Our draft left with US Congress Office of Technology Assessment on February 28, 1992

Gordon Cook, Editor Publisher COOK Report on Internet

Introduction

If private gain, is not to outstrip the public good, we must prioritize whom the National Research and Education Network (NREN) is to serve. Business or education? All education or just an elite? And within those cam puses that have the network: all disciplines or just the "techies?" Making informed decisions will be impossible without identifying the stakeholders and understanding the development and interplay of their interests. In late November of 1991 time to do this appeared to have been purchased by the decision of the National Science Foundation to rebid the cooperative agreement for the operation of the NSFnet Backbone. However, given the amount of acrimony displayed by the stakeholders since the rebid announcement, the matter may yet be in doubt.

Some suggest that thanks to the rebid decision, the NREN may become an important step on the way toward a national information infrastructure. Others point to the bickering that broke out between ANS and its competitors within two weeks of decision to rebid. They complain that the NSF seems to have become too biased towards the interests of ANS. This study summarizes the complex events leading to this important decision - one that unfortunately seems to have produced only a fresh policy muddle. It also identifies key policy issues that need to be addressed during the next five years and suggests some possible solutions.

The NREN is a planned nation wide computer data network that is also expected to have voice and video capability, By the end of the decade the NREN may become, on a national level, the most powerful tool ever created for finding, manipulating and disseminating information. Its existence is unknown and unimagined by most Americans. Yet, to the extent that information is power, the network's existence will shape their lives. How it does this depends on whom the NREN will serve. The choices are stark: all Americans on the basis of universally available and affordable telephone service, or comparatively few - those who at our wealthiest universities and richest corporations can afford the access costs of a system creat ed to serve elites? In tracing the development and evolution of the NREN, this short study focuses on the divergent policy goals of service to an elite through technology transfer and rapid commercialization and privatization, and service to research and education through wide availability. As the National Science Foundation (NSF) has allowed the network to slide perhaps prematurely towards privatization, the conflicts inherent in these increasingly divergent policy goals have complicated the development of the network to the point where hearings or legal action may yet be required to unravel the conflicting interests involved.

The NREN is a collection of significant, confused and sometimes conflicting interests. It encapsulates:

A search for a means of justifying the addition of new technologies to the network.

The definition of a high-tech "nirvana" initially focused on the perceived needs of a supercomputing elite.

However, by the time the legislation creating the network became law, supercomputers were increasingly common and for the most part did not require the high speeds of the network to be used effectively.

A top down imposition of new technologies on users who are not adequately coping with the old technologies.

A performance gap between the speeds of the backbone and the speeds available on most campuses. An awareness is just beginning that billions of dollars investment in campus networks will be needed to bring even 45 megabit per-second uses into the hands of students and faculty. An infusion of federal money into the hands of segments of the telecommunications and computer industry to use for furthering their own strategic agenda.

The network does represent a tool of immense power and immense potential value to every American. The complexities behind its development have however been over simplified and occasionally misrepresented.

Some of the questions that are unanswered are:

The extent to which the network will serve to further technology transfer for the development of new communications industries versus the extent to which it serves as a tool that enables improved research and education.

Whether it will emphasize speed at the expense of ease of use?

Whether it will be implemented as a public or a private good?

The NREN was first proposed in a November 1987 report to the Congress by the Federal Coordinating Committee for Science Engineering and Technology (FCCSET), a entity within the ex ecutive branch Office of Science and Technology Policy (OSTP). It was offered as a means of increasing the speed of the nation's academic computer networks, known collectively as the Internet, from 56,000 bits per-seco nd to over a billion bits per-second (gigabit) by the mid 1990s.

The NREN Vision

Over the next several years, in a series of reports, the NREN was presented as making possible a startling variety of "information age" applications. These ranged from making it possible for researchers at sites remo tely located from supercomputers to take full advantage of these powerful machines, to travel through "electronic" libraries, and to engage in video conferencing as a means of normal communication.

Carried to an extreme, the NREN was presented as a "vision" of a network that would become a means of "virtual transport" for its users through video, sound and data within a decade. To achieve this, some of the addi tional functions that supporters suggested would be built into existing networks were:

Transfer of electronic mail and data files in multi- media formats, including high resolution graphics, full motion video, and sound.

Multi-media computer conferencing capabilities. Development and use of utilities that rely on resources distributed across the network. For example creation of a single digital library that could automatically draw on geographically dispersed collections.

Facilities that would enable scholars to build, flexibly and collaboratively, data bases containing comprehensive collections of knowledge on chosen subjects.

Knowledge management systems that would provide standard, consistent and intuitive interfaces to network services and resources.

Intelligent programs called "knowbots" (knowledge robots) that could be sent out by users to travel through the network in search of specified kinds of information.

While some of these services could require for a single user, network speeds of close to Broadband ISDN (B-ISDN, or 155 megabits per-second), none would require gigabit speeds. Still, providing this broad range of ser vices across a national network that could be called on to serve tens of millions of users would be very costly. Therefore, some said that the only reasonable way to achieve these capabilities would be to enable users to choose and pay for only those tran smission services and speeds that met their needs.

One problem with this scenario is that, while a large number of academic and research users of the network might desire such capabilities, the tight budgets at our colleges and universities would permit very few to pay for them. Furthermore, because the current network has been established by computer specialists operating on shoestring budgets, its entire structure is now based on fixed transmission speeds and single levels of service. The culture of network use has become firmly rooted in fixed cost c harges to an institution for a fixed amount of network bandwidth (speed). Actual use of the network by individuals is unmetered and uncharged for. Installing individual usage-based charging capability would require a major and expensive effort and would meet with considerable resistance from present users.

An additional, more fundamental problem exists at the level of infrastructure. Federal money brings the network to the campus doorstep and not to the professor's computer. While it is possible that the NSF may be able to help the mid-level networks incr ease their speed, taking advantage of only some of the technologies of the NREN vision will require the replacement of most campus local area networks with FIDDI LANs capable of 10 0 megabit per second speeds. In December 1991, the Chronicle of Higher Education estimated that the cost of necessary campus upgrades to be between ten and 100 billion dollars - an amount of money that is simply unavailable to higher education in the nea r future.

Meanwhile, by early 1992, among industry executives, there was a growing awareness that the High Performance Computing and Communications legislation that had just passed was, in the words of Apple CEO John Sculley, "focusing on the narrow problems of res earch and engineering" and ignoring investment "in an infrastructure that can dramatically change people's lives." The Computer Systems Policy Project, a lobbying organization formed in 1991, began to suggest that money from the HPCC legislation be allo cated in such a way that it produce a network less oriented towards big science and having "more meaning to industry and individual Americans." This would be done by assuring broad access to a network that could accommodate a wide range of applications. W ith the HPCC legislation passed, this group was beginning to acknowledge the difference between the NREN vision and the reality of the current network.

The Reality of the Current Network

While few would deny the value of the current network, its use is much more mundane than the applications emphasized by proponents of the NREN vision. Nearly three quarters of the traffic that travels over the network expected by 1996 to become the NREN is attributable to : 1. electronic mail; 2. remote logins to distant computers, and 3. file transfers from or to data bases at remote file servers.

Almost all the remaining use of the network is devoted to the overhead functions needed to route network traffic. These conventional uses are likely to dominate the network for the foreseeable future.

While advocates of the NREN "vision" emphasize the sophisticated capabilities that would enhance the ability of the network to serve as a tool to facilitate the needs of the research and education community, they gene rally avoid discussion of the problems faced by users of the current network. Depending on their fields of specialization, somewhere between 5 and 20% of those who should be able to profit from the use of the network actually have made the considerable e ffort necessary to become users.

Network users currently encounter many problems that frustrate or discourage use of the network, A study by Charles McClure of Syracuse University, identifies such problems as having to cope with

complex procedures that network center managers believe to be easy and hence do not adequately explain to novices;

the difficulty of transferring materials between different networks - such as NSFnet and BITnet;

the existence of multiple editors. Files created with one editor may have to be cleaned up and reformatted by their recipients, if they happen to use a different editor;

a "savage" user interface having a large number of conflicting commands and procedures, all of which must be understood, if the network is to be used effectively;

unavailable training. A user must either be assisted by a more advanced colleague, or "gut it out" with inadequate manuals;

documentation that is either not available, or out of date, or too difficult to use. Online documentation, when it exists, is generally too difficult to find in order to be really useful.

Technological overkill becomes for some users a problem in its own right. McClure's study found that although a few users with requirements for moving vast amounts of data complained that network bandwidth was occasio nally restricted, for many users, the very rapid change in network capabilities became a barrier to use. Users would often state that more applications become counterproductive when they are "unable to use and understand the network technology and applica tions they currently have."

For many who are supposed to be able to benefit from the NREN vision, the current, much-more-limited network is unusable. Some suggest that network applications should concentrate, not on increased speed, but on a fe w common services, such as directories -- electronic white and yellow pages that list the electronic addresses of users and services -- user friendly interfaces and operating procedures, access to libraries and bibliographic services, documentation of ne twork operations, training in network use, and outreach to potential users who are not familiar with the network.

While the National Science Foundation does plan to open a new and much improved National Network Information Center sometime in 1993, this will be only the first of many steps needed in order to make the network available to significantly larger percentag es of the user population that it is ostensibly designed to serve. Some critics complain that network promoters, in a misguided emphasis on speed and technology "sizzle," are ignoring the real benefits that the current network can make available not only to higher education but to the K-12 community.

The Two "Drivers" of the NREN

The emphasis on "sizzle" is occurring, at least in part, because the story of the network is marked by dual conflicting policies. First is a policy effort to bring improvements to research and education. But there is also , increasingly in control, a po licy designed to enable the creation of a multi-billion-dollar-per-year commercial high-speed computer data networking industry having very little to do with the needs of the research and education community. Part One of this study tells that story.

While the NREN has been shaped by these two very different policy "drivers," it has also been marked by two implementation efforts occurring in parallel with each other without coordination and integration. At the sa me time that Congress was attempting to legislate the NREN's creation, the NSF, with backing from the President's Office of Science and Technology Policy, has begun its own administrative implementation. To summarize: from its inception to the present, t he NREN has been hampered by a confusion of direction and control that could prevent it from ever becoming the national resource that Congress appears to have intended.

The NREN began in 1987 as a mission to enhance the ability of government supported research and academic networking to support high-end computational science - in other words: supercomputing. In 1989, as a strategy for gaining broader political support, plans for it to serve all of higher education were announced. By 1991, although the legislative language remained vague, many advocates had extended the mandate from high school through kindergarten. Also vague was the question of who would have ultimate control over the network. The early legislation pledged privatization after 1996. The law as passed, while vague, gives the impression that privatization is an accomplished fact.

During this same period the National Science Foundation was already building the NREN along the lines of a three-stage plan outlined in 1989 by the Federal Research Internet Coordinating Committee. The first stage was a T-1 (1.5 megabits per second) backbone. The second was a T-3 (45 megabits per second) backbone scheduled for 1992 and the third, it was pledged, would be multi-gigabit speed.

In the spring of 1990 the NSF planned for the changeover of eight of 13 backbone nodes to T-3 speeds. Three new nodes and the conversion of all 16 nodes to T- 3 speeds were budgeted from FY 1991 funds. Stage two of the FRICC implementation plan would be complete by the end of 1991 - a year ahead of schedule. The network called for by the High Performance Computing and Communications (HPCC) legisl ation was being implemented by executive action. Until the last days of 1991 the network's implementation and plans for its rapid privatization were being carried out very independently of the intent expressed by the legislation which called for privatiza tion of the network as soon as practicable at the end of the five year period of the HPCC program.

The Emergence of ANS

On September 17, 1990, the NSF, MERIT, IBM and MCI announced the formation of a 501 (c) (3) entity called Advanced Network and Services (ANS). With former IBM Vice President Al We is as President and CEO, and ten million dollars from its corporate sponsors, ANS stated its commitment to use the NSFnet to build a gigabit network infrastructure for the benefit of American research and education. I BM and MCI had been "joint study appointees" to MERIT. MERIT was the Michigan Education and Research Infrastructure Triad that ran a state network in Michigan and held the "cooperative agreement" with the NSF to build and manage the T-1 backbone until November 1, 1992. MERIT contracted with ANS for operation of the T-1 backbone and the building of the new T-3 backbone. ANS obliged by contracting back to MERIT and the day-to-day oper ation of the network.

Under the new arrangement ANS was allowed to set up and run two virtual (independent) networks over the same physical equipment. The first (NSFnet) would be for those organization s that received government subsidies to connect. The second (ANSnet) would be for those organizations that purchased direct attachments from ANS. Such organizations could use their attachment for commercial purposes such as selling products and services t o the research and education community, or intra-company transactions having nothing to do with the academic community - actions that were barred from the government sponsored backbone. ANS consequently became the thir d commercial provider of internet services, joining Uunet Technologies of Falls Church, VA and Performance Systems International of Reston, VA.

There was, however, one significant difference between ANS and its two competitors. ANS had Fortune 50 corporate sponsors and, with a ten million-dollar-a-year contribution from the NSF for the upgrade to T-3 backbone speeds, ran the backbone that connected all 32 mid-level networks providing network attachments for more than 1000 educational, governmental and commercial institutions. Its compe titors ran small commercial backbones attached by gateways to the NSFnet backbone, connected primarily commercial clients, and received no Federal subsidies.

By the last days of 1991 some had begun to wonder about the "deal" that the NSF was getting in the new T-3 backbone. The new structure was embedded in 12 core nodes within MCI's national network with switches co-located at MCI "pops" or network "points of presence." Each of the 16 backbone nodes, referred to as end nodes, were located on university campuses . End nodes were linked to core nodes by a single T-3 connection. Consequently if a link between an end node and a core node failed the mid-level network or in several cases networks connected to the end node would be cut off from the rest of the nation . Within the MCI backbone core network, eleven sites had only two data paths, while one had three. The new topology was essentially a star with a ring at the center. Not what network designers would call highly redu ndant. In the event of failures alternate pathways would either be few or non existent. Furthermore the new T-3 network was continually plagued by crashes stemming both from hardware and software problems. Critics complained that the old T-1 network wa s not saturated. They noted its mesh topology with every node except one having three paths to every other node offered far more redundancy than the new failure prone backbone for which the NSF was paying 330% more. As the T-1 backbone had to be kept open and continued to carry two thirds of the network's traffic, complaints about the NSF's arbitrarily privatizing the T-3 backbone grew louder.

In addition to running the backbone, in 1991 ANS also began to sell network attachments. This occurred much to the discomfort of both its smaller commercial competitors and many o f the 32 mid-level networks, which were also free to sell attachments - having been encouraged by the NSF to do so in order to become financially self-sustaining. In December 1990 ANS gained the North Carolina state network (NCMC) as its first customer. NCMC had been a customer of SURAnet, the largest of the mid-level networks. Over the next six months ANS continued to make its presence felt by hiring, as vice presidents, directors of three mid-level networks and the chairman of the Internet Engineering Task Force (IETF) - the Internet's standards body. In June of 1991 it established a for profit subsidiary ANS CO+RE (Commercial + Research). Its by-laws stated that profits from CO+RE would be used by ANS solely to b uild the core of the national backbone devoted to the research and education community.

What was not immediately clear about ANS' strategy was that it was seeking to attach as many of the Fortune 1000 as possible to that backbone and that its ability to do this would depend on its being able to plow the necessary resources back into the backbone. In July 1991, Telecommunications published an interview with Alan Weis, the ANS CEO. Here Weis made clear ANS' intention to market the Internet to corporate users. Weis emp hasized that ANS would offer corporations running private T-1 networks the opportunity to out-source such demands to ANS, which would offer turnkey packages to increase connectivity of corporate local area networks. Nowhere was ANS' original research and education charter mentioned.

Confusion of Markets

In the meantime law makers were laboring under an enormous confusion about the need for and cost of high-speed data networks. The May 1991 Report on S272, the Senate version of the HPCC legislation states that "the Federal funding called for in the legisl ation will guarantee a market for commercial high speed networking services, thus stimulating private sector investment in multi-gigabit networking." Two sentences later we read that the private sector is "reluctant to make the multi-billion dollar invest ments needed to build a national multi-gigabit network in part because the technology has not yet been demonstrated and the market has not been proven." Part One below shows that any implication that a "market for commercial high speed networking service s" is in need of Federal funds to guarantee it is false.

Furthermore in its use of the term multi-gigabit networking without defining whether aggregate or clear channel is meant, the report is confusing. Aggregate gigabit speed will be attained by multiplexing hundreds and perhaps thousands of slower data tran smissions together over a single strand of fiber. The technology exists to do this now and the expense of doing it is not huge. In fact it is done routinely. Telephone switching and transmission on major inter city r outes currently aggregates at gigabit speeds thousands of 64,000 bit per second "voice" channels, that can also function as cariers for fax and high speed data transmissions

On the other hand, clear channel gigabit speed is the subject of the NSF and DARPA funded gigabit testbeds. It is understood as giving a single network user a bandwidth of 622 megabits per second or more. The technolo gy to do this does not now exist. If it can be acquired, it will be enormously expensive. David Farber, one of the principal investigators on the AURORA Gigabit Testbed, estimated in the summer of 1991 that an NREN offering such capability would cost a billion dollars a year to operate. In March of 1991 DARPA's Ira Richer at a public meeting of the Federal Networking Council, stated that a single clear-channel gigabit line from coast-to-coast would cost 31 million dollars per year. More-over what one would do with such capacity is uncertain. Distributed supercomputing where multiple supercomputers linked across a wide area network exchange data in "real time," is a major object of investigation in the gigabit test beds. Unfortunately, linking supercomputers in such a fashion is a worthwhile endeavor that can be accomplished much more cost-effectively by means of local area networks.

In 1991 the General Accounting Office undertook for the Congress two small studies, the results of which focused the mistaken emphasis on high-speed uses which had come to characterize the rhetoric surrounding discussions of the NREN. The first "Industry Uses of Supercomputers and High-Speed Networks" reported that none of the corporate users of supercomputers could envision an individual need for network transmission speed in excess of the 45 megabits (T-3) available commercially today. The second report "High-Speed Computer Networks in the US, Europe and Japan," pointed out that European networks are in general slower than the old T-1 NSFnet and have no current plans to migrate to T-3 speeds. The fastest speed available in Japan was T-1. While it is true that NTT's plan for fiber to the home in Japan by 2015 exceeds American plans, speeds involved are not expected to exceed the 155 megabits per second of B-ISDN. Moreover they depend on NTT generating sufficient operating revenue which, in turn, is dependent on its being able to deliver cable television to the home.

What is powering the development of the NREN? The answer is to be found in the needs of business communications. While the current market for wide area networking in the academic and research community is about $80 mi llion per year, the market for Fortune 1000 TCP/IP wide area interconnection is several orders of magnitude larger. In fact, it is so large as to render meaningless the market creation assertions of the Senate Report o n S272.

Part One

The Culmination of the Data Revolution

A decade ago when IBM marketed its first personal computer, critical data in Fortune 1000 companies resided on IBM mainframes or, to a lesser extent, on minicomputers - smaller machines where Digital Equipment Corp. (DEC), had the largest market share. D ata could be shared among these large computers by means of proprietary protocols: Systems Network Architecture (SNA) in the case of IBM and DECnet< /a> in the case of DEC. Computing was centralized and focused on large and rather expensive machines. With its proprietary protocols linking their mainframes and minis together, IBM and DEC each had strangle holds over their respective markets.

Between 1981 and 1991 an explosion of personal computers and local area networks (LANs) designed to link these computers together eliminated this strangle hold and changed corporate data communications forever. Mainfr ames and minis lost their monopolies on corporate data to these distributed networks of PCs. According to an AT&T survey, by 1993 70% of all businesses will have LANs. Twenty million terminals and PCs currently attached to LANs will grow by then to 50 million. At that point the total number of LANs will be somewhere between five and ten million. Most will need to be interconnected in order to complete the migration of data from the mainframe to the desktop and allow the corporation to both distrib ute completely and, when needed, interconnect its information. Furthermore, as Electronic Data Interchange (EDI - electronic ordering and invoicing via computer network) grows, customer and vendor LANs will also have to be able to interconnect on demand.

But data growth does not stop there. In the past year or two, as CAD CAM and medical use of networks has expanded, high resolution images of photographic quality have become frequent network travelers. At three hundred to one thousand times the size of a n average piece of electronic mail, such images are major users of network bandwidth. In corporations like Boeing, such applications are driving local area network upgrades from 10 megabit per-second Ethernet speeds to 100 megabit Fiber Data Distributed Interface (FDDI) speeds. The Eastern Research Group, estimates that spending for image transmission execeded $500 million in 1990 and is growing at 22% annually.

Extrapolation of these trends leads one to the conclusion noted in the GAO study that T-3 or 45 megabit per second network will be needed by some corporations. Aggregate the traffic of enough such corporations across trunks forming national backbones and you will also soon reach combined gigabit speeds.

These trends are pervasive and are changing the very foundation of telecommunications. In 1985 network traffic was 80% voice and 20% data. In 1990 the gap had narrowed to 55% voice and 45% data. By 1995 projections call for 69% data and 31% voice. Wha t's more in 1995 those data networks will be able to carry voice and data multiplexed together with ease. Faced with these trends, Local Exchange Carriers (LECs), the telephone companies that provide services to homes and local businesses, are gradually realizing that they must learn to become network data carriers or else see their market shares shrink and customer's rates rise. Furthermore, in facing this challenge, they realize that they are faced with significant regulatory constraints that competitors providing private alternatives can overlook.

Rise of a Commercial TCP/IP Market

Asked in 1991 to rank their most important service problem, network managers of Fortune 1000 companies predominantly identified their need to interconnect the LANs of their corporation's scattered offices. Another surv ey revealed that 45% of all 1992 capital expenditures would be for LANs and LAN interconnection equipment such as bridges and routers. Telephone equipment would account for only 20%. For the first time, half of 1992 operating budgets will be spent on data communications.

In view of these trends it should be no surprise that in early 1991 three separate reports by Forrester Research Inc., Frost and Sullivan, Inc., and International Data Corp. (IDC), found that the corporate market for connecting LANs will triple between 1990 and 1995. A Forrester report stated that $560 million in TCP/IP products and services were sold to the commercial market in 1990. It projected the 1992 figures at $1.1 billion. 1991 sales to all markets reached 1.2 billion.

In looking just at the market for bridges and routers as LAN interconnection products, the following estimates were given:

1990 1995

Frost & Sullivan $607 million $1.8 billion

IDC $588 million $936 million

Forrester (routers only) $156 million $565 million

Eric Buck, a securities analyst with Donaldson Lufkin & Jenrette Inc., remarked that the router market was growing so strongly in May of 1991 that demand exceeded supply and there were not enough companies availabl e to serve it.

The Triumph of the TCP/IP Protocol

Protocols are to network transmission what programming languages are to computer software. In view of the earlier emphasis by IBM and DEC on proprietary protocols, it is ironic that what made all this dramatic growth possible was the public domain TCP/IP protocol. TCP/IP had been developed in the government run ARPAnet during the 1970s as a protocol designed to allow computers from different manufacturers to communicate with each o ther across wide area networks. By the early 1980s it had become a standard for "internetworking," and had given academic local area networks the connectivity that their corporate counterparts were beginning to dream about. By 1990 70% of the Fortune 50 0 had private telecommunications networks. Eighty-five percent of the Fortune 500 used TCP/IP in some portion of their networking operations. Of the remaining 15%, one third used TCP/IP corporate wide, one third used IBM's SNA and one third were holding out for OSI, the new European developed international standard. TCP/IP converts included such companies as Bristol Meyers Squibb, K-Mart, and Wal-Mart.

Although TCP/IP came to dominate the market, its dominance had been expected to be only temporary. A 1989 article in Telecommunications stated: "the single most important and widespread vendor independent suite of comm unications protocols in commercial use today is that commonly referred to as TCP/IP." All indications are that within a few years, the single most important and widespread vendor independent suite of protocols in comm ercial use will be those defined by the International Standards Organization (ISO), referred to as Open Systems Interconnection (OSI)." Since then has OSI developed more slowly th an anticipated and TCP/IP, lacking any viable competitor, has become an international standard.

On June 19, 1991 IBM's Vice President and General Manager for Networking Systems, Ellen Hancock, stated that "TCP/IP has become a standard in its own right and [is] not simply an interim step to OSI." Acknowledging that many customers were adopting TCP/IP rather than OSI and that OSI would not intercept TCP/IP's growth path, she stated that IBM had "agreed to invest more than originally planned in TCP/IP, while continuing [its] OSI and SNA investments." Part of this change was also driven by the success of UNIX workstations using IBM's RISC/6000 technology. These workstations demanded TCP/IP as a means of communication between them and were begin ning to create an environment where SNA speaking mainframes could become isolated if their controllers were not promptly adapted to TCP/IP. Not surprisingly IBM announced that this integration would be available in the third quarter of 1992.

Two weeks earlier Digital Equipment Corp. announced that it had added TCP/IP support to its new generation of network products which was to be OSI based. The new product would be n amed Advantage Networks and would support DECnet, OSI and TCP/IP. While DEC envisioned a multi-protocol environment with OSI as the best choice, its customers showed no signs of mig rating away from TCP/IP which leads OSI in worldwide sales $1.2 to .55 billion in 1991. These changes mirrored what had been happening in the marketplace during the preceding year with the development of multi-protocol ro uters.

Multi-protocol Routers

Beginning in 1990 companies like Cisco, Proteon, 3Com, BBN Communications, UltraNetwork Technologies, and Wellfleet, which had developed routers for the connectionless data traffic of the Internet, began to add support for SNA, DECnet, Ethernet, token ring, FDDI, and other protocols. These products were welcome developments in the business community where network managers had been operating redundant physical networks each based on different protocols.

For the first time separate networks could be collapsed into common routes with traffic tailored to flow over reconfigurable links as network conditions demanded. The changes saved corporations money by enabling them to use (momentarily at least) less bandwidth and fewer wide area links. Network flexibility increased and downtime decreased as the average time to activate alternative network links dropped from over a minute to five seconds.

In September 1991 Wellfleet announced a new gigabit per second router offering four 256 megabit-per-second data paths designed to capitalize on these needs. The router offered redundancy in power supply, backbone, and interface logic. Most significant of all, failed components could be replaced without taking the router off line and having to interrupt network traffic. Users saw the router as the first that could accept multiple in puts from 100 megabit per-second FIDDI LANs and 45 megabit per-second T-3 wide area links.

Meanwhile, in August, IBM, lagging, as it often had in the past, well behind the demands of the market, announced plans for a family of multiprotocol routers. Early four port and eight port versions of the routers sch eduled for a December 1991 announcement would support Ethernet and token ring LANs and TCP/IP. They would route SNA via Data Link Switching that would greatly reduce SNA overhead. In the last week of January 1992 the routers were announced for a June 1992 release. Called 6611 Network Processors, they would support DECnet, NetBIOS, TCP/IP, and Novell's Internetwork Packet Exchange (IPX). FIDDI LANs would not be supported. Two higher level protocols developed in the academic Internet - Open Shortest Path First routing, and the Simple Net work Management Protocol would also be supported.

The network press reported that while users welcomed IBM's entry into the internetworking marketplace, the move may have come too late. As a director of network planning and design at Pennsylvania Blue Shield observed: "few users are willing to wait a ye ar for IBM like they did in the old days." However, an executive from Travelers Insurance said that while non-IBM shops have been investing in routers for some time, many major IBM users are just starting to examine t heir options. If IBM delivers its product by the promised spring 92 release date, it may not be too far behind the curve since about one third of the Fortune 1000 have yet to acquire routing technology.

The NSFnet - NREN and IBM's Strategic Direction

In the 1990s and beyond data networks and computing will merge in a way that will endanger the continued existence of any company - including IBM - that does not learn to compete in the networking arena. Al Weis, the ANS< /a> CEO, dramatized the importance to IBM of its participation in the building of the NSFnet backbone, in November of 1990 when he stated that, without its participation with MCI a nd MERIT in the construction and management of the NSFnet T-1 backbone, IBM wouldn't have learned how to incorporate TCP/IP into the networking of its mainframes. Failure to do th is would have isolated IBM's mainframes and could have endangered their continued survival.

Weis and Gordon Bell had been in 1987 among the earliest supporters of the NREN. With IBM having been selected to participate in the building of the NSFnet T-1 backbone, Gordon Bell authored an article in the February 1, 1988 issue of the IEEE Spectrum in which on page 57 he called for the government to select a single vendor to build the NREN and charge users for its services. When in Septem ber of 1991 ANS as operator of the T-3 NSFnet backbone released its plan for charging for access on November 1 of 1992, some began to see a degree of prophetic vision in Bell's thr ee and one-half year-old statement.

If IBM were going to become a significant player in the networking arena, it would need a network switch. Therefore, it was not surprising when Weis pointed out that Dr. Allan Baratz of the IBM T. J. Watson Research Center was in charge of the development of a gigabit switch that would play the major role in ANS' plans to contribute to the NREN.

IBM, long known for its interest in proprietary protocols, was beginning to embrace TCP/IP as an open standard at level 3 and 4 of the seven layer network stack. Level 2 the transpo rt layer could use several options on which TCP/IP could ride. The telephony community through CCITT was developing an international standard known as Asynchronous Transfer Mode (ATM - fixed length packets) to use over SON ET at speeds up to multiple gigabits in fiber optic transmission. IBM's switch, named PARIS, would use Packetized Transfer Mode (PTM - variable length packets) over SONET. IBM's design philosophy for a "private in tegrated voice/video/data network" was articulated in an article entitled "PARIS: An Approach to Integrated High-Speed Private Networks" received on May 4, 1988 by the International Journal of Digital and Analogue Cabled Systems and published on page 77-8 5 of volume one. PARIS was also desired to interact with a proprietary LAN known as MetaRing. In May of 1991 PARIS became a product and was moved from the laboratory into IBM's Educational and Multi-media Products Di vision.

The stakes with PARIS appear to be quite large. If the PARIS switch were to become extremely successful in the provisioning of high-speed networks, it could hinder the ability of the RBOCs to offer their own planned hi gh speed data service known as Switched Multi-Megabit Digital Service (SMDS) - a service dependent on ATM capable switching. A vice-president of Bell Atlantic said the following: w e want at least to be able to sell SMDS services where the possibility of reasonable profit exists. If we are locked out and can sell only "dumb" bit pipes (Level 1 service), it wil l be like selling hundred pound sacks of potatoes at ten cents profit per-sack. Gaining enough revenue to modernize the network upon which these services depend would depend on not being locked out of 90 to 95% of the profits. A manager at another RBOC i nvolved with IBM in the development of PARIS as part of the AURORA gigabit testbed agreed that the dominance of such a switch could have a serious economic impact on the Public Switched Telephone Network (PSTN). He als o acknowledged that by the summer of 1991 IBM had agreed to add ATM capability to PARIS. However, he was very sceptical that IBM was committed enough to the goal to ensure that the result would be successful. By late 1991 IBM was talking about an ATM "i nterface" for PARIS. The switch would remain a PTM based device and gain the capability to encapsulate ATM cells inside the much larger PTM packets. What this meant in terms of the issues of importance to the RBOCs and LEC s was unclear.

In early 1991 IBM announced a three year study of its PARIS technology with Bell South Services, Inc. a Bell South subsidiary and in October of 1991 it signed an agreement with Rogers Cable T.V. Ltd., a Toronto based subsidiary of Rogers Communications to use PARIS to better define the technologies needed to support interactive multi-media applications.

The time table for the addition of ATM as an interface to PARIS is not clear. The best indication came in an IBM announcement from Geneva Switzerland in October 1991. Further ATM development of PARIS now renamed PLAnet and MetaRing renamed Orbit would tak e place in La Guade, France with commercial release anticipated in 1993 or 1994.

IBM's long term strategic viability depends on its ability to move quickly and adroitly to stay at the leading edge of high-speed computer networking. In its partnership with MCI and MERIT it has enjoyed a center-ring seat in the development of the NSFnet, the world's largest testbed for the development of TCP/IP internetworking. As plans for NREN matured during 1991, it presumably continued to think that through its ANS subsidiary it would inherit the backbone for the Interim Interagency NREN. It is likely that the November 26 199 1 NSF announcement of the backbone rebid (where the NSF said that it would award a new backbone contract in 1993 to at least two different service providers) came as a major shock. Still there is nothing that would prevent ANS from being named as one of the two service providers for the new backbone. By January of 1992 Cisco has announced its intention to team with ANS in bidding on the new solicitation.

The NSFnet - NREN and MCI's Strategic Direction

With its participation in the NSFnet backbone MCI has gained a cooperative and very forgiving testbed in which to develop the first nationwide TCP/ IP LAN interconnect T-3 backbone where single application bandwidth can vary in size all the way up to and including 45 megabits per-second. Because academic and research users gain access to the net normally from fees paid by their institutions and don't have to pay themselves for individual use, they really have no choice but to be more forgiving of network glitches than their corporate co unterparts would be. A bandwidth-on-demand nationwide beta test group of several hundred thousand users is nothing to scoff at. Just ask those who get network trouble reports and have seen how buggy the T-3 backbone has been since its first links were i nstalled a year ago.

The NSFnet environment fits well with MCI's strategy of "bandwidth-on-demand - the ability for users to instantaneously call up and tear down bandwidth in increments of 64 kilobits per-second." "Our strategy," says Donald Heath, MCI's Vice President of Data Networking, "is to put up a platform . . . [that lets]customers do frame-relay today, SMDS tomorrow and then [move] on to asynchronous trans fer mode and broadband ISDN." MCI is well on its way. As of December 1991, it is the only carrier to offer switched (on-demand) T-3 service. The rate is $127 per-minute prime time.

According to Computerworld's annual premier 100 rating which ranked MCI as its number one telecommunications company, MCI spends as much on the development of software systems as it does on communications. These systems are giving MCI enormous flexibility in tailoring its services on a world wide basis and in pushing intelligence down into the network, something that should be useful if the NREN ever began to bill on a usage basis.

In the late spring of 1991 MCI announced two services that, while dependent in part on what it had learned in the NSFnet cooperative agreement, were independent of its direct role in the Internet. One was a Virtual Pr ivate Data Network service using TCP/IP. Corporations could come to MCI for LAN interconnection service without being directly connected to the academic Internet. The other service was christened Infolan and offered by Infonet, a multi-national company owned 35% by MCI and 65% by 15 European and Asian PTTs.

Infolan offers TCP/IP internetworking for corporate LANs located in the United States and 115 foreign countries. Companies contracting for the service will have Cisco routers installed on their premises. These transmit over leased lines to the nearest entry points in the Infonet global network. Infonet is also a major sup plier of EDI services. Infolan will enable Infonet to offer EDI to a wider customer base. Infonet also manages its customers' connections 24 hours a day using Cisco's Net-Central Station software which based on the I nternet developed SNMP protocol. In managing its customers connections, unlike most LAN to WAN services, Infonet is offering out-sourcing to global companies with widely scattered relatively low volume data networking needs.

Out-sourcing

As the dominance of LANs changes the technical landscape of corporate communications, major corporations are beginning to hire specialists to provide for their telecommunications requirements. This practice is known a s out-sourcing. Among large corporations, most out-sourcing is focused on data network management because of management's difficulty in finding enough personnel skilled in local area networks and in network management.

The size of the movement to out-sourcing was brought home in September 1991 when General Dynamics Corp. signed a ten year contract valued in excess of 3 billion dollars with Computer Sciences Corp. for the management of its data processing and network ope rations. By placing the responsibility for computer and network modernization on Computer Sciences, the move was expected to save General Dynamics large sums of money over the life of the contract. Computer Sciences meanwhile having taken over ownership of General dynamics computing facilities was able to generate additional income by selling off the surplus capacity on its mainframes.

At the time of the CSC-General Dynamics deal, British Telecom underscored the growing importance of out-sourcing on a global scale by launching Syncordia, a new global out-sourcing company with headquarters in Atlanta, and customer service centers in Lond on, Paris and Tokyo. Targeting a global profile of 4,000 large corporations, Syncordia will use BT's existing Global Network Services which furnish data services to customers in 1,000 locations worldwide.

Private Versus Public Solutions

Out-sourcing is but one more example of a pervasive movement toward the provision of Private solutions that ride on top of the "dumb bit-pipes" of the Public Switched Telephone Netw ork. Value Added Networks (VANs) provided by Enhanced Service Providers (ESPs) is the regulatory terminology for what is happening across the nation and, indeed across the world, as computer data network traffic growin g at rates of 30% or more per-year meets and overwhelms a telephone and regulatory system ill prepared to comprehend what is happening to it. To the sorrow of "Aunt Millie" when she gets her phone bill five to ten years from now, the revolution in comput er data networks just described may be preparing to make a mockery of the division by the Modified Final Judgement (MFJ) of our phone service into 198 local areas or LATAs served by L ocal Exchange Carriers (LECs) and long distance routes that would be served by long distance companies known as Inter Exchange Carriers (IXCs).

For many reasons, including regulatory and rate of return restrictions on their lines of business, the LECs have so far failed to become viable providers of data network services. They all view Switched Multi-Megabit Digital Service (SMDS) using ATM cells over SONET carriers as their final major opportunity. SMDS will give both single personal computers and en tire local area networks the opportunity to plug into Metropolitan Area Networks with data transport rates in 64 kilo-bit per-second increments up to a total of 45 megabits per-second on demand. By 1995 the top SMDS s peed is expected to equal the 155 megabits per-second of broadband ISDN. Unfortunately for an LEC SMDS ends at a LATA boundary. Barred by the MFJ from carrying inter-LATA traffic, the LEC must rely on an IXC to offer the inter LATA service and buy local SMDS access from it.

Not surprisingly most IXCs are also getting ready to do SMDS. Here the only restrictions they face are access fees to be paid to the LEC. In prep aring for SMDS meanwhile, the LECs are also inhibited by the rate setting structures of no less than 50 state Public Utility Commissions (PUCs) which, in order to protect Aunt Millie's phone bill in the short run, have tended to price such business orient ed services rather high. In order to kill the LEC's offering of SMDS, all the unregulated IXCs and private bypass providers such as Metropolitan Fiber Systems have to do is set their rates 20 to 30 % lower than the LEC s.

Although the price of the current regulatory mismatch of good intentions with new technologies may take a few years to be truly brought home, what these trends point to is Private solutions that may continue to drive technological modernization out of the PSTN, ensuring in turn the continued growth of more Private solutions, hindering the ability of the PSTN to modernize and, as its user base becomes nibbled away by new wireless as well as fiber technologies, eventually weakening it to the extent that further modernization will have to be paid for by sharp rate increases for private subscribers, eventually endangering universal access to the network.

The stories of Metropolitan Fiber Systems and the Enterprise Integration Network offer examples of how these trends both interlink with the current development of the NREN and make for strange bed fellows as the speed of technology change overwhelms the capacity of our political and regulatory system. The likely result threatens to be the development of a two tier communications system - a feature rich, corporate financed, private system and an expensive but impoverished system for those unfortunate enough not to have access to the upper tier.

Metropolitan Fiber Systems and EINET

Metropolitan Fiber Systems (MFS) was founded in January 1988 as a fiber-optic telecommunications provider running private networks in the high density business districts of Chicago, Philadelphia and Baltimore. By the summer of 1991 MFS had expanded its networks to eleven cities. In August it announced a significant new venture. Using TCP/IP and FDDI local a rea network technology it would begin to offer 100 megabit per-second speed LAN interconnection service in each of its city networks. In effect it was offering - with some significant differences - the equivalent of < A HREF="glossary.html#29">SMDS Metropolitan Area Networks (MANs) at least a year before the LECs could offer them. The network speed offered was more than twice that of SM DS. However, in the form of by-pass of the PSTN commonly referred to as cream-skimming, the private networks offered by MFS would reach only the largest and potentially most profitable subsets of urban users. The LA N interconnect service would become available first via MFS' 849 fiber mile 30 building network in downtown Houston - a complex characterized by a heavy concentration of oil companies and medical centers. Within the next six to twelve months these service s will be extended to the remaining ten MFS networks. Late in 1991, anticipating FCC clearance, Royce Holland, the MFS CEO, said that he believed that his business could expand into a total of about 50 cities.

MFS' announcement had a missing link that was identified in conversations on the Internet mailing list known as Com-Priv during August. There a Houston area affiliate of MFS explained that it had designed its intercon nect service so that the ANSnet or a similar national backbone could be used to connect its eleven metropolitan area networks into a single TCP/IP based national network. ANSnet Vi ce President Guy Alms (formerly the Director of SESQUInet, the Houston based mid-level network) attended the MFS press conference announcing the new service. While MFS would be a very significant account for ANS, sig ning up with ANS was by no means the only option that Holland has. The ANS backbone is embedded in twelve core nodal system locations that make up the critical links of MCI's national network backbone. Nothing would prevent MCI from offering the service directly.

Finally at the end of the second week of November another substantial service provider for businesses emerged. The Austin Texas based Microelectronics and Computer Technology Corp. (MCC) announced that it would begin to spend one quarter of its annual bu dget of $55 million to oversee the development and operation of the Enterprise Integration Network (EINet), a nonprofit TCP/IP based network. According to Allan Schiffman, Chief Technology Officer at Enterprise Integr ation Technologies Corp. which is helping to design the network, EINet, architecturally similar to the Internet, would carry commercial traffic and offer better security features.

Four industrial consortiums with an emphasis on American industrial competitiveness pledged support for EINet stating that they would use it for internal communication and encourage its use among their members and associates. These four are: Sematech, als o located in Austin; the Iacocca Institute of Lehigh University in Bethlehem, PA; the National Center for Manufacturing Sciences and the Industrial Technology Institute both of Ann Arbor, Michigan. EINet, which will begin operation in March 1992, will off er a direct link with an X.500 directory being developed by DARPA on a pilot basis for the academic Internet. Companies will pay an annual fee to access EINet in addition to transmission charges according to the bandwidth of their connection. The fee will be according to the user corporation's annual gross revenues:

Revenues Annual Fee

Greater than $500 million $60,000

$50 to 500 million $30,000

Less than $50 million $10,000

Additional ripples from the EINet announcement were felt when during the second week of December plans for a Factory America Broadband Network were announced at the Agile Manufacturing Conference in Lake Beuna Vista, Florida. The net was promoted as a pa rt of the critical infrastructure needed "to ensure that U. S. manufacturers remain competitive 15 years from now." A six month long $500,000 study by the Iacocca Institute called for the establishment of the network possibly as a part of the EINet. Fin ally in January Paul Hurray of the University of South Carolina, announced a version of an industrial competitiveness network. In early 1992 business use of wide area networks appeared to be growing even more rapidly than academic use.

Many observers, in view of the size of the ANSnet backbone and its strong corporate orientation, expected the award to go to ANS. On January 8, 1992 in a major surprise, MCC annou nced the award of the backbone to Uunet Technologies of Falls Church, Virginia. The contract value on an annual basis was not announced. Further questions revealed that Uunet would be paid only a percentage of the revenues generated by each EINet member. If EINet is very successful, it success is likely to strengthen the CIX (see page 13 below) and help level the playing field. This outcome is one scenario that could lead to the CIX backbone being upgraded in such a way that it emerges as a true competitor for the NSFnet-ANSnet backbone. It would not be surprising to see a significant number of mid-levels switching to the CIX backbone. Some observers predict that ANS will eve ntually either have to join the CIX and give up the commercial charging plans discussed in Part Two of this study, or find itself completely isolated and largely without customers.

With the arrival of what are being called the Public Data Internets (ANS, PSI and Uunet) corporations may now use not only the academic Internet solely for enterprise networking, but also have available a host of purely private solutions. MFS, EINet and Infolan either have opened or are about to open private, Fortune 1000 focused efforts not directly connected to the academic Internet. Backers of the privatization of the NSFnet have expressed the hope that enough corporations would join the network to enable large economies of scale and profits that can be used to cross subsidize academic uses of the highe r speed more expensive network. A major question, in view of these new arrivals in the marketplace, is how many corporations will opt for this solution instead of a private TCP/IP, non Internet solution where they are likely to find more options to ensure the security of proprietary data, greater reliability and more services tailored specifically for their business needs.

Part Two

What Is an Academic Network Doing in a World Like This?

The Evolution of ANS

Meanwhile in the academic Internet, events during 1991 moved at a very rapid pace as ANS installed, what for all practical purposes was a privatized T-3 backbone, a portion of which it provided to the NSF to connect the 32 mid-level networks at 16 sites around the country, and the remainder of which it worked assiduously at marketing to corporate users. During 1991, ANS moved somewhat imperiousl y into a culture that it did not understand. This culture was based on a tradition of openness and cooperation that dated back more than twenty years to the beginning of the ARPAnet. By emphasizing voluntarism and sharing, the Internet had developed as a very fertile testbed for new technologies and had created a whole series of networking standards affecting all layers of the protocol stack - standards based on what worked best for the community not on what one stoo d to profit by, as former Lotus Development Corp. founder Mitch Kapor observed.

A tradition of the Internet community was its willingness to discuss issues affecting it by means of open electronic mail lists where statements made or questions asked would be reflected in writing, dated and time stamped to hundreds and sometimes thousa nds of other network users. Performance Systems International (PSI) had started one such mail list in the spring of 1990. Called Com-Priv, it was devoted to a discussion of the commercialization and privatization of the Internet. ANS read the messages sent to the list but refused to participate otherwise in the conversation.

In a parallel effort, from November 1 1990 through April 15, 1991, the author, in his role as NREN Assessment Director at the United States Congress, Office of Technology Assessment, moderated a restricted on-line con ference that included over 100 policy makers involved in the creation of the NREN or in the issues that its creation raised. ANS received the traffic from this discussion, but until some questions about its role were raised in a way that it felt impossible not to answer, it did not join in the discussion. During the first two weeks of January, it provided some useful input before falling silent again. In the meantime, by the time an ANS technician began a regular an d quite skillful series of replies to the Com-Priv list on behalf of his employer in June of 1991, the network community had decided that it neither trusted nor liked ANS. For the community this was a serious matter because the general expectation was th at ANS would inherit complete control of the backbone and hence the network when the cooperative agreement between MERIT which had subcontracted backbone operation to it and the NS F expired on November 1 of 1992.

In the meantime ANS attended countless academic and professional association meetings where it described its role as one of building an American research and education networking infrastructure second to none in the w orld. While some wondered whether investments in the network designed to make it easier to use should have priority over increases in speed, CEO Alan Weis emphasized that ANS would build faster network infrastructure indicating a desire to move to 622 me gabit per-second backbone speeds by late 1992 or early 1993.

To Privatize or Not to Privatize? The Great Backbone Rebid Announcement

On November 26, 1991 the NSF announced its decision taken the preceding Friday with the blessing of the National Science Board to rebid the backbone. The decision was clearly reluctantly arrived at on the part of the NSF in the face of strong opposition by FARNet members and Educom's National Telecommunications Task Force to the NSF's desire to complete the privatization of the network by giv ing control of the backbone to MERIT and ANS at the end of the five year cooperative agreement on November 1, 1992. The NSF's change of face left insufficient time for planning a transition to a new cooperative agreement by the time the old one expired. Therefore the NSF stated its intent to extend the current cooperative agreement with MERIT at the same rate of $10,000,000 per year for up to another 18 months. ANS would be lef t in its position of uncontested power until early 1994 when the cutover to new backbone operations could be made.

The agreement, which will be a complicated one to implement, called for at least two network service providers and for the contracting of network routing authority to a neutral third party. (MERIT currently has routi ng authority.) If routing authority were left in the hands of either service provider, it could be used to the detriment of the other provider.

The NSF mystified some observers by stating that it expected the costs for provision of the new higher speed backbone (presumably 622 megabits per second) to decline dramatically in comparison to the old one. In year one it anticipated paying $6 million, in year two $5 and in year three $4 million. While it did not explain why it anticipated such a bargain, some observers noted that some of the current problems with ANS appeared t o come from the fact that MERIT and its partners had apparently made a below cost bid for the network and appeared to be able to extract significant concessions down the line for having done so. They wondered if the NSF were looking for a repeat scenario. They also wondered whether the figures cited by ANS would effectively freeze out a successful bid by AT&T or any of the RBOCs, which as regulated carriers are severely constr icted in how low they can bid on such contracts. Was the desire to get the best monetary deal by the NSF tantamount to ensuring that NREN would be deployed as a private network and was this in the best interests of na tional telecommunications policy? These were questions without ready answers.

Among FARnet's November 1 1991 recommendations to the NSF on the provision of backbone services was a plea that the mid-levels be given a choice of backbone providers. The promise made by the NSF to provide multiple providers presumably could mean dividing the network into Eastern and Western halves. However under this scenario, mid-level networks would, by virtue of geography, be likely to have a backbone provider thrust upon th em. Giving them a real choice could mean that the U.S. Government would find itself in the awkward situation of providing two backbones. A third possibility exists. For $3,200,000 per year (or $100,000 per network) the 32 mid-levels could connect to bot h the East and West coast CIX interchanges. This would provide aggregate T-3 bandwidth to the mid-levels. The NSF could decide whether to subs idize clear channel T-3 access to ANSnet for the nation's top 25 research universities. The NSF's decision is expected to become known when the new backbone solicitation is released on May 4, 1992.

Uneven Foundation for the Commercial Market

While the NSF decision to rebid the backbone may eventually help to level the playing field, at the beginning of 1992, the only reasonable conclusion is that it is not level. Two small commercial national providers (PS I and Uunet) and a mid-level (CERFnet) run by a commercial company (General Atomics) have formed an alliance called the Commercial Internet Exchange (CIX). They have interconnected their T-1, 1.5 megabit per-second ba ckbones forming a commercial backbone that connects a small but growing fraction of the commercial and NSFnet community. During the first quarter of 1992 the CIX gained three new members: BARRnet (a California based mid-level), U.S. Sprint, and Unipalm of Cambridge U.K. While the CIX remained a T-1 network with only one interconnection point, it nevertheless connected more than 3,000 commercial firms - including the top 20 computer companies in the US.

Against the CIX in early 1992 stands ANS, another small company. But ANS provides service to the NSFnet back bone carrying 28 times the traffic of the CIX backbone and connecting all 32 mid-level networks covering the entire United States. Thinking in terms of broad connectivity, it is as though ANS were positioned to compete to become a national airline wit h routes that included Boston, New York, Washington, Atlanta, St Louis, Chicago, New Orleans, Pittsburgh, Philadelphia, Detroit, Denver, Salt Lake City, Seattle, San Francisco, and Los Angeles while the CIX opened only from Boston, New York, Washington, C hicago, Denver, San Francisco and Los Angeles.

Figure 1 on the next page shows the current T-3 ANSnet backbone. The heavy black lines are the "core" nodes and are a part of the MCI national backbone. The names in ellipses are the 16 backbone "end" nodes of the NSFnet. RTP stands for the Research Triangle Park where ANS connects CONCERT, the North Carolina State Net work. West Lafeyette identifies ANS' connection of Purdue University and Blacksburg its connection of the Virginia Polytechnic Institute. What the map does not make clear is that two of these three ellipses represent single campuses while the remaining 16 represent 32 mid-level networks with over 1000 attached institutions.

Those who say the playing field is not level point out that ANS has behind it two major corporate sponsors. One of them, IBM, appears to be investing in the network as a major strategy to revive sagging parts of its o wn business. The other, MCI, seems to be using the network as a testbed in order to develop and market, before AT&T and the RBOCs are able to, data services that are projected to become dominant within the telephon e industry. PSI and Uunet, the competitors of ANS, are very roughly the same size. However, they do not, by virtue of a Federally-funded cooperative agreement, run a backbone

connecting over 1,000 institutions

costing their parent companies, by some estimates, as much as fifty to sixty million dollars, and

bringing them an annual operational cash flow of ten million dollars.

Possession of the major network backbone is strategically important for an additional reason. The network is based on the operation of trunks, known as bit pipes, of various capacities at fixed cost regardless of the traffic that goes through them. Traffic has never before been metered. Whether or not it should be is a matter of controversy within the research and education (R&E) networking community. Regardless of whether metering traffic would be a wise thing to do, from a technical point of view, the potential for installing metering in the next year or so is small. Therefore economic viability in the network community is a function of how many customers a provider can attach at fixed bandwidth.

The size and direct reach of the backbone connecting customers to each other, if all other costs were equal, would likely be a determining factor in the decisions of customers in search of a network provider. Since ma ny network costs tend to remain fixed no matter how many customers a provider acquires, economies of scale become significant. If ANS can connect a sizable number of new organizations to the network, the foundation p rovided by its larger infrastructure will make it a difficult force to compete against.

Under these circumstances PSI and Uunet and the mid-levels must hope to compete with ANS primarily in terms of cost. Since, courtesy of the Federal government, ANS has a larger immediate cash flow, and, courtesy of i ts corporate sponsors, a core backbone network embedded within MCI's network, not to mention 32 mid-level networks with over 1000 customers dependent on it for connectivity, its competitors feel the contest is uneven. From the point of view of the mid-levels matters are even worse since all three of their commercial competitors (ANS, PSI, and Uunet) can act both as long distance and local carriers. In other words they can provide both local and national connectivity by linking a customer directly to a backbone instead of to a mid-level.

Competing Backbones: CIX Versus ANSnet

"Appropriate use" of the government funded NSFnet backbone has been a driving factor behind the changes taking place. Commercial traffic is banned. While well over 100 commercial companies are on the network, they may not use the network to do business with universities or each other. Appropriate use is defined as communication with other network members about research and education. With most of the network's commercial members having products or services of interest to the networks non commercial members, this creates many grey areas. For example, the manufacturer of a new workstation may send a new product announcement to a mailing list discussing engineering advancements in workstations (informing) but not to individuals (selling). To cite another example, the publisher of a major legal database, makes its wares available via the network free of charge to law students while law firms may not use the network to retrieve mate rial even for a fee. Although commercial use restrictions are enforced primarily by an honor system (the NSF does not inspect packets), there are so many grey areas that all parties are eager to be rid of the restrictions.

Unfortunately the dual (on the one hand government subsidized while, on the other hand, commercially available) backbone run by ANS creates another problem of imbalance among the n etwork players. With commercial traffic only legitimate over certain parts of the network, commercial customers of the CIX, or of the mid-levels, wishing to do business with academic institutions reachable only via th e NSFnet backbone, would need to route their traffic via the ANSnet backbone - something for which ANS has established charging procedures. In support of its charging procedures, ANS announced that commercial custome rs signing up at T-1 rates will pay contributions of almost $5,000 per year to a "national infrastructure pool" designed to support research and education use of the network. Such a connection has a total annual cost of about $75,000 per year.

Analysis of the likely annual corporate contribution to this infrastructure pool reveals that it would be unlikely to play a major role in cross subsidizing academic use. The total number of corporations able to afford such costs would be unlikely to exce ed one thousand. Of these about 100 to 200 are already connected to the network. Many are unlikely to connect because of security concerns. (These may opt for TCP/IP based services isolated from the national network. ) Given that PSI and Uunet charge considerably less for the same connection, and that Union Carbide and Abbott Laboratories, ANS' most recent customers purchased 56 kbs connections, estimating 100 customers per year for ANS at T-1 speeds would seem exceedingly generous. This would generate less than $500,000 per year for the "national infrastructure pool." Considering that the NSF's current contribution to the network on behalf of its research and education users i s about $18 million per year, one would not expect $500,000 to go very far.

As the network moved towards the termination of the cooperative agreement with Merit in November of 1992 and what participants assumed would have been full operation by ANS, two c amps developed. On the one hand ANS was telling each mid-level network that it could sign agreements that will permit commercial service over ANS net for an annual fee of approximately $80,000 per year at T-1 speed and $325,000 a year at T-3 speed. (The amount that each mid-level will have to pay ANS for its connection to the backbone depends on how many commercial customers it has and at what band width they connect. A mid-level with 15 commercial customers could be forced to pay about twice the basic $80,000 connect fee.)

On the other hand, at the beginning of August the CIX began to invite TCP/IP service providers to join it for a one-time fee of $10,000 at T-1 speeds plus the annual cost of a circuit to either the east or west coast CIX connection point. Those who joined would pledge to forgo settlement fees for traffic passing over each other's backbones. This condition pu t the approach of the CIX into stark contrast with that of ANS which was anticipating significant income from what would, in effect, be mid-level settlement fees.

As if this were not complex enough, it created what participants assumed would be only a hypothetical problem thanks to the decision to rebid the backbone. Most had assumed that the AN S charging procedures announced in September would not go into effect because of the decision to rebid, With the end of the year dispute over whether the mid-levels could be forced to sign connectivity agreements, the ANS pricing agreements no longer appear to be moot. Whether the NSF has the power to prevent ANS from enforcing them is not clear. Certainly the problem is worth examining to understand the potential chaos into which the network may be heading.

If ANS is able to insist on imposing these commercial use fees, the mid-levels could attempt to escape them by joining the CIX. Unless the CIX paid an annual gateway fee to ANS, the mid-levels who did join the CIX would not be able to reach those who remained with ANS . Such a fee would be directly proportional to the number of commercial users within the CIX member networks. Some observers believed that it could have been as much as a million dollars per year.

The CIX was saying to ANS join us. But, for ANS, because of the greater cost of running its higher speed backbone and its commercial ambition s, the prospect of joining the CIX was unthinkable. At the same time the CIX let it be known that it would not accept ANS' stiff payment provisions for the right to send traffic over its backbone. By September 1991 two rather hostile camps had developed. "Flaming" on the Com-Priv List grew so intense that PSI's Bill Schrader and ANS' Al Weis felt compelled to call a truce saying that ANS and the CIX had agreed to negotiations. By January of 1992, the acrimony had reappeared and it was clear to all that unless someone took forceful action, the CIX and ANS would likely refuse to interconnect if the NSF let ANS' pricing agreements stand, and approved its continued privatization of the network.

At Risk: The Research and Education Users

Privatization would put education and research users at risk by raising costs to pay for higher bandwidth and the reliability of a production quality network. Disruptions in service could occur due to the balkanization of the network by a possible dispute between CIX and ANS and/or disruptions of the mid-levels. An increased possibility of the imposition of met ered service could mean further disruption of a network culture based on the flexibility to experiment. Privatization could make the improvements of the NREN vision less likely to happen because the commercial sec tor would pay for these only if there is money to be made from providing them.

Before these conditions became so starkly apparent, the NSF had seemed willing to gamble that the commercial growth in the network would be so great that costs will be driven down by economies of scale. Furthermore it assumed that profits will be so sizab le that they can be used to cross subsidize the academic users. Lower costs and higher profits would buy the NREN vision and ensure that users will be able to afford to stay on the network. Unfortunately, while the hope is alluring and the price of T-1connections may be declining, the overall cost of network use is unlikely to decline as new applications encourage higher speed and hence more expensive connections. Even more important, the creation of commercial TCP/IP internets like EINet, not to mention choices expected to be available from the public switched network with the next 12 to 24 months, means that the commercial customers that the academic Internet was hoping to attract are likely to have multiple means of wide area networking and are much less likely to be attracted to a privatized NSFnet.

The Policy Dynamic in 1991

The NSF in attempting to reach a final decision about the fate of the backbone faced a very complicated decision. Examination of the context for the decision shows that the vision for a National Research and Education Network emerged out of a successful two-decade-old Federal policy of supporting the development of computer networking technology. This policy has provided a neutral environment where the network development could be driven by two separate but loosely sy nergistic policies. First the network was treated as an enabling technology to link academic communities to each other and to remote computers. In doing this it became a tool that could be used to facilitate research and education. Second the network fac ilitated technology transfer to the private sector by serving as a testbed for the development of new communications protocols and the associated computers and programs necessary to exploit them. With the network paid for largely by Federal funds and under the control, first of DARPA and then the National Science Foundation, the entity that provided the money was able to balance the goals of technology transfer and enabling technology for improving research and educat ion should they ever threaten to conflict. With the expectation of rapid network privatization in 1991, ANS' needs began to dominate the decision making process, making this balancing act increasingly difficult. The process that is now unfolding appears to be placing academic and research users of the network in an environment structured for commercial needs and driven by commercial costs.

Despite the NSF's decision to rebid the backbone, its bias towards ANS appears to remain so strong that the building of a high speed network designed for academic and research use may yet take place as an overlay of a high speed network crafted to suit the needs of the Fortune 1000. Network policy makers should be using the time before the award of the new backbone contract in April 1993 to ascertain whether the two needs can be o verlaid on each other in any cost effective manner or whether research and education networking needs are sufficiently different as to make their provision by a publicly funded networking entity desirable.

The Dichotomy of Mixed Commercial and Academic Use

Corporate uses of a commercial NREN would be significantly different than those of academia. The difference is highlighted by the list of the twelve most popular ISDN applications compiled by the North American ISDN User's Forum in October 1991.

1. Videoconferencing 2. Telecommuting 3. Telephone and workstation integration 4. Multi-point screen sharing 5. Customer service call handling 6. ISDN access in geographically remote locations 7. Image communications 8. Automatic number identification 9. Multi-document image storage and retrieval 10. Multi media services 11. Multiple ISDN telephones on a single BRI loop 12. ISDN interface to cellular, mobile radio & satellite systems

Of the twelve applications only the seventh, ninth and tenth are of significant importance to the academic and research community. Nowhere to be seen are such NREN dreams as multi-media electronic mail, and access to vast shared library databases by intelligent user agents known as knowbots. In an environment of low communications costs and vast bandwidth, the academic community has shown some interest in video conferencing. How ever, such an interest would fall far behind the ones just mentioned. Finally we find no mention of supercomputing, the original raison d'etre for the NREN.

One of the problems of overlapping an academic network onto a network designed for and driven by corporate use is that the two are very different worlds and use information in very different ways. The corporate world's data is generally proprietary. Whil e data does need to be shared within a corporation in controlled ways, management must guard, at all costs, against leakage of proprietary data to the outside. To cite but one example, security concerns would now prevent a corporate supercomputer from be ing placed on a publicly available network. It is true that companies increasingly will need to communicate across corporate boundaries in order to use computer networks to buy and sell goods and services from each other. Known as Electronic Data Interc hange (EDI), such services will be tightly controlled by the companies that use them and will be only marginally useful to the academic community.

In contrast to the corporate world, the academic and research community will use the computer network to share information. Within this community, the network benefits research and education in direct proportion to the extent that it enables sharing of i nformation across institutional boundaries. Network resources created to protect and guard proprietary information are not high on the list of academic needs. By the same token, network wide white and yellow pages, and universally accessible databases ar e not high on the list of corporate needs.

Of course, given enough Federal subsidies, the academic network planners can buy or have created by the corporate network proprietors, anything that the academic world does want. However, the consequence of overlapping the networks with their different n eeds would be likely to mean that the cost of overlaying academic tools on a network driven by the needs of corporate users may be far greater than the cost of building these tools into a network designed for academic users from the ground up. Moreover g iven our chronic budget deficits, the high costs are likely to mean that the academic community may have to do without the tools it desires.

Under such conditions, it seems likely that a significant percentage of the NSF's NREN appropriations could yet be spent in support of the higher costs of a faster commercialized network. In other words the NSF could find that providing the same level of support to network users in pursuit of science and education policy goals would have cost significantly more, because of the changes brought to the network by privatization.

The Mid-levels' Role as the "Glue" for the NREN in the Face of the Changing Marketplace

The NSFnet is made up of 32 mid-level networks connected to each other by the national backbone. These networks provide connectivity to over a thousand educational institutions, go vernment agencies and corporations. If ANS' commercial pricing agreements are allowed to stand, the mid-levels cost of access to the backbone would increase, as would their operating costs. The NSF could provide fres h subsidies to the mid-levels to offset such costs. This would be likely to anger PSI and Uunet which would see such subsidies as yet another form of subsidization for ANS. In the absence of such subsidies, the mid-levels' most likely significant source of new revenue to offset these costs (and indeed their best chance for financial viability) would come from connection new corporations to the network. Unfortunately they can no longer count on what, as recently as 1989, was a monopoly in providing thes e connections.

Their monopoly is gone because the mid-levels are no longer the only "glue" between the local campus networks and the backbone. The old tripartite division of the network in a backbone and campus networks connected by the mid-levels is no more. ANS, the backbone operator is now competing with the mid-levels in selling direct connections, as are PSI and Uunet, the smaller commercial competitors of ANS. With such competition it is unlikely that mid-level income from new commercial connections will increase significantly.

The mid-levels provide perhaps 90% of the connectivity for academic and research users of the Internet. The NSFnet backbone connects not only the 32 mid-levels but also, through th e Federal Internet Exchange (FIX), the NSI backbone with about a dozen NASA laboratories and the ESnet backbone with a similar number of DOE national laboratories. MILnet, as the open portion of the Defense Data Network is known, connects a larger number of military bases here and abroad.

If a scientist doing work for NASA, DOE or DARPA is not located at one of the respective labs or military bases, he or she will generally be found at one of the 1,000 institutions connected to the NSFnet and will obta in connectivity to other researchers and to sponsors at NASA, DOE, or DARPA through the NSFnet mid-level, the backbone, and the Federal Internet Exchange gateway at the appropriate site or sites on the other three networks. Therefore, even to the mission agencies, if the NSFnet is to become the NREN, and is to fulfill the goal of making the network available to scientists and educators, the cont inued existence of the mid-level networks, in the absence of some other means of affordable connection for their client institutions, is of critical importance.

However, current market forces are bringing the role of the mid-levels into question. Barring a government decision to subsidize massively the attachment of secondary schools, and two and four year colleges, the market for new network growth is predomina ntly commercial. The mid-levels are predominantly run by academics most of whom have no knowledge of the commercial market and technical demands therein for standards of service. If the mid-levels are ever to become independent from NSF subsidies, they m ust continue to be able to attach new commercial clients - something that throughout 1991 they had been able to do, on the whole, more cheaply than ANS and PSI. (Some say that Uunet's prices are generally quite low be cause they assume considerable customer expertise and provide relatively little customer assistance. But others assert that Uunet has now begun to offer a full range of services.)

One of the problems to be solved by any plan for an orderly transition to a completely commercial market will be the need to decide under what conditions the mid-levels are or should be sustainable. The mid-level's viability is currently made more difficu lt by ANS's aggressive pursuit of the connection of the same commercial firms. With the T-3 backbone in place as an overlay of the MCI national network backbone, ANS has considerable flexibility in attaching its commercial clients. Therefore it will be able to compete more effectively with the mid-levels on the basis of cost. The mid-levels' problem is further compounded by the fact that PSI an d Uunet are also selling connections to commercial clients.

What does this situation mean when viewed from the framework of the policy goal of making the network available as a tool for the use researchers, educators and students? If one or more mid-levels should collapse, how can one predict the impact on the us ers? If the policy goals that Congress chooses for user connection to the network include only the largest of our universities, the answer is that the impact on these users would probably not be serious. Such institutions would buy high-speed connections from ANS or another commercial provider. The next tier of universities would be assisted by the NSF in doing the same thing. However, at some point the NSF's ability to help would run out.

On the other hand if Congress desires to promote broad access to the network, the value of the mid-levels begins to become apparent. Aided by approximately 40 million dollars in NSF subsidies over the past five years, the mid-levels have installed their own infrastructure that allows them to bring low cost, low bandwidth service to small schools and to small, technically oriented, businesses. ANS does not offer network connection at less than 56kbs.

Unfortunately for the small clients of PSI and Uunet which do offer slow speed connection, all of their clients depend on connectivity through the ANS backbone to reach much of the total network. Consequently, the ability of PSI and Uunet to connect these institutions would depend on two things: First on the willingness of ANS, a competitor, to allow them use of its backbone to offer them access to the entire network; Second on t heir ability to connect them as cheaply as the mid-level had done - something likely only for those located very near to one of their backbone nodes.

Consequently, unless a mid-level were taken over and its operation preserved largely intact, its demise would very likely harm its low end users. What is not possible to predict is precisely where the line would be drawn between those who could afford to connect to a commercial provider and those who could not. A reasonable estimate would be that at least 40% of the more than 1000 institutions connected to the NSFnet would be at risk if their mid-level collapsed. U ntil the commercial market matures to a point where it can offer ubiquitous, low cost, low speed connections to the network, the conflict between the goal of technology transfer for commercial development and the goal of expanding the network as a tool fo r the improvement of research and education is likely to continue.

No Clear Financial Picture by Which to Project the Financial Strength of the Mid-levels

In looking at the evolution of the mid-levels in a privatized market, policy makers could hope to be better guided in their decisions on a case-by-case basis if they had an accurate picture of the finances of each mid-level. Readily available summary hist ories of what kinds of grants for what purposes had been made by the NSF to each of the mid-levels on an annual basis would make it easier to see which mid-levels were fiscally the healthiest and to draw conclusions as to why. With such a picture it woul d be easier to make decisions about how to apply continuing subsidies with maximum effectiveness. Unfortunately this data has never been separated from individual grants and aggregated by the NSF into a form that would make such conclusions easy to draw.

In meetings in January and May of 1991, NSF officials explained that their relationship with the mid-levels has been driven by proposals originating from the mid-levels. That is to say they do not assume to know what the communications needs are in the t erritory of any given mid-level and issue a request for proposals to fill them. Instead the mid-levels come to them with proposals for cooperative agreements to connect new groups of colleges or to build new and faster links between existing schools. (C olleges and universities wanting to connect to the network may apply directly to the NSF Connections Program or may negotiate their connectivity through a mid-level.)

The NSF has the proposals initiated by the mid-levels reviewed. If they are judged to be reasonable and money is available, they are funded. Records of what has been funded are kept by the NSF. How