From
July 1991
is Sharon Fisher's investigation of international networking provisions:
STATE OF THE ART: Whither NREN?
Everyone agrees that a high-p
erformance network for research and education is vital if the U.S. is to remain competitive in the world economy. The question is, who should provide the network?
by Sharon Fisher
T0 those who prefer to limit the amount of government in their lives, one of the three great lies is "I'm from the government, and I'm here to help." But others believe that for certain large tasks, only an entity with the scope of the federal government can get things done.
These two philosophies are currently in conflict over the U.S's future networking needs, or, more precisely, how such needs should be satisfied. One camp proposes a federally funded National Research and Education Network (NREN), comparing it to the interstate highway system, while the other believes that sufficient commercial services already exist to provide such a network.
Ironically, both groups agree that this type of network should end up being provided commercially. The question, then, is whether such a network nee
ds a federally sponsored jump start.
History Behind NREN
The forerunner of NREN is the Internet, a network devoted to research and educational use, the development of which was originally funded by the Advanced Research Projects Agency of the Department of Defense.
The first step toward a more sophisticated national network was in the late 1980s, when the National Science Foundation funded a high-speed network backbone, NSFnet, linking first 12 and now 16 supercomputer centers around the U.S. (
see figure 1
). As a result, links between these centers were upgraded to Ti speeds (1.5 megabits per second), and a central-management network-operations center was established by the Merit organization in Ann Arbor, Michigan. Merit, with IBM and MCI, won a five-year contract in 1987 to manage the network and oversee upgrades.
The value of such a network is apparent in its staggering growth (
see figure 2
). For the past three years, the amount
of network traffic has grown at an annualized rate of 20 percent per month, according to Joel Maloff, former vice president of client services for Advanced Network and Services (ANS), which manages the network along with Merit. In March alone, the network transmitted over 7 billion packets of information.
At the same time, members of government and industry were becoming concerned over what they perceive to be a high-technology lag in the U.S., compared to other countries. In response, in 1989 the President's Office of Science and Technology Policy issued its Federal High-Performance Computing Program report. This report proposed four areas for R&D in technology that could bring the U.S. back to the forefront: high-performance computing systems, advanced software technology and algorithms, basic research and human resources, and a national research and education network.
NREN and Congress
It's this last area, NREN, that has attracted most of the attention. The report proposed a
three-step process whereby the federal government would fund a network, based on the Internet and NSFnet, that could eventually transmit data in the gigabit range (more than 1 billion bps), and that would extend not only into the traditional research, university, and government arenas, but even into selected elementary and secondary schools. Once the network was built, it would gradually be commercialized, or run by commercial organizations rather than the government.
Proponents note that the entire program is projected to cost $988 million in the first year, and a total of about $2 billion over a five-year period. In comparison, the total funding for science each year is some $70 billion. Or to put it in more readily understandable terms, the whole project would cost about the same as one B-2 bomber, according to a congressional source.
In response, bills funding the proposed project have been brought to the Senate Committee on Commerce, Science, and Transportation, spearheaded by U.S. Senator Al G
ore of Tennessee. In 1990, such a bill unanimously passed the Senate Commerce Committee and the full Senate, but it died in committee in the House of Representatives, mainly due to the lack of a strong champion. Gore resubmitted the bill, now known as S. 272, the High-Performance Computing Act of 1991, to the subcommittee on March 5, and it is slowly working its way through the process.
The Question of Control
But while there's little disagreement that such a network would be useful, there's some argument over whether the federal government should be paying for-and controlling-it. While the national data superhighway comparison is often made, the analogy breaks down when you consider that the infrastructure for such networks already exists in the facilities of organizations like AT&T, US Sprint, and MCI.
"It depends on your view of NREN," says Martin Schoffstall, vice president and chief technical officer for Performance Systems International (PSI), which provides a commercial ne
twork service and a portion of the existing NSFnet backbone. "If it's a funding device for the grade schools or the high schools or the local libraries or the community colleges to participate in the Internet, then it is needed. If it means building a big network, owned and controlled by the government or by some contractor, then it probably is not [needed]."
Many believe that the best way for the government to develop a network is to support existing services rather than to build another network. "The Internet, up to now, has been this wonderful socialwelfare state," says Geoff Goodfellow, president of Anterior Technology, in Menlo Park, California, which provides E-mail and news feeds. "Not to say that the Internet hasn't been a good thing," Goodfellow adds, "but we're now at that turning point when those commercial carriers can provide the same level of service that the government backbone can. Then it's time for the government, rather than funding the backbone, to put the money in the hands of the su
bscribers of the network and let them pick the network service provider they wish to go with, along the same lines that they would pick a long-distance carrier.
Commercialization of the Internet
A couple of years ago, the question of commercial use of the Internet would have seemed heretical. Traditionally, its use and access have been free. But as organizations have grown to depend on it more and more, they have not wanted to rely on its somewhat free-form management.
Consequently, a number of organizations are already in the business of providing access to the Internet-and even providing their own networks. Many of the supporters of these organizations believe that the best way to produce an efficient, high-performance network is through the time-honored American method of competition rather than the time-honored American method of federal subsidies.
The commercial companies providing Internet access include General Atomics, of San Diego, through its CERFnet; PSI of Reston
, Virginia; and UUNET Technologies of Falls Church, Virginia. It's worth noting that both CERFnet and PSInet were originally NSFnet regional networks, with CERFnet covering parts of southern California and with PSI starting out as NYSERnet, the middle Atlantic regional network. UUNET, for its part, has long been in the business of providing commercial access to the Internet.
These groups compete with ANS, which is headquartered in Elmsford, New York, and has offices in Ann Arbor, Michigan. ANS is a nonprofit corporation, funded originally by IBM and MCI, with the mission of supporting research and education networking in the U.S., says Maloff.
"Where ANS fits in is that, several years ago, the NSF ran a solicitation for the management of the NSFnet backbone to upgrade to Ti," says Maloff. This contract was won by a consortium of MCI, IBM, and Merit, the University of Michigan-based organization that also runs Michigan's statewide network.
But it was later determined that this type of relationsh
ip would not effectively bring the network into the 1990s, Maloff relates, so a new organization was required. "The first concern of industry was that if IBM had the ability to control and dominate it, that was not a good thing. That ruled out a straight partnership or stock corporation."
The result is the current not-for-profit corporation, funded by $5 million each from MCI and IBM. Both companies have a seat on the eight-person board of directors, although neither can ever have more than one. In addition, the original agreement between Merit and NSF for management of the existing NSFnet has been assigned to ANS as well (with the engineering portion then subcontracted back to Merit).
The result, too, is that portions of the NSFnet backbone have already been upgraded to 45-Mbps T3 speeds, Maloff says. The T3 infrastructure (now referred to as a
cloud
rather than a backbone, to indicate its redundant, nonlinear nature) is managed by ANS.
ANS, at this point, consists of the NSFnet bac
kbone and its own commercial service.
At present, organizations that use the NSFnet backbone-which includes ANS-are not supposed to use the network for commercial traffic, only for traffic in support of research and education.
In other words, although commercial organizations like Sun Microsystems or Apple Computer can be on the network, they are not allowed to use the network for commercial services. For example, a number of PC vendors use services such as BIX to provide technical support to their customers; on the NSFnet, this is not allowed.
These particular restrictions are in place because of commercial information and networking vendors' concern that a federally subsidized network could compete with them unfairly. "The concern with commercial use over the NSFnet is that commercial organizations will be using the facilities that are subsidized by the federal government," Maloff explains.
Maloff indicates that work is under way to loosen these restrictions, but that the final determina
tion rests with NSF. "If, in fact, there are ways to be able to fairly cost and allocate costs back to the appropriate place, there is less of a concern [about competition and commercial use]," he says. "We want to eliminate, if possible, the chilling effect that some of these policies have. Folks are worried about whether they are in compliance with the acceptable-use policies."
Policing the Internet
In practice, making acceptable-use determinations is very difficult to do when up to 45 Mbps can be transmitted, each routed a different way depending on destination and network traffic, and when neither service organizations nor users are interested in having the service organization check every data packet and determine whether it's acceptable traffic.
Consequently, there's a question even now about just how much of a restriction there really is on commercial use of the NSFnet backbone. "The use rules, for the most part, are illusory," says Goodfellow. "Enforcement is really a pro
blem unless you look inside the information itself, and no one's willing to do that." Maloff concurs, noting that ANS has not, to date, taken any action regarding any breaches of the acceptable-use policy.
In an attempt to eliminate the problem, several commercial services banded together this spring to set up an alternative backbone that would not use NSFnet links and consequently would not have to follow acceptable-use policies.
The Commercial Internet Exchange allows all users of AlterNet (UUNET's service), CERFnet, and PSInet to exchange Internet traffic directly at no additional cost. This would mean that CERFnet users could communicate with AlterNet or PSInet users without having to traverse the NSFnet backbone at all.
Moreover, the organizations said, the agreement means that they can continue to provide service even if the NSFnet backbone were to fail completely-a point of vulnerability, after the 1988 Internet Worm incident, of which many users are aware. The three commercial services
claim they provide nearly 100 percent of the commercial TCP/IP- and Open Systems Interconnection-based internetworking services in the U.S.
Another possibility is that NSFnet's purpose may be redefined, says Douglas Van Houweling, vice provost for information technology at the University of Michigan and chairman of the board of ANS. "It's entirely possible to imagine the net being targeted to serving the research and education community, including the library community, but not restricted in terms of the type of traffic it can carry in service to that community." In other words, the network would be able to support commercial services like Dialog as long as the customers were in the research and education community.
Content is a subset of the issue of who should control the network, says Van Houweling. "The current environment is one that works because of the cooperation among a very large group of organizations all over the world," he says. "We need to continue that highly decentralized, cooperativ
e model of organization delivering the network."
"Quite a few people have suggested that we need some agency or organization that is 'in charge,' "Van Houweling continues. "I believe that we do need an organization that draws together the involved service providers with the diverse user community to set overall objectives for the NREN. But that is very different than saying that group should be in charge."
Creating the Future
At this point, there are two major actions that will help to determine the future and form of NREN. The first is the success or failure of the NREN funding bills that are currently before the Congress. It's always difficult to predict when such actions might hit the floor. It is possible that this decision will have already been made by the time this article appears in print.
The second is a little further in the future. Merit's contract with the NSF for running the NSFnet backbone expires in October 1992. "What will happen after that will be determined
in part by what the NSF concludes is the follow-on activity to the current cooperative agreement," says Eric Aupperle, president of Merit Network, in Ann Arbor, Michigan.
To provide for an orderly transition, a preliminary decision may be made as early as the end of this year, Aupperle says. Such a decision might be a continuation of the current scheme, passing control to ANS, setting up a competitive bid between ANS and whatever other cornmercial organizations might feel qualified, or something else. "Among the possibilities are, rather than funding an organization like Merit, using NSF dollars that would be available to provide funding to the regionals to acquire backbone services from whomever," he explains.
Once all the issues have been settled, then the NREN-in whatever form it eventually takes-will indeed be able to become the broad-based network that it was originally intended to be. "We haven't really yet become a marketdriven community," says Maloff. "We have been technology-driven. At the
point where we become more concerned with the secondary-school teacher in the middle of North Dakota or the astronomer who no longer has to once a month fly into a city to pick up a set of tapes, then we'll be a real community with real needs. Until then, we are an esoteric kaffeeklatsch."
illustration_link (174 Kbytes)
Figure 1.
illustration_link (119 Kbytes)
Figure 2.
photo_link (66 Kbytes)
Sharon Fisher is a San Francisco-based freelance writer specializing in data communications. She can be reached on the Internet as
slf@well.sfca.us
, and on BIX as "sharonfisher."
5 Years Ago in BYTE
