12 March 1992 /internet/legislative.actions/hearing.12mar92/wolff.testimony Testimony of Dr. A. Nico Habermann and Dr. Stephen S. Wolff Committee on Science, Space and Technology Subcommittee on Science March 12, 1992 Part 2: Testimony of Dr. Stephen S. Wolff Mr. Chairman, thank you for this opportunity to appear today before this committee to discuss the NSFNET and related activities. There are three parts to my testimony. I will discuss first the current state of the NSFNET Backbone project, including its relationships to other networks that actually, or potentially connect to it, and also the management controls the NSF has in place with its awardee, Merit, Inc. Second, I shall report on the progress we have made in implementing the Project Development Plan for continuation and enhancement of NSFNET Backbone services which was approved by the National Science Board in November last. Finally, I shall briefly discuss the relationships between the NSFNET and NREN programs, including the interagency management structure now evolving for the NREN as an Administration program with a legislative authority. Current State, Other Networks, and Management Controls a. Current State The five year cooperative agreement between the Foundation and Merit, Inc. for management and operation of the NSFNET Backbone was signed in November, 1987, after a five month period of competitive announcement and merit review of proposals. Merit, and its partners IBM and MCI, put in place a 1 3-node, 1.5 mb/s (million-bits-per-second), or T1, network in a very short time. The new Backbone began to carry traffic in August, 1988. In that month, traffic doubled over the July figure for the original Backbone network that the new one supplanted. Since August, 1988, traffic on the Backbone has increased more than fifty-fold, from 200 million to 11 billion packets per month. This increase in traffic has been accommodated by hundreds of minor engineering improvements to the network and two major upgrades. The first upgrade increased the number of links in the network from 14 to 19. This increased the robustness of the Backbone by multiply connecting all 13 nodes, and it increased capacity as well. The second upgrade increased the number of Backbone nodes from 13 to 16 (the three new nodes were competitively selected), and raised the transmission speed from T1 to T3 (1.5 to 45 mb/s). All the engineering improvements and both major upgrades were clearly foreseen and discussed in Merit's original farsighted proposal to the NSF. Such are the economies of scale in telecommunications that the upgrades to accommodate a fifty-fold traffic increase have been achieved with only a doubling in cost to the Foundation - from the original $14 Million over five years to the present five-year project cost of $28 Million. The NSFNET Backbone is the linchpin of the overall NSFNET project, which includes establishment of and assistance to regional networks that deliver Backbone service to every state in the union. Other significant measures of the size and success of the NSFNET project include: More than 600 of the 3-to-4,000 two-year and four-year colleges and universities in the nation are interconnected, including all the schools in the top two categories of the Carnegie Foundation classification of major research universities. Several hundred high schools are also connected, but the exact number is difficult to determine since regional networks have widely leveraged NSF funds to connect the smaller institutions without NSF's direct involvement. Many industrial research organizations and commercial establishments that support the nation's scholarly enterprise are connected; indeed, the so-called ".COM" domain is the fastest growing segment of the network. The NSFNET Backbone is the default infrastructure for the nation's research and education community. It carries, for example, ten times the traffic of the Department of Energy's ESnet Backbone which interconnects many NSFNET client sites with national laboratories and other DoE facilities. By selecting a proven set of open communication protocols ("TCP/IP") and mandating their use in the NSFNET, the Foundation catalyzed an entire industry in which there are now upwards of a half dozen US manufacturers. US made packet switches and gateways dominate the world market, and a T1 packet switch can now be bought for well under $10,000. (By contrast, before NSFNET, the most widely used network packet switch operated at a speed of only 56,000 bits per second and was priced at $120,000. A further effect has been to substantially increase the connectedness of the scientific community as several other large networks, e.g., MFENET, the forerunner to ESnet, and European HEPNET, the European High Energy Physics network, have switched in recent years from their own proprietary communication protocols to those (TCP/IP) compatible with the NSFNET.) NSFNET's selection of TCP/IP has led to it becoming the most widely used set of open communication protocols in the world. Procedures for transporting these protocols over emerging telecommunications services, such as the Switched Multi-megabit Data Services (SMDS) and Frame Relay have recently advanced to Draft Standard status. Because of this, NSFNET and the Internet will be able to benefit from whatever economies may be available fromusing the new offerings of the telecommunications carriers. Scientists and educators on NSFNET can now collaborate over the network with their peers in 39 countries on 7 continents, and every month brings new requests for connection to the US network of which the NSFNET and its Backbone is the principal component. b. Other Networks Another measure of the success and influence of the NSFNET project has been the emergence and rapid growth of private sector offerors of TCP/IP network services. These include: UUNET Technologies, which indeed predated the NSFNET, but has grown rapidly in recent years; Performance Systems International (PSI), a spinoff from the NSF funded regional network NYSERNET; Advanced Networks and Systems (ANS), who provide NSFNET Backbone Services under contract to Merit; US Sprint; InfoNet, a multinational TCP/IP provider; and CERFnet, which functions as a regional network in Southern California. Several of these private providers have formed a cooperative for interchanging traffic known as the Commercial Internet Exchange, or CIX, of which Mitch Kapor is Chair. The NSFNET Backbone is limited to uses compatible with the NSF enabling legislation, as amended. There is an "NSFNET Backbone Services Acceptable Use Policy" (the "AUP", a copy of which is attached to this testimony) which was developed in consultation with an NSF Advisory Committee and the NSF General Counsel and expresses this limitation. The general principle is worth stating, "NSFNET Backbone services are provided to support open research and education in and among U.S. research and instructional institutions, plus research arms of for-profit firms when engaged in open scholarly communication and research" By contrast, the private providers, have no such limitations. Although much of the traffic on their networks need not conform to the AUP, it is NSF policy to allow the private providers to use NSFNET Backbone services to exchange AUP-conformant traffic between their customers and NSFNET clients. However, the NSFNET Backbone may NOT be used by the private providers as a "transit network" - i.e., to interconnect their fee paying customers. In this traffic sharing environment, ANS occupies an especially sensitive position since NSF indirectly, through Merit, is one of its customers. Accordingly, NSF has made special arrangements with Merit to monitor the quality of service afforded to NSFNET and to ensure that the traffic of ANS' private customers does not adversely impact NSFNET Backbone services. c. Management Controls The NSF participates with Merit, IBM, MCI, the State of Michigan, and (since its formation in 1990) ANS in three series of regular meetings which collectively form the primary means of oversight and control. There is a biweekly "Partner Conference Call" which functions at the tactical level, a monthly "Engineering Meeting" for technical desiderata, and a quarterly Executive Committee meeting which considers strategic issues. During the transition from the T1 Backbone to T3, the Executive Committee also scheduled weekly conference calls. As provided for in the Cooperative Agreement with Merit, NSF convened a blue ribbon review panel of academic and industry experts and conducted a two day long review of Merit's Backbone performance at the eighteen month anniversary. The panel rated Merit's performance "excellent". The Project Development Plan In November, 1991, the National Science Board (NSB) approved a plan for continuation and enhancement of NSFNET Backbone Services beyond the expiration of the current cooperative agreement with Merit in November, 1992. The NSB also approved an extension of the agreement for a period not to exceed eighteen months in order to allow new providers to be competitively selected and to provide for an orderly transition. A copy of the Plan is attached to this testimony. The Plan was developed after more than a year of external consultation. During this year of consulting the external community, NSF supported two workshops at the Kennedy School of Government at Harvard - one in March 1990 and the second in November, 1990. These workshops involved university networkers, economists, specialists in public policy (especially telecommunications policy), telecommunications carriers, and others. NSF's sister Federal agencies involved in the NREN were consulted at a meeting convened for this purpose in July, 1991, since the NSFNET Backbone is the most heavily used Backbone network among the several agency networks that are developing the NREN. The Foundation sponsored a workshop in August, 1991, by the Federation of American Research Networks (FARNET), a trade association that was inaugurated in 1987 to act as the voice of the regional networks, the "users" of Backbone services. The workshop was also attended by all the private providers of Backbone services, as well as telephone company representatives. In addition, the Networking & Communications Research & Infrastructure Division Advisory Committee was consulted at its meeting in November 1991. That Committee includes leading researchers in the communications and networking field, private network providers, and telephone company representatives. Moreover, NCRI staff participated at public meetings of the networking community, such as meetings of the Internet Engineering Task Force (sponsored by industry), Net '90 and Net '91 (sponsored by the academic and user community), and others. The Plan has a schedule that includes release of a draft Solicitation in February 1992, a three month period for public comment, followed by release of the final solicitation in May. Owing to unexpected delays in releasing a separate but related solicitation, and the technical complexity of the proposed new NSFNET Backbone architecture, it has not been possible to adhere to the original schedule. The other solicitation has been released, NSF's engineering experts have been consulted, and it now appears the draft solicitation will be ready at the end of March, so the schedule has slipped by about eight weeks. We believe there is still adequate time to accomplish the solicitation-review-award-transition process within the eighteen month extension authorized by the NSB. The technology permits a planned, gradual, and orderly transition of traffic from one provider's facilities to another's. The transition, now in progress, of moving traffic from the T1 Backbone to T3 provides practical experience for the future. The Plan provides for a degree of continuing competition among two or more TCP/IP service providers in furnishing NSFNET Backbone Services. There will however be no significant changes in the rules for access to NSFNET Backbone Services by commercial service providers. The Acceptable Use Policy, developed in consultation with the NCRI Division Advisory Committee and the NSF General Counsel represents, in the opinion of Counsel, the most liberal interpretation possible under the NSF enabling legislation, as amended. This current policy allows access to commercial services for the support of open scholarly research and education under the General AUP Principle stated above. NSF believes the next award will clarify the issues in free and open competition for the provision of Backbone services, and will conclude with at least two fully qualified and experienced providers of bulk services. It is likely, therefore, that NSFNET Backbone funds may - after the end of the next award (i.e., by FY 1996) - be distributed competitively to those organizations (currently the regional networks) who require Backbone services so that they may procure them competitively on the open market and free of Federal intervention. NSF had wished to employ this model at the expiration of the Merit award, but was advised at the FARNET workshop that the regional networks (the "users") were unprepared for that degree of operational complexity on their part. Moreover, sister Federal agencies felt in addition that such a procedure would, at the current state of technology, result in serious routing instability in the network, prejudicial to the accomplishment of their missions, since they depend heavily on the NSFNET to reach many of their grantees and contractors. NSF will continue working with the regional networks and the sister Federal agencies to overcome these obstacles. In a separate, but closely related activity, the NSF has just released a competitive solicitation for Network Information and Registration Services. These are services which have traditionally been provided for the worldwide Internet by Network Information Centers (NlCs) associated with the major US Backbone networks (i.e., ARPANET, NSFNET, ESnet, and the NASA Science Internet) as well as by Centers operated by NSF regional networks, by campus network organizations, and by the private TCP/IP network providers. The principal NIC, however, was for many years operated by SRI International under contract to the Defense Communications Agency (now the Defense Information Systems Agency, DISA). In a recent re-competition held by DISA, SRI lost the contract to another firm. DISA is funding the new contractor, GSI, to serve only the Defense Data Network; accordingly, NSF is funding GSI on a month-to-month basis for service to the rest of the Internet (including, of course, its largest component, the NSFNET) until NSF's recently released solicitation can result in a new Network Information Center. During the month-to-month funding, NSF is closely monitoring GSl's operation. It is interesting to note that the commercial users of the Internet, many of whom are clients of the private TCP/IP providers, form the largest single user class of GSl's services. Relation to NREN Finally, I would like to turn briefly to the relation of the NSFNET to the overall NREN program that is part of the HPCC Program described earlier by Dr. Habermann. The planning process for the HPCC Program is coordinated by the HPCCIT Subcommittee. This subcommittee meets regularly to coordinate agencies' HPCC programs through information exchange, common development of interagency initiatives, and review of individual agency HPCC proposals and budgets. This process provides for agency participation through agency proposal development and review, budget crosscut development and review, and interagency program coordination. Agency programs are reviewed against a set of evaluation criteria for merit, contribution, readiness, linkages to industry, and other factors. During 1990, in order to provide for broader and more inclusive coordination of research and education communities, the NSF, as part of its HPCCIT network task group activities, created the Federal Networking Council (FNC) and initiated the creation of an FNC Advisory Committee (FNCAC) as an NSF advisory committee. The FNC consists of representatives from Federal agencies that have requirements for operating and using networking facilities, mainly in support of research and education, and for advancing the evolution of the Federal portion of the Internet. Membership lists of the FNC and FNCAC are attached to this testimony. Achieving the goals of the NREN will require close coordination of the NSFNET, NASA Science Internet (NSI) and Energy Sciences Network (ESNet) programs to meet the expectations of scientists working on the Grand Challenge problems. At the same time, however, the NSFNET program will vigorously pursue wider NREN goals of developing the technologies that will enable access by libraries, use for lifelong education, and connection to health care systems, etc. The NSF will continue to involve the private sector to the greatest extent possible for meeting the goals of public policy in this arena in the most cost-effective and technically responsive way. NSF is participating with the other agencies in the FNC in the drafting of the NREN report required of the Office of Science and Technology Policy by the High Performance Computing Act of 1991 (P.L. 102-194.)