the Internet into the heart of a
computer and communications convergence. Central to the Internet’s immersion into digital
convergence has been the effectiveness with which the Internet Protocol (IP) has played the
role of “spanning layer.” [1]
The IP abstraction enables applications to request network services independent of
underlying, physical network technologies. Moreover, new underlying network technologies
may either substitute for or co-exist with existing network technologies without
significantly affecting the broader system. Based on this abstraction, the National Research
Council recently articulated the "Open Data Network (ODN)" as an architecture for the
networks of the future that generalizes the principle of separating service offerings from
infrastructure facilities as demonstrated in the Internet [2]. In the same way that IP serves
the Internet, the ODN relies upon a "bearer service" to function as a technology-independent
network layer that resides above the technology substrate and enables interoperation
between diverse, high-level applications and various underlying network infrastructures
(figure 1).
The NRC report describes an Open Digital Network (ODN) as a four-level layered
architecture: "i) at the lowest level is an abstract bit-level service, the bearer service, which
is realized out of the lines, switches, and other elements of networking technology; ii) above
this level is the transport level, with functionality that transforms the basic bearer service
into the proper infrastructure for higher-level applications (as is done in today’s Internet by
the TCP protocol) and with coding formats to support various kinds of traffic (e.g., voice,
video, fax); iii) above the transport level is the middleware, with commonly used functions
(e.g., file system support, privacy assurance, billing and collection, and network directory
services); and iv) at the upper level are the applications with which users interact directly.

This layered approach with well-defined boundaries permits fair and open competition
among providers of all sorts at each of the layers”. [2]
Certainly, the Internet demonstrates the technical and functional robustness of a
technology-independent bearer service abstraction [3]. The bearer service is intended to
support requests for service from all applications and to recognize all substrates, but as both
application and infrastructure innovations turn increasingly towards user-oriented models of
network architecture, technology and policy considerations related to the generalization of
this abstraction should be carefully studied. Such a service blurs the boundaries of
telecommunication markets.
Figure 1: The Bearer Service Concept
For example, the promises of Internet telephony to combine the benefits of the public
switched telephone network (PSTN) and the Internet would be possible through a bearer
service – even though Internet telephony would weaken market boundaries and challenge
the regulatory environment [4]. Regulators have a difficult time categorizing bearer service
providers as belonging to any one existing market because such providers can offer services
that cut across many existing telecommunications markets. Businesses that develop
ubiquitous services or tailor applications for customers may be threatened by market
entrants who have competitive products built around a new technology architecture that is
able to provide great flexibility in applications design — the bearer service. Finally,
customers can benefit from an integrated services environment because their data and voice
communications can be transmitted across multiple telecommunications infrastructures. A
new era of interoperability [5] is possible through the bearer service.
However, questions for this new market abound. Not only is the business model in
question, but the technology is also in flux. In this chapter, we hope to shed some light on

the bearer service issue by answering some of the questions arising from this new bearer
service market. First, what does a technology independent bearer service look like? What
are the technical characteristics of a bearer service that satisfies the requirements of both
elastic and real time applications? Second, is a segregated network that separates
infrastructure facilities from service offerings by a bearer service an economically viable
market organization? This chapter spans both technical and economic issues to ascertain
the potential of a market for bearer services. Specifically, this chapter will address the
following subset of bearer service market questions:
i) What comparative advantage does a technology-independent bearer service provide
over homogeneous infrastructures that support multiple service classes (e.g. ATM)? This
question is especially important given the increasing interest in Integrated Networks
solutions,.
ii) In a recent paper Gong and Srinagesh [6, 7] suggest that layered network
architectures may inherently engender vertically integrated markets for network services. Is
this a general statement or a phenomenological one? And if the phenomenological reading
is correct, what factors tend to suggest the viability of disaggregated markets for network
services both at and above the spanning layer as described above?
The remainder of the chapter is organized as follows. Section 2 employs a
comparative analysis to define concepts of the bearer service and the layered network
architecture concepts. Considering IP as a bifurcation point in the evolution of network
design, the bearer services of traditional communications infrastructures and the Internet
(both the current best-effort and the future Integrated Services Internet) are surveyed to
elicit design characteristics and functional differences between a technology-dependent and
independent bearer service. We suggest that in a network design with a technology-
independent bearer service, the communications network supports a flexible organizational
model capable of dealing with unanticipated (applications’) variability.
Section 3 associates the technology-independent bearer service with the Internet
organizational model: a flexible system of regional or more extended backbones and access
links (or access networks) to these backbones, managed by the Internet Service Providers
(ISPs). Specifically, we address the question of whether the ISP model, which is
characterized as a model where network operators exhibiting varying degrees of vertical
integration compete in an open market, can sustain itself, or if one monolithic, integrated
firm will emerge from mega-mergers?
To answer this question, we begin by considering the work of Gong and Srinagesh [6],
who argue that a stable and sustainable equilibrium for healthy bearer service market
growth might not be possible. Our analysis closely follows the definition for a bearer
service formulated in section 2. Arguing that the bearer service is not a commodity product,
we identify differentiable service attributes upon which an independent bearer service
market could form. Through differentiation, the bearer service market can avoid a Bertrand
equilibrium (i.e. pure price competition). Furthermore, we challenge assumptions about
perceived trends towards vertical integration, by noting the relative independence of assets
between bearer service providers and potential buyers of this service (i.e. connectivity
providers who are situated at the edge of the network versus other kinds of service
providers).
The chapter concludes in section 4 by underlying the role of the bearer service market
in creating the potential for competition between independent and integrated providers.
However, as revealed in section 3, while a market for bearer services may be economically
viable, it will also likely be fragile. The conditions for sustaining a bearer service that
exhibits "spanning" capabilities and promotes efficient interoperability (as in the Internet),
requires an alignment of both technical and economic considerations. In addition, new
institutions, not necessarily government-dependent, may be required to enable the market to
form.
2. From the railroad gauge to information bitways: the evolution of the bearer
service functionality

Given a pre-specified set of applications and a physical network which may include
more than one substrate technology (e.g. a Ethernet-based LAN connected to a Frame relay
network), the bearer service (BS) constitutes those common1 functions which are
implemented throughout the network rather than in the network end nodes and are
necessary for pairing each application’s communication requirements with the performance
characteristics of all components of the heterogeneous network.
For example, consider voice and fax data services transmitted via traditional, analog
telephony between the United States and Canada. Neither within nor between each country
do the customers need to share the consumer premises equipment (CPE). Likewise, at the
level of network technologies, materials like cables, switches, and even local numbering
conventions might vary. However, since all CPE and network subsystems observe a
common, international numbering convention and all networks and CPE transmit, carry,
and receive signals in the 4 kHz frequency spectrum, voice conversations are conducted and
faxes are conveyed. For voice and fax data services, the international numbering scheme,
combined with the 4 kHz channel constitute the bearer service.
More abstractly, computer and communications technologies may be separated into
three layers. The physical infrastructure (e.g. wires, switches, etc.) resides at the lowest
layer. At the top lies the set of applications and service offerings supported by the
underlying infrastructure. A spanning layer2 bridges the two [1]. An application requests
network services through the spanning layer to the substrate technologies.
In the early public switched telephone network (PSTN), telephony was tightly coupled
to a specific infrastructure so the spanning layer supported only a single application and one
technology substrate (typically, the spanning layer was located in the wires). The
development of newer applications for computers and communications as well as advances
in substrate technologies, prompted a refinement of the spanning concept. In the presence of
a diverse suite of applications and a heterogeneous network, the bearer service (BS)
constitutes a spanning layer which escapes from the wires and thus supports all applications
over the entire network. This section uses Piore’s model of organizational flexibility and
production system transformation as a methodological framework for tracing the evolution
of the spanning layer towards a technology-independent bearer service. Advances in
shipping and the transport of physical goods are used as a metaphor for the transformation
of yesterday’s PSTN into tomorrow’s ODN.
2.1. Production technologies and flexible specialization
Piore [8, 10] describes the on-going transformation of industrial production systems
towards greater variety and flexibility as a four-stage evolution3. The products in such
systems are comprised of both independent and interdependent design features; changes in
design features mark the different evolutionary stages. Independent design features "can be
varied in isolation without complementary changes in other features of the design" while
interdependent features "require a number of complementary adjustments" [8].
The initial stage, mass production is characterized by a production system tailored to a
single product. There is no room for variation. In Mass production with cosmetic variation,

1
If every application uses the function, then it is certainly a function in common and
unambiguously a component of the BS. If only one application uses the function, then perhaps it is more
appropriately considered part of the application (or of the transport layer). If two or more applications utilize
the function but not all applications in the set use the function, then we need to question whether the function
belongs in the BS. Recall also that a separate metric for distinguishing BS functionality is whether that
function can be implemented in an endnode. BS functions include only those functions which cannot be
implemented in an endnode.
2
As suggested by Clark, "a spanning layer is characterized by three key parameters that
characterize the sort of interoperation it supports: i) the span of infrastructure options over which the
interoperation occurs, ii) the range of higher level applications the spanning layer supports, iii) the degree of
symmetry in the services of the spanning layer and its interfaces, and in the higher layers up to the application
definitions". [1]
3
This work draws on The Second Industrial Divide [9]

product design may slightly vary existing or may introduce new independent design
features. "The notion of cosmetic variation seems to imply a sharp dichotomy between
design changes which are easy to make and those that are not" [8].
Flexible mass production extends cosmetic variation by introducing the potential for
change in interdependent design features. Flexible mass production explicitly identifies, a
priori, both the set of product design features which is subject to change and the set of
values which each design feature may take. Therefore, the flexible mass production system
represents a finite number of products, which vary in more than simple cosmetics.
Diametrically opposite mass production is flexible specialization where variation is
virtually infinite. However, closed flexible specialization includes those systems where the
set of design features that varies is defined a priori, but the domain over which each varying
feature ranges is unknown. By contrast, open flexible specialization where both the set of
variable design features and the domain over which each variable ranges is potentially
infinite.
A different cognitive model applies in each stage, with flexible specialization
involving a balance between "a deepening of understanding within a given cognitive frame
and the pull to reintegrate back (in the production process) to a different frame in order to
produce a sellable commodity" [11]. Similarly, a technology-independent bearer service
offers more than a predesigned set of services. Rather, it supports an “application-blind
interface” that enables the introduction of new applications independent of the initial
strategies and service offerings of Telecommunicationos Operators (TOs). given (ex-ante
designed) set of services: it provides a basic functionality, an application-blind interface in
the network jargon. This functionality can be easily reintegrated to the application vendors
and users' cognitive frames, thus allowing them to introduce new applications and operate
independent of the strategies and the service offerings of the Telecommunications Operators
(TOs). To illustrate the continuum that spans well-defined, mass production systems and
flexible, application-blind interfaces, we will discuss two contrasting metaphors, the gauge
and the container.
2.2. The gauge metaphor
The gauge of a railroad is defined as the distance between rails or between the flanges
of the wheel sets on a railroad car. The gauge determines the tracks upon which a given
railroad car may travel. By extension, the gauge therefore also determines which railroad
companies may exchange rolling stock and the transparency with which a customer may
transport freight across boundaries between different railroad companies. Accordingly,
diversity in gauge standardization implies transaction costs and other inefficiencies as
customers and freight traverse rails of different gauges. Thus, for reducing technical
complexity and transaction costs to internalizing mutual network externalities, railroads
have been progressively converging towards a gauge standard [12, 13].
The emergence of the spanning layer concept may be derived directly from the
convergence towards a rail gauge standard. Railroad tracks comprise the physical network.
Differences in rail cars represent distinct applications from which a customer might choose.
Gauge dimensionally is therefore a layer which resides between the physical track and the
applications. Gauge standardization expanded the services that a particular rail system could
offer by extending the reach of the rail network and expanding the scope of traffic (the
kinds of cars) that could be carried. Standardization reflected a shift towards some
flexibility where rail car design could vary (as long as it conformed to the gauge standard)
and traffic could move across network boundaries.
2.3. The spanning layer in traditional communications networks
Communications networks have traditionally been vertically bundled. Whether for
telephony, radio, broadcast television, or community access television (cable),
infrastructures have long been closely coupled to service provision. As Tennenhouse et al.
notes [14], "telephone and cable services are each carried over their own wired systems.