Internet Host: nic.cerf.net
Directory: farnet
Subdirectory: farnet_docs
Filename: nsf-backbone-rec
Last Updated: Nov 8, 1991
Recommendations
to the National Science Foundation
from the Board of FARNET, Inc.
Regarding Inter-midlevel Connectivity
after the Expiration of the
Current NSFNET Backbone Agreement
November 1, 1991
Introduction
With support from the National Science Foundation, FARNET (the
Federation of American Research Networks) conducted a workshop
and electronic discussion in the late summer and early fall of 1991
on the question of how connectivity among midlevel networks
should be implemented after the current agreement for NSFNET
backbone services expires in November, 1992. Participants included
representatives from FARNET member networks, other NSFNET
stakeholders (including carriers and leaders in university
information technology), Federal agency representatives, and legal
and economic experts. We gratefully acknowledge the participation
of all those who helped to make this a fruitful and instructive
process. As a result of this extended dialogue, the FARNET Board is
pleased to make the following comments and recommendations to
the National Science Foundation.
Section I CONTEXT: The NREN, the NSFNET Backbone,
and the Midlevel Networks
In November, 1992 a five-year agreement between MERIT, Inc. and
the National Science Foundation for the operation of a national
backbone network for the NSFNET will expire. Viewed from
almost any perspective, the pace of activity in the networking arena
since 1987 has been astonishing. Network traffic has grown
exponentially; bandwidth has increased by a factor of nearly 700;
the number of networks connected to the backbone has increased
from a few hundred to more than 3,000; and the user population is
now estimated at more than 2,000,000 people nationwide.
In addition, Congress and the Executive Branch, with a level of
support from higher education and industry that is remarkable,
have created a cohesive High Performance Computing and
Communications program. A key component of this program,
which is designed to push the frontiers of computer
communication technology and maintain a U.S. leadership position
in high performance computing, is the construction of a National
Research and Education Network (NREN). According to the Office
of Science and Technology Policy [1], NSF is expected to coordinate
the "harmonization of existing agency networks" into the NREN
and to support research into high-speed protocols, switches, and
other technology.
Because of the extraordinary growth of the NSFNET and the
visibility of the HPCC program in government, industry, and
academia, the next generation of the network (the "interim
interagency NREN," or IINREN) will be developed and deployed in
an environment very different from that of the mid 1980s.
The NREN user community now includes many Federal agencies,
libraries, hospitals and health care professionals, and a growing
number of educators at the elementary and high-school levels. The
list of stakeholders includes the RBOCs, the interexchange carriers
(the three largest of whom now have operational roles in the
NSFNET), several resellers of value-added network services (such
as ANS, Alternet, and PSI), more than 25 state and regional
networks, many publishers and others in the commercial
information industry, and virtually all of the companies that supply
the telecommunications industry with equipment.
At the same time, the pressures for commercial use of the backbone
facilities, from both potential consumers and potential providers,
are tremendous. Vendors are clamoring to deliver software,
technical support, instruction, news, and information across the
network, and buyers are ready to purchase. But the market for
network-based services is immature, in part because the value of
such services is not well understood and also because restrictions on
the commercial use of government-sponsored networks have
discouraged such use.
Because of this immaturity, and in view of the ambitious goals of
the NREN program, a completely market-driven approach to the
evolution of the network at this time is widely viewed among
FARNET members as unlikely to satisfy some of the most
important desiderata described below.
Section II DESIDERATA: Critical Factors in the Evolution
of the NSFNET to the Interim NREN
FARNET members believe that the following are the critical issues
surrounding the continued development of the NSFNET:
� Establishing a strong Federal NREN program, which will
leverage significant private investment in the next generation of
the network and will attract state and local funding and use
� Maintaining or improving U.S. technology leadership in
the areas of high performance computing and communications,
and developing strong mechanisms for technology transfer
� Providing an adequate level of stability and predictability,
both in the operation of the network and in any transition to new
technology or management structures, with particular emphasis on
the requirements of the mission agencies (NASA and the
Department of Energy)
� Greatly improving support for user and information
services network-wide, including access to both public and private
information resources, with early resolution of copyright and other
intellectual property issues
� Ensuring adequate levels of performance, which will
require that more attention be directed to network management,
routing arbitration, cross-connect mechanisms, monitoring and
problem resolution tools and procedures, etc.
� Offering a variety of choices for state and regional networks
in backbone services (providers, location, access speeds,
technologies, costs, etc.)
� Rapid elimination of restrictions on delivery of
commercial services across the network, and structured transition
to commercial provision of network systems and services
� Availability of widespread and equitable access to the
network at reasonable cost, with a minimum level of guaranteed
interconnectivity among service providers
� Providing mechanisms to ensure global interconnectivity
without undue restrictions
Section III RECOMMENDATIONS: FARNET
Recommendations to NSF Regarding Inter-midlevel
Connectivity in the Interim NREN
1. The multi-tier model for providing network services is valid and
should be preserved.
Experience confirms the validity of the three-tier approach
(backbone, midlevel networks, and campus networks) that NSF
adopted in the mid-80s. The vibrant infrastructure of state,
regional, national, commercial, and non-profit providers that has
been created around the NSFNET backbone since 1986 is the best
argument for this approach.
Regional and state networks have been able to leverage resources,
both public and private, that may not have been available (or even
apparent) to a national organization. The multiplicity of providers
has generated considerable innovation in services and products and
has increased the number of knowledgeable and committed
network experts. We do not believe that the NSFNET could have
expanded as rapidly and successfully as it has without strong local
and regional involvement. At the same time, the network has
profited from ambitious and successful efforts by Merit, IBM, and
MCI (and now ANS) in establishing the NSFNET backbone.
We expect that the structure of the IINREN will continue to evolve
and change as current providers reassess their missions and new
providers emerge.
For example, we are not sure that future connections between
midlevels will necessarily be of the type we have today, with
dedicated private lines linking centrally managed routers. Other
emerging topologies deserve consideration, including CIX/FIX-like
structures (shared FDDI-based interconnection points) and shared
use of national cell relay or frame relay networks.
For convenience, in this paper we will refer to both traditional
backbones and the use of alternate topologies in the top level of the
three-tier hierarchy as "top-level backbone services."
2. Strong NSF support for top-level backbone services must
continue.
NSF must continue to provide strong support for the development
of the top level of the IINREN, including support for robust and
capable backbone services. As called for in the HPCC program, this
should include funding for higher-level applications and user
support, as well as new protocols, switch and transmission
technologies, and higher bandwidth. In addition, all targeted users
should have access to the IINREN at appropriate bandwidths, to be
determined in conjunction with users and midlevel providers.
NSF has provided critical leadership for the NSFNET backbone and
seed funding for the midlevel networks, as well as support for the
connection of hundreds of campuses to the network. Its actions,
particularly in the backbone arena, have galvanized the response of
industry. We expect that continued NSF support for top-level
services will preserve this important leveraging effect.
3. As the agency responsible for NREN facilities coordination and
deployment under the HPCC plan, NSF must assume a strong
management role vis a vis the core of the NREN (which
presumably will evolve from the current NSFNET).
We believe that NSF is well qualified to be the lead agency in this
area. Its leadership is committed to the HPCC program. It has
established excellent relationships with industry and higher
education, both in the NSFNET program and in the gigabits
research program. It has demonstrated vision and skill in executing
these programs.
During the next phase of development, NSF will need to take an
assertive position vis a vis the management and oversight of the
backbone if it is to meet the NREN/HPCC goals as established by
Congress and the Executive Branch. This is true whether there is a
single, or multiple, providers of backbone service. In particular,
NSF must play a proactive role to accelerate the harmonizing of
multiple agency networks and protocols into a single shared NREN.
4. The operation of the backbone network should be recompeted in
the GFY92 timeframe with multiple awardees.
As described earlier, many conditions have changed since 1987,
when the current agreement for backbone operations was made. In
view of these changes and of the emergence of the HPCC program,
we recommend that NSF issue a new solicitation for the provision
and operation of production quality midlevel interconnection
services. A variety of approaches would be encouraged and
evaluated through peer review. The new providers would begin
service in GFY93.
The award should include at least two service providers who would
be required to cooperate in the delivery of services, with specific
attention to the resolution of administrative, legal, technical, and
pricing issues associated with interconnection of facilities operated
by different vendors. The goal is to promote neutrality at these
cross connects. Experiments with pricing, transition to commercial
services, etc. should be conducted as part of the operation of the
network.
The redesign of the backbone should be based on engineering and
economic criteria. That is, the topology and capacity of the network
should be derived from the distribution and usage patterns of the
target population (existing and projected), and the design should
make efficient use of existing and planned carrier facilities.
The establishment of NSF-supported multi-provider backbone
services will encourage the development of market mechanisms to
ensure performance and enhance competition among providers
and will lead more rapidly to a structured transition to commercial
provision of network systems and services.
5. NSF should ensure that new technology is deployed in the
backbone very carefully, to protect the quality of service to the end-
user.
Because hundreds of thousands of users already rely on the
NSFNET for day-to-day support of research and educational
activities, it is clear that we require a production-quality network
today. The service level should approach the same standards as we
used to expect from voice service.
This means that the network operators must introduce new
technology very carefully into the backbone, balancing the need for
improvement with the need to maintain production-quality
service. To keep pace with technical changes and demand for new
services, they will have to upgrade facilities and equipment. At the
same time, they must be held to clear performance standards. NSF
should develop those standards in cooperation with the user
community (per Recommendation 8, below) and should enforce
them as part of its oversight function.
6. Midlevel networks should be able to exercise choice among
vendors of top-level backbone services.
To further the development of the market for commercially
provided network services, midlevel networks should be able to
exercise choice among providers. Mechanisms to allow choice by
midlevel networks of a production backbone awardee could be
implemented in a variety of ways, ranging from direct funding of
the midlevels for that purpose to designation of a preferred awardee
by the midlevel with NSF funds flowing directly to the backbone
operators.
7. The backbone awardees should not be able to take advantage of
their position to inhibit competition or to compete unfairly.
Because the provision of reliable top-level services is essential to
the success of the IINREN and the entire HPCC program, we
emphasize again that NSF should maintain a strong, central
oversight role in the provision of these services. Oversight should
include both technical and management issues. In particular, NSF
should guard that the winners of any new backbone solicitation do
not use their position to inhibit competition or compete unfairly.
8. Provider accountability for performance should be ensured
through the NSF award process.
All organizations that receive NSF funding for the delivery of
network services, from the campus level through the midlevels to
the top level, must be held to clear performance criteria. These
should be established by NSF in concert with the users and the
providers of the service. The criteria must be objective and
measurable and should be designed to ensure an acceptable level of
service end-to-end throughout the IINREN [2]. Reliability and
availability should be emphasized. Where the tools and systems to
measure performance and resolve network outages are inadequate,
NSF should provide funding to develop improved versions.
9. NSF should take a leadership role in developing mechanisms to
permit commercial traffic to be carried on the IINREN.
In the FCCSET report on "Grand Challenges," NSF is assigned the
task of "initiating the exploration of pricing mechanisms for
network service and network applications and structured transition
to commercial service." NSF should actively promote and provide
explicit guidance for this transition. To the extent that this requires
research into legal or policy issues, or into techniques for
performing accounting functions, NSF should support such
research. In particular, it is most desirable from our point of view
that the interim NREN be used in part for the delivery of
commercial information and other services.
The number of commercial companies already involved in the
operation of IP-based computer communication networks --
including US Sprint, AT&T, MCI, PSI, ANS CO+RE Services, and
Infonet -- along with developments such as the emergence of CIX,
Inc. (the commercial Internet exchange) indicate that commercial
vendors are already alert to the possibilities that the NREN program
offers. The emergence of a larger market for network connectivity
and services and the entry of new providers will, in the long run,
lead to lower prices for all consumers as the marginal cost of
delivering services diminishes and economies of scale come into
play. Moreover, this expanded market will attract additional
investment by commercial companies since it offers greater
potential returns on investment. Additional investment will lead,
in turn, to more rapid technology development. Coupled with the
explosive growth in business use of IP-based networks, the NREN
program in the public sector can help to drive the
commercialization of wide-area network technology much as the
NSFNET program has.
10. NSF should explore the feasibility of connecting midlevel
networks using a FIX or CIX model as an alternative to a traditional
backbone. Direct inter-regional links may also be desirable when
such direct links reduce costs and/or improve reliability.
NSF should explore the feasibility of linking midlevel networks
using a FIX or CIX interconnect model as an alternative to a
traditional backbone. Furthermore, NSF should seriously entertain
proposals based on direct inter-regional links where such links can
reduce costs and/or increase the end-to-end reliability and
redundancy of the IINREN.
11. NSF should support the development of software tools for end-
user applications and network management and operations.
NSF should issue one or more solicitations for the development
and deployment of tools for network management and operations,
end-user applications, routing protocols, etc. Practical solutions to
existing problems should be emphasized. For example, we believe
that the lack of useful tools for information retrieval and display is
one of the biggest impediments to the productive use of the
network and has impaired the credibility of the NREN in the eyes of
the target user populations.
NSF should consider the issuance of several separate solicitations
for for the development of software tools to ensure that this area is
given adequate attention. Operators of NSF-supported networks
should work closely with the awardees to ensure the rapid testing
and deployment of new software tools.
NSF should continue to emphasize open architectures and
standards in these solicitations. Its early decision to specify TCP/IP
as the standard networking protocol for the NSFNET was a
profoundly effective incentive for the extension of networking
services. Where standards are not adequately understood or
developed, NSF should support programs to test, evaluate and
improve them.
12. NSF should issue a new solicitation aimed at midlevel and
campus providers, with award criteria based on policy goals such as
improving the ease of use of the network and leveraging private
and non-Federal public funds.
Finally, we recommend that NSF, working with the user
community and the providers, define and implement clear criteria
for the award of additional funding to midlevel and campus
networks (as distinct from the top level) and issue a new solicitation
in this area. In the early stages of the deployment of NSFNET, this
funding was appropriately focused on "connectivity." The new
criteria should be designed to further specific programmatic and
policy goals such as the extension of network services to new or
underserved communities (for ubiquity), the improvement of
network operations procedures and tools (for reliability), the
enhancement of existing services through development activities,
upgrading of existing connections to "have-not" institutions,
leveraging of state, local, and private funds (to maximize the impact
of Federal investment), training and support for end-users (in
cooperation with national and local programs), etc.
Notes
1 Grand Challenges: High Performance Computing and
Communications. The FY 1992 U.S. Research and Development
Program. A Report by the Committee on Physical, Mathematical, and
Engineering Sciences, Federal Coordinating Council for Science,
Engineering, and Technology, Office of Science and Technology Policy.
2 This recommendation evolved from a discussion of possible
"certification" of network service providers. The goal of the
certification process was the establishment and enforcement of
minimum performance requirements across the network. We felt
that implementing the requirements via NSF's award criteria and
existing review process would be more direct and less bureaucratic.
