From pozar@kumr.lns.com Sat Apr 24 07:05:18 1993 Received: from [140.174.7.1] by fido.wps.com (5.67/1.34) id AA00542; Sat, 24 Apr 93 07:00:47 -0700 Received: by kumr.lns.com (/\==/\ Smail3.1.25.1 #25.3) id ; Sat, 24 Apr 93 06:59 PDT Message-Id: From: pozar@kumr.lns.com (Tim Pozar) Subject: NSF report To: tomj@fido.wps.com Date: Sat, 24 Apr 1993 06:59:41 -0700 (PDT) Content-Type: text Content-Length: 177416 Status: OR |--------------------------------------------------------------------| NOTES ON THE TEXT: The following is the text of the report "Review of NSFNET" by the Office of Inspector General of the National Science Foundation, followed by the text of NSF's response to the OIG report. The format is somewhat different from the printed versions of the report and response, to acco--odate normal screen width and the limitations of the ASCII format (only one font, no underlining or boldface). When language in the printed report was italicized for emphasis, in this version that language has asterisks ("**") placed on either side. The page numbering is different than in the printed version, but the Table of Contents has been corrected to conform with this version. |--------------------------------------------------------------------| Office of Inspector General National Science Foundation Review of NSFNET 23 March 1993 1 |--------------------------------------------------------------------| Review of NSFNET I. Introduction . . . . . . . . . . . . . . . . . . . . . . 3 II. Background: What NSFNET Is . . . . . . . . . . . . . . 4 III. Progra--atic Issues A. The NSFNET Solicitation, Evaluation of the Proposals, and the Award . . . . . . . . . . . . . . . . 7 B. Expansion of NSFNET and Conversion to T3 1. Necessity . . . . . . . . . . . . . . . . . . 11 2. Amendment or Resolicitation? . . . . . . . . . 16 C. Spinning Off: ANS & CO+RE 1. CO+RE and the Infrastructure Pool . . . . . . 21 2. Co--ercial Use . . . . . . . . . . . . . . . . 25 3. Co--ercial Access . . . . . . . . . . . . . . 29 4. The Decisionmaking Process . . . . . . . . . . 31 5. Public Announcement . . . . . . . . . . . . . 35 D. Acceptable Use Policy . . . . . . . . . . . . . . . 37 IV. Administrative Issues A. Cooperative Agreement or Contract? . . . . . . . . 41 B. Conditions of the Award . . . . . . . . . . . . . . 42 C. Prior Approval . . . . . . . . . . . . . . . . . . 44 D. Documentation . . . . . . . . . . . . . . . . . . . 48 E. Compliance with OMB Circular A-110 . . . . . . . . 49 F. Accessibility . . . . . . . . . . . . . . . . . . . 51 G. NSF Funding for the Provider or the Users? . . . . 52 V. The Future of NSFNET A. The Proposed Solicitation . . . . . . . . . . . . . 53 B. Co--ents on the Proposed Solicitation . . . . . . . 56 C. Interagency Coordination of the NREN 1. The New NASA/DOE Network . . . . . . . . . . . 60 2. The High Performance Computing Act . . . . . . 61 3. Coordination of the New NASA/DOE Network . . . 63 4. A New Paradigm . . . . . . . . . . . . . . . . 64 VI. Conclusions and Reco--endations . . . . . . . . . . . . 66 Appendix: NSFNET Acceptable Use Policy . . . . . . . . . . . 78 NSF RESPONSE TO OIG REPORT . . . . . . . . . . . . . . . . . 80 2 |--------------------------------------------------------------------| Review of NSFNET I. Introduction In response to a Congressional request, we have reviewed the National Science Foundation's administration of the National Science Foundation Network ("NSFNET") program. In general we were favorably impressed with the NSFNET program and staff. Nevertheless, we make reco--endations to correct certain deficiencies and strengthen the upcoming re-solicitation. This review was initiated at the request of Congressman Rick Boucher, Chairman of the Subco--ittee on Science of the House Science, Space, and Technology Co--ittee. Congressman Boucher identified certain concerns after conducting a hearing on NSFNET.1 This report addresses the specific concerns raised by Congressman Boucher in the context of a general review of NSF's administration of NSFNET. In the course of this review, we interviewed, among others: the Assistant Director for NSF's Directorate for Computer and Information Science and Engineering ("CISE"), the Director of CISE's Division of Networking and Co--unications Research and Infrastructure ("DNCRI"), the Deputy Director of DNCRI, the former General Counsel of NSF (until February 1993), the former Executive Officer of CISE (until December 1991), the Director of ____________________ 1 Hearing on Management of NSFNET, 102d Cong., 2d Sess. (12 March 1992). These concerns included: modification of the cooperative agreement for expansion of the NSFNET backbone to T3 without re-bidding the award; privatization of the backbone; co--ercial use of the backbone; allowing conflicts of interest in the contractor-subcontractor relationship; and failure to properly oversee the agreement. 3 |--------------------------------------------------------------------| the Division of Legislative Affairs of NSF's Office of Legislative & Public Affairs, and individuals from the National Aeronautics and Space Administration, the Department of Energy, Merit Network, Inc. ("Merit"), Advanced Network & Services, Inc. ("ANS"), Performance Systems International, Inc., and the Co--ercial Internet Exchange. The documents we reviewed include: NSFNET and other program files from DNCRI, the Division of Grants and Contracts, and the Office of General Counsel; additional documents we requested from Merit and ANS; electronic mail files from the Director and Deputy Director of DNCRI; archives from the "com-priv" electronic mail bulletin board established by Performance Systems International, Inc.; co--ent letters received in response to the proposed solicitation; and various publicly available reports. II. Background: What NSFNET Is "The National Science Foundation Network (NSFNET) provides the opportunity for students, scientists, business people -- individuals from literally all walks of life -- to access resources ranging from electronic co--unity bulletin boards to supercomputers scattered across the continent and around the world. The NSFNET offers access to the nation's largest and fastest network for research, education, and technology transfer."2 "The NSFNET was designed to support the data networking needs of the research and education co--unity. It has become an essential infrastructure for the co--unity used daily to facilitate co--unication among researchers, educators, and students and to provide them with remote access to information and computing resources. The number of users, the number of connected networks, and the amount of network traffic continue to grow rapidly. ____________________ 2 "NSFNET: Bringing the World of Ideas Together", Merit Network, Inc. (1992). 4 |--------------------------------------------------------------------| "NSFNET also supports the goals of the High Performance Computing and Co--unications (HPCC) Program which was delineated in the President's Fiscal 1992 and 1993 budgets and became law with the passage of The High Performance Computing Act of 1991 (PL 102-194). One component of the HPCC Program is the National Research and Education Network (NREN) Program which calls for gigabit per second speed networking for research and education by the mid 1990s."3 As an example, imagine Dr. Smith, sitting at a computer terminal at NSF in Washington, D.C., sending a message to Dr. Jones, at a terminal at Stanford University in Palo Alto, California. Dr. Smith is connected to NSF's network, which connects about 1200 people. The NSF network cuts the message into pieces called "packets" and sends the packets through a type of switch, called a router, which is part of a regional network called "SURANET".4 SURANET has a router in College Park, Maryland, that is linked to a router called an external nodal switching subsystem ("ENSS"), which is an entry point for NSFNET. The ENSS routes packets such as Dr. Smith's from SURANET -- on NSFNET, now -- to a router called a core nodal switching system ("CNSS") at an MCI switching center in Perryman, Maryland.5 The CNSS sends the packet(s) that contain Dr. Smith's message out over MCI's long distance lines, via other CNSS's as necessary, to the CNSS at Dominguez Hills, California, which is linked to an ENSS in Palo ____________________ 3 "Public Draft -- Network Access Point Manager / Routing Authority and Very High Speed Backbone Network Services Provider for NSFNET and the NREN Program," 57 Fed. Reg. 26,692 (1992) (hereafter "Public Draft"). 4 Southeastern University Research Association Network. 5 The locations where routers can connect to MCI's long distance lines are referred to as MCI "points of presence". 5 |--------------------------------------------------------------------| Alto, where Stanford's regional network, "BARRNET",6 is connected. Leaving NSFNET, Dr. Smith's packets continue via BARRNET to Stanford's campus network, and finally to Dr. Jones's electronic mail box. Dr. Smith's message travels on NSFNET to the distant regional network used by Dr. Jones virtually instantaneously.7 The system of MCI's long distance lines and the CNSS's, ENSS's, and the local telephone lines connecting them convey traffic for the "NSFNET backbone". Moving from Dr. Smith's terminal, onto the NSF local network, onto SURANET, and so on -- at each link, getting closer to MCI's long distance lines -- there are more and more packets moving over the system. The routers need sufficient speed to ensure that the packets don't get bottlenecked along the way. Speed is generally expressed in standard telephone company units of multiples of megabits per second (mbps); the unit designations relevant to this discussion are "T1", which is 1.5 mbps, and "T3", which, at 45 mbps, is 30 times faster than T1. In general, the lines transmit information at very high speeds; the big MCI lines can handle thousands of mbps. Routers, which analyze each packet's destination and choose the appropriate path, limit the speed of the system. NSF and the other institutions on SURANET use T1 routers. Capacity ____________________ 6 Bay Area Regional Research Network. 7 It works the same way when what is being transmitted is large amounts of data, the only difference being there are many more packets. 6 |--------------------------------------------------------------------| can be increased by using multiple routers, in parallel, to feed messages into the lines.8 III. Progra--atic Issues A. The NSFNET Solicitation, Evaluation of the Proposals, and the Award The project solicitation for a cooperative agreement for the management and operation of the NSFNET backbone was issued by the Division of Networking and Co--unications Research and Infrastructure of the NSF Directorate for Computer and Information Science and Engineering.9 The solicitation closed on 14 August 1987, with the receipt of proposals from six offerors, including Merit.10 A panel of ten reviewers evaluated the proposals on the basis of nine weighted criteria: 1. Understanding NSFNET objectives and improving infrastructure (20%). 2. Capability to develop, manage, and operate NSFNET including prior experience managing networks and providing information services (15%). 3. Likelihood of availability of services in FY 1988 (15%). ____________________ 8 In theory, thirty T1 routers could provide the capacity of one T3 router; however, the software involved in coordinating thirty routers would be a logistical nightmare. 9 This solicitation envisioned that the award would be structured as a cooperative agreement, as opposed to a contract. Cooperative agreements and contracts are used by the government in different circumstances, as discussed below in Section IV.A. 10 Merit was the Michigan Educational Research Information Triad, a non-profit corporation managed by a consortium of eight Michigan universities: University of Michigan, Central Michigan University, Eastern Michigan University, Michigan State University, Michigan Technological University, Oakland University, Wayne State University, and Western Michigan University. The company's name has been legally changed to Merit Network, Inc. 7 |--------------------------------------------------------------------| 4. Capabilities of key personnel (15%). 5. Quality of circuit facilities (10%). 6. Quality of educational and information services (10%). 7. Quality of planned interaction with affiliated subnetworks (5%). 8. Quality of migration for NSFNET from the TCP/IP protocol to the Open Systems Interconnection protocols of the International Standards Organization (5%). 9. Plan for interconnection with other government networks including the ability to develop agreements (5%). Evaluated with the above criteria, the proposals received the following overall scores from the reviewers:11 Offeror: Reviewer: 1 2 Merit 4 5 6 1 VG G G F P F 2 E E VG G F G 3 E VG VG F G F 4 VG G G VG F F 5 VG VG VG F F P 6 G VG VG G F P 7 VG VG G G F F 8 VG --12 VG G G F 9 VG VG G F F P 10 VG VG VG G G F On the basis of these scores, the reviewers determined that the proposals from Offerors 4, 5, and 6 were technically unresponsive. The remaining three proposals were then considered by the reviewers in light of their pricing information, as set out below for the full five year period of the award: Offeror 1 $40,012,850 13 ____________________ 11 "E" = excellent; "VG" = very good; "G" = good; "F" = fair; "P" = poor. 12 This reviewer did not review Offeror 2's proposal. 13 Offeror 1's proposal included budgets that ranged from $34,862,370 to $43,927,503. The six different proposed budgets each depended on the speed of the backbone and circuits and the level of redundancy. The price quoted is for a backbone comparable to that proposed by Merit. 8 |--------------------------------------------------------------------| Offeror 2 $24,556,000 Merit $13,996,543 For comparable capacities and speeds, Offeror 1's and Offeror 2's proposals were substantially more expensive than Merit's. Offeror 1's proposal was much more costly in part because NSF would pay for purchasing all equipment including the routers, Offeror 1's indirect cost rate was high,14 and salaries and phone lines were expensive. Offeror 2's proposal was more costly primarily because its cost of phone lines was much higher than Merit's.15 Merit's proposal included contributions from MCI and IBM, which the Review Analysis prepared by NSF discussed as follows: "When pricing information was restored to the Merit, Inc. proposal, the Panel was initially concerned because the proposed cost to the NSF was unrealistically low for the level of service promised. The Panel then noted the extraordinary degree of cost-sharing proposed; this included: " $5M from the State of Michigan for facilities and personnel, " approximately $6M from MCI in reduced co--unication charges, and " $10M from IBM in equipment, installation, maintenance and operation. "When these cost-shared amounts were added to the proposed cost, the value of the Merit, Inc. proposed effort was seen by the panel to be fully comparable to the other two technically highly-rated proposals."16 ____________________ 14 Offeror 1's indirect cost rate ranged from 67.08% for FY88 to 73.25% for FY92, while Merit's was 58 percent. 15 MCI, the provider of lines under the Merit proposal, accepts a lower-than-market rate for access to its lines and connections. 16 Review Analysis at 2 (27 October 1987). 9 |--------------------------------------------------------------------| On the basis of the panel's evaluation, NSF staff made the following reco--endation: "Although the technical features of the Merit, Inc. proposal were not the most highly rated, the differences among the point scores of the three highly rated proposals were slight (less than 0.25); because of the large cost-sharing proposed, however, the Panel felt that the Merit, Inc. proposal offered the NSF and the scientific co--unity the best cost:benefit ratio, that the risk level of the proposal was acceptable, and that therefore an agreement should be negotiated between the NSF and Merit, Inc. for approximately the amount requested, to carry out the proposed work."17 The Memorandum to Members of the National Science Board in support of funding Merit's proposal stated: "A merit review panel was convened to evaluate six proposals obtained in response to the solicitation. After the six proposals were rated, three proposals were rated `technically competitive.' The panel was then provided with the cost portion of the proposal. Consensus was quickly reached when the Merit, Inc. proposal costs were shown to be 50 to 65 percent less than the other two technically competitive proposals. "Merit, Inc. was thus rated as the only proposal in the competitive range and was reco--ended for funding pending satisfactory responses to questions and concerns of the panel. The panel unanimously agreed that Merit, Inc. had excellent understanding of the objective of NSFNET and had proven capability of managing and operating an intercampus network for more than fifteen years. Their key personnel were also given high scores for the interaction with and work done for the current interim manager of NSFNET Backbone. The joint study partners IBM and MCI were seen as clearly co--itted to the success of the project. The experimental nature of the nodal switch equipment and the information service software were questioned by several reviewers. These concerns have been addressed by Merit, Inc. in their responses to questions. Overall, this was seen as an excellent proposal directed by highly competent and reputable personnel. The i--ediate expansion of their current Network Operation Center at Merit, Inc. would assure its availability in FY 1988. The cost sharing by the State of Michigan and Merit, Inc.'s joint study ____________________ 17 Review Analysis at 3 (27 October 1987). 10 |--------------------------------------------------------------------| partners of IBM and MCI make this a highly leveraged project."18 At its 16 October 1987 meeting, the National Science Board approved the award to Merit, and in November 1987 NSF awarded to Merit a five-year cooperative agreement (the "Cooperative Agreement") to implement and manage the NSFNET backbone. The NSFNET backbone, operating at T1, came on-line in July 1988. A Mid-Term Performance Review of the project was conducted by a panel of outside reviewers on 17-18 May 1989. Merit's performance during the first eighteen months of the agreement was generally deemed to meet or exceed expectations; the review panel found that Merit did not fail to meet expectations in any area. We conclude that NSF's decision to award the Cooperative Agreement for NSFNET to Merit was reasonable.19 B. Expansion of NSFNET and Conversion to T3 1. Necessity In February 1989, at the request of the Division of Networking and Co--unications Research and Infrastructure, Merit proposed an expansion of NSFNET primarily through the addition of up to fifteen new nodes in order to improve access for campus ____________________ 18 Memorandum to Members of the National Science Board Re Support for the Management and Operation of the NSFNet Backbone Network at 3-4. 19 For the purposes of this review, it is not relevant whether Michigan, MCI, and IBM have in fact contributed the precise amounts estimated in the Merit proposal, or somewhat greater or lesser amounts. What is relevant is that Merit's proposal was substantially less costly to the government than the other technically responsive proposals, and Merit's performance under the award has met or exceeded expectations as evaluated by both NSF staff and a panel of outside peer reviewers. 11 |--------------------------------------------------------------------| sites that had no access to a backbone node.20 The location of each new node would be determined in cooperation with NSF; the total projected budget if fifteen nodes were added was approximately $24.5 million,21 with implementation scheduled to begin in late 1989 and last six months. The proposal explained that, "[w]ith [T1 speeds already achieved], the NSFNET backbone is well positioned for the implementation of T-3 (45 Mbps) in the early 1990s reco--ended by [the Office of Science and Technology Policy (OSTP)]."22 In March, the five peer reviewers of the Merit proposal "were unanimous in their highly supportive co--ents relating to Merit's qualifications and of their current operations and management of the network . . . ."23 The reviewers also unanimously agreed that the NSFNET backbone needed to be expanded; however, they were concerned that T1 technology would soon become obsolete and that, therefore, investing large amounts of money in adding T1 nodes would not be cost-effective. As one reviewer stated: "The hardware technology proposed is relatively expensive and cannot easily migrate to support the T3 environment. Caution is advised in making too great an ____________________ 20 "Proposal for the Management and Operation of the NSFNET Backbone Network: Expansion," submitted by Merit (February 1989) (hereafter, "Merit Expansion Proposal"). 21 MCI agreed to waive its installation fees although it would pass on costs for the installation of local access lines. The main reason for the increase was the cost for the additional lines. 22 Merit Expansion Proposal at 3. The OSTP report reco--ended that the NREN be upgraded to T3 in the early 1990's. OSTP, "The Federal High Performance Computing Program," at 32, 36 (8 September 1989). 23 Review Analysis at 5 (1 June 1989). 12 |--------------------------------------------------------------------| investment in this technology at this time with a possible limited useful life." Another reviewer expressed similar concerns: "The new NSS nodes will increase access/connectivity, but this will also increase the requirements for testing of new versions and enhancements to the NSFNET. . . . Expansion of the NSFNET is needed but must be balanced with the planned advancement to new technologies for the NSFNET which will bring T3 network capacities to the users in the 3 year period of this proposal." In June 1989 the National Science Board approved an increase in the authorization limit for NSFNET from $14 million to $20 million, as reco--ended by the Division of Networking and Co--unications Research and Infrastructure. This decision was made on the basis of the increase in traffic from September 1987 (75 million packets for the old network) to February 1989 (600 million packets) and the increase in the number of institutions connected to regional networks. The additional money allocated was intended to provide increased capacity and connectivity and reduced delays by adding new nodes and changing the physical topology of the backbone. After approval of the funding level increase, locations for the new nodes were proposed and were assessed by a panel of peer reviewers, and studies of new network configurations, including possible inclusion of newly-available T3 technology, were undertaken by Merit at NSF's request. In December 1989, traffic projections indicated that the increase in network use could reach the limits of NSFNET's capacity during the next year. Therefore, instead of adding more T1 routers and fifteen new nodes to reduce this bottleneck, Merit proposed that only one to three new nodes be added at T1, two nodes be added at T3, and 13 |--------------------------------------------------------------------| portions of the existing backbone be upgraded to T3.24 T3 nodes provide 30 times the speed of T1 nodes, and, therefore, T3 nodes can be used for applications where very high speeds are necessary such as moving pictures. Thus, the expansion to T3 would enable the use of network applications that were impossible under T1. The panel of peer reviewers gave Merit high praise for its past performance and supported Merit's proposal. In April 1990, Merit submitted a revised statement of work, "based on the input received from the National Science Foundation, in particular the need for adding nodes to and expanding the switching and transmission capacity for the NSFNET backbone."25 Merit then submitted a scope of work revision, proposing the addition of three new nodes at T3 combined with the T3 upgrade of six nodes instead of adding any new T1 nodes.26 On 29 May 1990, an amendment to the Cooperative Agreement provided funding for this revision.27 In August 1990 Merit submitted a proposal for the "Expansion of T-3 Capabilities for the National Science Foundation Network (NSFNET)". Thereafter, in November 1990, the Board approved an increase in the Cooperative Agreement's authorization limit from $20 million to $28 million so the entire backbone could provide ____________________ 24 Memorandum from Merit to NSF (20 February 1990). 25 "Revised Statement of Work / NSF Supplemental Proposal No. 8944037" (20 April 1990). 26 "Scope of Work Revision / Proposal No. NCR-8912875" (30 April 1990). 27 Amendment No. 4 to Cooperative Agreement No. NCR-8720904. 14 |--------------------------------------------------------------------| T3 speeds. The conversion to T3 routers was completed in mid-1992.28 At the same time as the T3 expansion, a new network architecture was implemented for the T3 technology under which CNSSs were linked by multiple T3 circuits. In addition, MCI assumed the responsibility of maintaining the connection of each ENSS to its CNSS at MCI's point of presence (through local telephone company lines or MCI's own lines). When the network was expanded to T3, the cost increased dramatically. Specifically, the additional cost to provide T3 to the first eight sites was about $8 million, of which almost $7 million was for the direct costs of transmission services (including circuits, equipment, installation, and maintenance). The three primary reasons for this cost increase were the new high-performance router technology, the costs of the additional capacity and speed on the MCI lines, and the requirement that the local lines from the CNSS's to the ENSS's be able to handle T3 speeds. We believe NSF's decision to upgrade NSFNET to T3 **before** the T1 network was saturated was reasonable. We also believe the price was not unreasonable: even with the additional costs, Merit provided NSFNET with more nodes and complete T3 expansion in 1992 at a total cost of $28 million29 -- considerably less ____________________ 28 The current network is "T3" in the sense that it can be upgraded to T3 whenever there is a need to do so. The routers are presently able to handle 22.4 mbps (half of T3), and they can be readily configured to provide full T3 speed. The routers have not yet been upgraded to T3 because the volume of traffic has not been sufficient to saturate them at half that rate. 29 Not including $95,401 to connect DARPA to the backbone, and $754,701 to connect NASA. 15 |--------------------------------------------------------------------| than Offeror 1 proposed to provide only T1 and about fifteen percent more than Offeror 2 proposed to implement the slower technology. Also, although T3 costs substantially more than T1, it provides a **30-fold** increase in network speed. 2. Amendment or Resolicitation? Nonetheless, it is legitimate to question whether NSF should have issued a new solicitation for the node expansion and/or the T3 conversion, rather than increasing the existing award. We must therefore consider whether the public was on notice, when the original NSFNET solicitation was issued, that the awardee of the NSFNET cooperative agreement would be responsible for modifications and expansions of the backbone to acco--odate increased use and advances in technology. The project solicitation for the management and operation of the NSFNET set out NSF's expectations: "NSFNET is dynamic. It will change with evolving network affiliations, improved technologies, competing co--unications costs, varying traffic load, and other similar factors. It is necessary that NSFNET management be able to advise and acco--odate changes in the top-level, or backbone, architecture."30 The network engineering, protocols, and standards in the solicitation required that the managing organization "provide or arrange for the following services", including: "d. Plan, and with the approval of the NSF/DNCRI install, modifications to the network topology and node assignments to improve overall network performance and reduce operations and/or circuit costs to the government. Alteration of Backbone ____________________ 30 Project Solicitation for Management and Operation of the NSFNET Backbone Network at 1 (hereafter "Project Solicitation"). 16 |--------------------------------------------------------------------| nodes and/or Tail31 links are specifically permitted. "e. Coordinate, and with the approval of the NSF/DNCRI install, changes to the NSFNET Backbone and Tails as required to acco--odate network growth and traffic pattern changes as additional affiliated networks are connected to the backbone or as traffic from existing networks grows."32 The solicitation also required that the proposals contain "[p]lans and procedures to monitor traffic in order to plan for and acco--odate growth including working with new sites and phasing in new equipment."33 As requested in NSF's solicitation, Merit's proposal anticipated increases in required speeds: "Our solution utilizes technology designed to work at T1 speeds and well beyond that as needed in the future. This design can be extended quickly, to stay ahead of the anticipated, increasing network load."34 "[Merit's structure p]rovides a foundation for future networks with broader user bases and greater capabilities as projected by the FCCSET Report to Congress."35 "[W]e believe that the structure proposed will provide a flexible base for further network development and thereby position NSFNET to evolve into the type of network envisioned by the FCCSET study."36 ____________________ 31 Tails are "the circuits, gateways, taps, etc. extending from the Backbone to [regional] networks, each of which is an independent entity serving its own end-user co--unity and providing its own facilities and user services." Project Solicitation at 1. 32 Project Solicitation at 3. 33 Project Solicitation at 6. 34 "Management and Operation of the NSFNET Backbone Network," at vi, submitted by Merit (hereafter "Merit Proposal"). 35 Merit Proposal at vi. 36 Merit Proposal at viii. 17 |--------------------------------------------------------------------| "At the beginning of 1990 we expect to begin the transition into NSFNET's second phase, incorporating into it [T3]-capable software. If demand for higher rates exists, and funding for the additional hardware and transmission becomes available, we have targeted the beginning of the deployment of a higher speed network for that time."37 One reviewer co--ented that Merit's proposal was an "[i]nteresting hybrid approach that allows adjustment of traffic **and** [speed] . . . ." The other two competitive proposals, per the solicitation, also anticipated enhancements to the network. Offeror 1's proposal stated: "The NSFNET Operations and/or Engineering staffs, in consultation with NSF and other appropriate organizations, will plan changes to NSFNET. Network Information Services will announce and document changes well in advance."38 "Major changes in co--unications facilities will be carefully planned to overlap with existing facilities, providing cutover to new facilities."39 "3.3.6. Site Addition Procedures "We urge NSF to ask us to review proposals for connection to NSFNET as part of the peer review process. Once the determination has been made that the new site or network is to be connected, we would expect to follow [certain] procedures . . . ."40 "11. Experimentation on the NSFNET Backbone "[We] are deeply involved in the ongoing Internet engineering efforts. Our strategy for protocol development will be to work with the various task forces of the Internet Activities Board to define areas where research and experimentation need to be done; to foster and participate in that research and experimentation as much as possible; and to develop ____________________ 37 Merit Proposal at 13. 38 Offeror 1 Proposal at 12. 39 Offeror 1 Proposal at 12. 40 Offeror 1 Proposal at 13. 18 |--------------------------------------------------------------------| deployment strategies for a new function or architecture when the task forces reach a consensus. We want to be sure that the NSFNET backbone is a state-of-the-art network to support the tremendous requirements of its clients . . . . "The ongoing evolution of the Internet will require integration of experimental components from time to time . . . . "[We] will maintain a test lab for testing the interoperability of any experimental routers before they are used on the backbone . . . ."41 Similarly, Offeror 2's proposal anticipated expansion and enhancements: "[A]s the needs of NSFNET grow and change, additional processors can be added or co--unications interfaces replaced at minimum cost."42 "We expect that NSFNET traffic volume and patterns will change as mid-level networks are added and as new networking applications come into use. [We] will monitor backbone traffic patterns during the life of the agreement to insure that the backbone meets service requirements at minimum cost. We will study traffic measurements and interview NSF officials, mid-level network managers, and supercomputer site administrators to estimate future traffic loads and network growth. Twice a year, starting three months after the network is installed, network designers using [our] expert system will analyze this information and make reco--endations to provide adequate capacity and redundancy to meet then-current and near-term future co--unication requirements. [Our] staff will also be available to plan for the addition of new mid-level networks or other changes to NSFNET."43 "4.4.11. Modifications to Network Topology. Changing levels and flows of data traffic can require modification to network topology, as would adding or deleting mid-level networks. The network analysis services described in 4.4.4 can be coupled with topological design to validate the results of previous designs and to provide data for network modeling. "In addition to routine performance monitoring, network performance (network delay, throughput and circuit ____________________ 41 Offeror 1 Proposal at 18. 42 Offeror 2 Proposal at 2. 43 Offeror 2 Proposal at 6-7. 19 |--------------------------------------------------------------------| utilization) and reliability will be reassessed every six months. Changes to the topology will be reco--ended to account for changes in traffic volumes and traffic patterns. Topological changes might include the addition or removal of specific circuits or devices, as well as redeployment of existing circuits and devices. Internet packet routers will be reconfigured as necessary to meet requirements for throughput and to interface to appropriate co--unications lines."44 Thus, the NSFNET solicitation explicitly envisioned expansions of and improvements to the network, and the public was on notice that the successful offeror would be responsible for expansions of the NSFNET backbone within the period of the award. The NSFNET solicitation sought proposals that anticipated expansion and upgrading of the network, and Merit's winning proposal did so. In our view it was reasonable for NSF to modify the Cooperative Agreement to fund expansion of NSFNET and the T3 upgrade. Indeed, particularly in light of the extremely favorable evaluations Merit's work on NSFNET received in its peer reviews, NSF would likely have had no justification to cut short the term of the Cooperative Agreement and re-compete it.45 C. Spinning Off: ANS & CO+RE 1. ____________________ 44 Offeror 2 Proposal at 10-11. 45 It has been alleged that the NSFNET backbone was unreliable and that the speed was not that which had been promised. NSFNET was always expected to be, and has been, a testbed for the cutting edge of networking technology -- a work in progress. Throughout the period of performance of the Cooperative Agreement, NSF and Merit have pushed and extended the limits of networking technology. Each upgrade of the network has presented problems that had not been encountered previously, and NSF and Merit, IBM, and MCI have worked together to solve them. We have no basis to disagree with the judgment of NSF staff -- shared by the outside reviewers who evaluated Merit's performance at midterm and for each proposed upgrade -- that the performance of NSFNET has been within reasonable expectations. 20 |--------------------------------------------------------------------| CO+RE and the Infrastructure Pool Before the expansion of NSFNET and conversion to T3, in September 1990, Merit, IBM, and MCI formed a nonprofit corporation called Advanced Network & Services, Inc. ("ANS") to take over management and operation of NSFNET. In ANS's Certificate of Incorporation, the purpose of ANS is set forth as follows: "The Corporation is a non-profit organization dedicated to the advancement of education and research in the interest of improving the ability of the United States to compete in the global economic environment. The Corporation will concentrate on computer networking and related services, an area clearly recognized as a vital component of United States competitiveness. The Corporation shall help establish a high-speed computer network which will be maintained at the leading edge of technology, and which will eventually feature multi-gigabit per second data transfer rates. The Corporation will also help to expand the access to and interchange of information technology resources among academic, government and industry users. In addition, the Corporation will engage in research and development work which will support the academic and research co--unities and contribute to United States preeminence in high speed network technology and related services." NSF agreed to the assignment of Merit's responsibilities under the Cooperative Agreement to ANS, and further agreed that ANS could seek new customers, as follows: "NSF agrees that [ANS] may solicit and attach to the NSFNET Backbone new users, including co--ercial users, and may connect them to new or existing nodes on the Backbone, with the understandings that: "1) such users will reimburse [ANS] for at least the full average cost of the connection, the added traffic, and additional related support, and "2) the reimbursements will be used to enhance the network infrastructure and services, in order that the level of service provided by MERIT under its Cooperative Agreement with NSF not be diminished. 21 |--------------------------------------------------------------------| "NSF and MERIT will agree on the technical means of compliance with 2) above."46 In May 1991, ANS created "ANS CO+RE Systems, Inc." ("CO+RE"), a for-profit corporation, to: "engage in activities that are not solely in support of scientific research and education, but include providing network services for (1) mixed co--ercial and scientific research and educational traffic, (2) co--ercial traffic, and (3) any other activity which would constitute an unrelated trade or business if carried on by [ANS]. For example, co--ercial users who want direct connectivity with the scientific research and educational co--unity would connect through [CO+RE]. Their attachment fees would be paid to [CO+RE] . . . . Network consulting services also will be provided by [CO+RE]. After-tax profits from [CO+RE] generally will be dividended to [ANS] to the extent they are not needed for reinvestment in [CO+RE]."47 The NSFNET Acceptable Use Policy (the "AUP"), which is set out in the appendix to this report and is discussed in greater detail in Section III.D., prohibits purely co--ercial traffic from using NSFNET.48 When CO+RE was formed, NSF program staff agreed with Merit and ANS that the Merit/ANS network operations center, IBM-provided routers, and MCI-provided lines were not subject to the AUP because NSF was not paying for equipment and facilities, but was instead paying for the conveyance of NSFNET traffic and the provision of network support services. Nonetheless, NSF did place the following conditions on the use of the network for conveying co--ercial traffic: ____________________ 46 Letter from NSF to Merit (10 September 1990). 47 Attachment No. 4 to ANS's Filing with the Internal Revenue Service for Non-Profit Status at 3. 48 Under the Acceptable Use Policy, research and education users can use NSFNET to contact co--ercial entities and receive co--ercial services and products over NSFNET (such as Dialog), but co--ercial entities cannot affirmatively use the backbone for advertising co--ercial activities. 22 |--------------------------------------------------------------------| "NSF agrees that ANS may move co--ercial traffic in both directions across the NSF sponsored Backbone gateways, providing that: "(1) ANS recovers at least the average cost of the co--ercial use that traverses the NSF sponsored gateways. "(2) Excess revenues recovered above costs for this use after tax will be placed in a pool to be distributed. "(3) An ANS resource allocation co--ittee will be formed with representation from the participating NSF sponsored gateway management, other network organizations, the NSF and ANS to distribute those funds with the objective of further building national and regional infrastructure, and "(4) MERIT and ANS ensures that the attachment and service sponsored by the NSF under Merit's Cooperative Agreement with the NSF is not diminished. "NSF, MERIT and ANS will agree on the technical means of compliance with the points outlined above."49 In our view, the establishment of the infrastructure pool is consistent with the requirement in the Cooperative Agreement that "project income received or accruing to the [awardee] during the period of this award shall be retained and added to the funds co--itted to the project by the Foundation and used to further project objectives." We reco--end that NSF ensure that the infrastructure pool is funded properly. ANS subsequently explained how the "infrastructure pool" would work: regional networks may sign "connectivity agreements" with ANS, thereby "simply . . . agree[ing] to accept traffic from ANS CO+RE customers." ANS explained further: "Networks that sign ANS connectivity agreements pay nothing, and, because they agree to let co--ercial traffic travel across their gateways and through their ____________________ 49 E-mail message from NSF to Merit (24 May 1991). 23 |--------------------------------------------------------------------| networks, they are eligible to receive grants from the National Infrastructure Pool."50 "*ALL* of the additional revenues due to [co--ercial] charges go into the infrastructure pool and get distributed back to the service providers that participate in the connectivity agreement to build infrastructure."51 Thus, the revenues in the infrastructure pool are to be distributed as determined by the resource allocation co--ittee, which consists of one representative each from ANS, NSF, and each network service provider that has signed a connectivity or gateway agreement. We reco--end that NSF ensure that the infrastructure pool is equitably distributed among the networks that are connected to NSFNET and that this money is used to build national and regional infrastructure as required by the agreement between NSF and Merit. We further reco--end that, at the close of the term of the Cooperative Agreement, NSF ensure that Merit conducts a cost and compliance audit of ANS and CO+RE to verify that the infrastructure pool has been funded and distributed appropriately.52 ____________________ 50 E-mail message from ANS to com-priv@psi.com and other addressees (12 December 1991). 51 E-mail message from ANS, cc'd to com-priv@psi.com (13 January 1992) (emphasis in original). 52 As discussed below, Merit has been allowed to permit co--ercial traffic over the T3 network created with NSF support; this use was not anticipated by the parties at the time of the original solicitation or Cooperative Agreement. Although we have concluded that it was not inappropriate to allow such co--ercial use of this NSF-funded resource, the unanticipated benefit that Merit received resulted in the formation of the infrastructure pool and hence the need to audit it. Therefore, we feel it is equitable for NSF to ensure that Merit conducts a cost and compliance audit of ANS and CO+RE **at Merit's own expense** (an expense which should not decrease CO+RE's contribution to the infrastructure pool). (continued...) 24 |--------------------------------------------------------------------| 2. Co--ercial Use We have been told by NSF staff that in 1986-87, when NSF conducted the NSFNET solicitation and entered into the Cooperative Agreement with Merit, it was not clearly foreseeable when, or if, high-speed networking would become co--ercially viable. Therefore, NSF did not address the issue of co--ercial use of the network by the awardee in the solicitation or the Cooperative Agreement. Similarly, **none** of the offerors addressed co--ercial use in its proposal. As discussed above, we have concluded that the decision to proceed with the T3 expansion by amending the extant cooperative agreement with Merit was reasonable largely because the expansion of the network was anticipated in the NSFNET solicitation and was therefore anticipated by Merit as well as the other offerors.53 In this case, neither the solicitation nor the proposals anticipated or ____________________ 52 (...continued) The Office of the Inspector General contracted for an independent audit of Merit for the period from 1 December 1987 to 30 June 1991. No questioned costs were found, and no instances of internal control structure weaknesses were detected. "Questioned costs are costs for which there is documentation that the recorded costs were expended in violation of the law, regulations or specific conditions of the Cooperative Agreement and the grant or those costs which require additional support by the grantee or which require interpretation of allowability by the NSF grant or contract officer." "Financial Audit of National Science Foundation Awards to Michigan Educational Research Information Triad, Inc.," at 3, Office of Inspector General Report No. 92-1017 (20 December 1991). Non-statistical sampling was used. However, this audit did not include CO+RE and the infrastructure pool, which should be a particular focus of the audit performed after the termination of the Cooperative Agreement. 53 See Section III.B.2. 25 |--------------------------------------------------------------------| prohibited co--ercial use.54 Because of this ambiguity, the question of whether co--ercial use should have been permitted is properly resolved by reference to the objectives of NSF in general and NSFNET in particular: if co--ercial use would have a negative effect on either NSF's objectives or NSFNET's objectives, then co--ercial use must be prohibited. NSF decided that co--ercial use would not have a detrimental effect on the objectives of NSF or NSFNET. In our view NSF reasonably concluded that allowing co--ercial use of the network -- with the conditions NSF imposed -- is consistent with NSF's overall statutory mandate "to foster the interchange of scientific and engineering information among scientists and engineers" and "to foster and support the development and use of computer and other scientific and engineering methods and technologies, primarily for research and education in the ____________________ 54 Merit's original and subsequent proposals, as well as the amendments to the Cooperative Agreement, do provide detailed information about equipment and facilities to be used for NSFNET and subsequent enhancements. This level of detail is consistent with the requirement that proposals demonstrate the offeror's ability to meet NSFNET's technical requirements. However, we do not read these documents as implying that the equipment and facilities would be **dedicated** to NSFNET. The network created by Merit under the NSFNET Cooperative Agreement is composed of routers, lines, and network operations support. There is certainly no question that the lines to be provided by MCI were not going to be physical wires and cables dedicated to NSFNET. The network operations support consists for the most part of staffing at a network operations center and service personnel provided, as needed, by MCI and IBM. Again, there is nothing in any of the documentation to imply that MCI and IBM were going to provide specific, dedicated individuals, and the Merit proposal explicitly stated that its network operations center would be shared with non-NSFNET uses. Merit Proposal at 51-52. There is no discussion in the solicitation or the proposals about shared use of the routers. Taken as a whole, however, we conclude that shared use of the routers, lines, and network operations support for NSFNET is not inconsistent with the Cooperative Agreement and subsequent amendments. 26 |--------------------------------------------------------------------| sciences and engineering".55 NSF further concluded that co--ercial use of the network -- along with concomitant creation and funding of the infrastructure pool -- would further the objectives of NSFNET, by enhancing connectivity among co--ercial users and research and education users and by providing for enhancements to the network as a whole.56 Under these circumstances, we conclude that it was not unreasonable for NSF to decide that allowing Merit to permit some co--ercial traffic over the network created by Merit was consistent with the objectives of NSF and the NSFNET program.57 It is clear that Merit, ANS, and CO+RE have benefited from CO+RE's ability to sell use of the network to co--ercial entities, in a manner not anticipated when Merit and NSF entered into the NSFNET Cooperative Agreement. However, in view of the fact that the objectives of the program were furthered by co--ercial use of the network, the mere fact that an unexpected ____________________ 55 42 U.S.C. 1862(a)(3)-(4). 56 As set out above, in the agreement with NSF allowing co--ercial traffic, Merit and ANS were required to "[ensure] that the attachment and service sponsored by the NSF under Merit's Cooperative Agreement with the NSF is not diminished." ANS has in fact gone well beyond that requirement, by providing a full T1 backup system while the T3 network was being deployed and debugged, as well as adding redundancy to the T3 network. 57 Our understanding of NSF's reasons for allowing co--ercial use of the T3 network is based on oral explanations provided by NSF staff during the course of our review. As discussed in Section III.C.4., appropriate documentation for this decision does not exist. 27 |--------------------------------------------------------------------| benefit accrued to the awardee is not objectionable58 and would not require a re-solicitation. In our view, it is also significant that ANS is a not-for-profit entity created to operate for the public good. While Merit, IBM, and MCI are all closely associated with ANS (Merit, IBM, and MCI each have one seat on ANS's Board), such association does not seem to us to be significant, particularly in light of ANS's tax-exempt status and its corporate mission to further the good of the networking co--unity.59 Once ANS ____________________ 58 As project income, any **financial** benefit accrued during the life of the award must be applied to further the purposes of the award; for example, under the NSFNET award, CO+RE's after-tax profits fund the infrastructure pool. See Section III.C.1. 59 ANS has qualified for tax-exempt status under 26 U.S.C. 501(c)(3), which states: "Corporations . . . organized and operated exclusively for . . . scientific, . . . or educational purposes, [and] . . . no part of the net earnings of which inures to the benefit of any private shareholder or individual . . ." are tax-exempt. We see no reason not to rely on the determination of the Internal Revenue Service that ANS meets this definition. The Articles of Incorporation of ANS are also consistent with ANS's non-profit status because they prohibit any financial gain (except Merit's potential gain from liquidation of ANS) to Merit, IBM, and MCI (its members) as follows: "ARTICLE FOURTH "No part of the earnings or assets of the Corporation shall inure to the benefit of, or be distributable to, its members (subject to the provisions of Article Fifth below), directors, officers or other private persons, except that the Corporation shall be authorized and empowered to pay reasonable compensation for services rendered and to make payments and distributions in furtherance of the purposes set forth . . . above. . . . "ARTICLE FIFTH "In the event of the liquidation, dissolution or winding-up of the affairs of the Corporation . . . , (continued...) 28 |--------------------------------------------------------------------| decided that co--ercial use was also desirable, a second spin-off -- this time a for-profit entity which would thus pay taxes on its co--ercial profits -- was used. At least for the term of the NSFNET Cooperative Agreement, CO+RE's after-tax profits will fund the infrastructure pool, to be distributed according to the principles discussed above; even after the term of the Cooperative Agreement, if the profits are received by ANS in the form of dividends, they must be used in a manner consistent with ANS's tax-exempt status. Considering all of these factors, we conclude that NSF's decision to allow Merit to permit co--ercial traffic over the network created by Merit under the NSFNET Cooperative Agreement was not unreasonable. 3. Co--ercial Access Was CO+RE granted an advantage over other network providers by virtue of ANS's relationship with NSF? CO+RE is able to sell access to the T3 network (which was created jointly -- cooperatively -- by NSF and Merit/ANS/IBM/MCI) for co--ercial traffic. However, NSF has publicly stated that if another company wishes to send co--ercial traffic over the T3 network set up by NSF and Merit/ANS, it can do so on the same terms accorded ____________________ 59 (...continued) the Board of Directors will . . . distribute all the assets of the Corporation in the manner the Board of Directors determines to be best suited to the accomplishment of the purposes of the Corporation . . . , but in no event shall any portion of such transfer inure to the benefit of any member of the Corporation (other than any member that is a not-for-profit organization and will use such assets to accomplish the purposes of the Corporation)." 29 |--------------------------------------------------------------------| to CO+RE.60 Furthermore, although a few research and education entities (such as the NSF Supercomputer Centers) presently have use for T3 service, there is not yet any **co--ercial** demand for T3 service, and at this time CO+RE has no co--ercial T3 customers. In our view, the very fact that no co--ercial provider has pursued NSF's offer of equal access to the T3 ____________________ 60 After learning that CO+RE was marketing co--ercial access to the T3 network, a co--ercial network provider inquired: "Now given that [the network] is a government resource there, can [we] sign a contract with [a regional network connected to the T3 network] to provide them with co--ercial access through that interface?" E-mail message to NSF, cc'd to com-priv@psi.com (16 December 1991). NSF responded: " It's a government-*sponsored* resource. NSF doesn't own the gateway. "But the short-form answer is `Yes, under the same conditions as ANS.' There are enough variables, however, that we need to meet to find out exactly what you have in mind. . . ." E-mail message cc'd to com-priv@psi.com (19 December 1991). The NSF position was reiterated publicly: "Anybody else wanting to provide regional-to-regional interconnect for non-R&E traffic between two regionals who allow such traffic can . . . use the NSFNET Backbone gateways under the same conditions as ANS CO+RE." E-mail message cc'd to com-priv@psi.com (17 January 1992). The open-access policy was also reiterated by NSF internally: "NSF will permit other private providers to attach to Backbone nodes and pass non-conforming traffic to NSF Backbone service clients who elect to receive it, on the condition that they, like ANS, contribute to the fund that is used to increase network capacity so that R&E service is not impacted." E-mail message (11 February 1992). 30 |--------------------------------------------------------------------| network further indicates that CO+RE's access to the T3 network is not, in and of itself, co--ercially significant.61 At the end of October 1992, the original period of the Cooperative Agreement expired: NSF and Merit have amended the Cooperative Agreement to extend its period for eighteen months, to give NSF more time to undertake the solicitation for further enhancement of NSFNET and implementation of the NREN (as discussed below). We reco--end that, for the remaining period of the amended Cooperative Agreement, NSF ensure that other network providers continue to be offered access to the T3 network on the same terms as CO+RE, and, if the offer is accepted, then access is provided fully and fairly. 4. The Decisionmaking Process The record is utterly barren of documentation of NSF's reasoning for allowing co--ercial use of the network. The only documentation of NSF's decision to allow co--ercial use is an electronic mail message and a letter, both from the Director of DNCRI, describing the conditions under which ANS/CO+RE could ____________________ 61 The co--ercial network provider who had inquired about equal access to the T3 network dismissed NSF's offer of equal access: "To get real serious -- [we] do[] not want to meet with NSF alone to discuss how we can take pecuniary advantage on millions of tax dollars. You need to discuss this with the co--unity [in] an open manner, you need to fix the current wrongs -- quickly. I want a level playing field." E-mail message to NSF, cc'd to com-priv@psi.com (1 January 1992). Nonetheless, we are satisfied that other co--ercial providers were publicly offered co--ercial access to the network on the same basis as CO+RE, so NSF has not given CO+RE an advantage over its co--ercial competitors. 31 |--------------------------------------------------------------------| convey co--ercial traffic over the network -- and even those documents were not in the program files. As a result of the lack of documentation of the decision regarding co--ercial use, we were required to reconstruct the reasoning behind the decision from interviews with the program staff, more than two years after some of the events in question. Such post hoc explanations can lead to inaccuracies. For example, we were told by the program staff that Merit's proposed price for the T3 upgrade was conditioned upon ANS's ability to sell co--ercial access to the T3 network. This led us to initially conclude that the decision to allow co--ercial applications was based on an assumption that NSF would save funds if it allowed co--ercial use. However, representatives of both Merit and ANS have unequivocally stated that the price of T3 was not conditioned on co--ercial use. 32 |--------------------------------------------------------------------| Important program decisions,62 particularly those involving a program as large as NSFNET, must be both well-reasoned **and** well-documented.63 Ultimately, clear documentation facilitates clear reasoning and results in better decisions. Requiring documentation necessarily results in more careful analysis -- including evaluation of whether peer review, public co--ent, supervisory review, and/or evaluation and approval by the National Science Board are appropriate. Peer review of, and/or public co--ent on, important decisions ensure that the views of the affected co--unity are considered. Documented supervisory review ensures that no one individual acts beyond his or her mandate. Pursuant to NSF's Proposal and Award Manual: "Prior approval by the National Science Board (NSB) is required for all reco--ended awards involving any of the following conditions: ____________________ 62 Importance is determined by reference not just to cost or project performance, but to **all** relevant factors. The decision to allow the network service provider spun off by Merit -- ANS -- to convey co--ercial traffic over the T3 network was, in our view, indisputably an important decision requiring significant, documented progra--atic evaluation before it was made. NSF has a long-standing policy prohibiting NSF awardees from using NSF-supported facilities co--ercially in a manner that "may have a material and deleterious effect on the success of private companies engaged in the provision of equivalent services." NSF Important Notice Number 91 (11 March 1983). Congress addressed this issue recently when it amended the NSF Act: "The Co--ittee . . . does not intend that, under the authority provided by the legislation, NSF will institute policies that result in improper competition between NSFNet, or the interim NREN, and other co--ercial networks." H.R. Rep. No. 567, 102d Cong., 2d Sess. 4 (1992). As explained in Section III.C.3, we have concluded that the manner in which NSF allowed co--ercial traffic over the T3 network did not give ANS/CO+RE a competitive advantage, because all co--ercial providers are allowed access on equal terms; nonetheless, the decision was clearly an important one requiring thorough consideration and documentation. 63 See also Section IV.D. 33 |--------------------------------------------------------------------| "a. a total co--itment of $2 million or more; "b. an annual co--itment of $500,000 or more; "c. a policy issue that has not previously been resolved by the Board; and "d. a new program or novel set of circumstances."64 Although the initial decisions to (1) award NSFNET to Merit and (2) expand the network to T3 were submitted to the Board for approval, the decision to (3) allow co--ercial use of the T3 network was not reviewed by the Board. In our view, prior approval by the Board would have been appropriate because the decision to allow co--ercial use of the T3 network was "a policy issue that ha[d] not previously been resolved by the Board" and involved a "novel set of circumstances." As explained above, we have concluded that the decision to allow co--ercial use of the T3 network was within NSF's discretion.65 However, as the Supreme Court observed in an analogous situation, "the orderly functioning of the process of review requires that the grounds upon which the administrative agency acted be clearly disclosed and adequately sustained."66 Thus, the dearth of documentation of NSF's underlying reasoning -- as well as the lack of evidence of peer, supervisory, or National Science Board review of this decision -- reduces our confidence in this conclusion. This contrasts sharply with our confidence in our conclusions regarding the reasonableness of the well-documented decisions to award the NSFNET Cooperative ____________________ 64 NSF Manual 10 at II-12. 65 See Section III.C.2. 66 Securities and Exch. Co--'n v. Chenery Corp., 318 U.S. 80, 94 (1943). 34 |--------------------------------------------------------------------| Agreement to Merit and to upgrade the network to T3,67 because those decisions were based on written proposals, peer reviews, and staff evaluations, as well as final review and approval by the National Science Board. We reco--end that, in the future, all significant program decisions by DNCRI and the reasoning behind them be well-documented (including documentation of significant meetings and telephone conversations with people outside NSF) in a manner that reflects the level of NSF supervisory review and approval that occurred, and we also reco--end that, when appropriate, issues involving the program be reviewed by scientific peers and/or submitted to the National Science Board for review and approval. 5. Public Announcement In addition to the lack of formal documentation, it is legitimate to question whether NSF's decision to allow co--ercial use should have been publicly announced. A review of the chronology of events is useful: NSF gave formal permission for Merit to make a subaward to ANS for the operation and management of the NSFNET backbone in September 1990, and in May 1991 NSF agreed to the specific terms under which co--ercial traffic would be allowed over the T3 network. ANS began marketing co--ercial services in 1990. CO+RE was formed by ANS in May 1991 and began more active solicitation of co--ercial customers; its first co--ercial customer was signed up in September 1991 and connected to the T3 backbone in November 1991. CO+RE's co--ercial ____________________ 67 See Sections III.A. and III.B. 35 |--------------------------------------------------------------------| competitors soon became aware that CO+RE was marketing co--ercial access to the T3 backbone, and on 3 October 1991 one of CO+RE's co--ercial competitors asked in writing for NSF to provide copies of the agreements between NSF and Merit that allowed such marketing. The agreements were provided to the co--ercial competitor on 31 October 1991. On 7 December 1991 that co--ercial competitor announced on the "com-priv" computer bulletin board that it had obtained, and was disclosing for the first time publicly, "Secret NSF-Merit/ANS Agreements." The government is under no general duty to **announce** modifications of agreements between itself and its awardees. The government **is** required by the Freedom of Information Act ("FOIA") to provide certain information, upon written request.68 We do not believe the actions of NSF can fairly be viewed as reflecting an intent to keep "secret" the agreements regarding co--ercial use. As noted above, when copies of the agreements were requested in writing, NSF provided them within four weeks. Since that written request was expressly **not** made under FOIA (which requires a response within 10 days69), we have no basis for objecting to the time taken in producing them. Because NSF intended other co--ercial network providers to have access to the T3 backbone on the same terms accorded to CO+RE,70 however, we believe NSF should have affirmatively announced **this** development to the networking co--unity. By ____________________ 68 5 U.S.C. 552. 69 5 U.S.C. 552(a)(6)(A)(i). 70 See Section III.C.3. 36 |--------------------------------------------------------------------| making a public announcement, NSF could have avoided a controversy which in our view was generated primarily by (1) ignorance of the facts regarding the co--ercial access available to the T3 backbone, and (2) a mistaken perception that the agency was endeavoring to keep its actions from the public. 37 |--------------------------------------------------------------------| D. Acceptable Use Policy Under Section 3(a)(4) of the National Science Foundation Act of 1950, as amended (the "NSF Act"), NSF was given the authority "to foster and support the development and use of computer and other scientific and engineering methods and technologies, **primarily for research and education in the sciences and engineering**."71 In 1989, NSF drafted an Acceptable Use Policy ("AUP") to define research and education traffic that may properly be conveyed over NSFNET under Section 3(a) of the NSF Act. This policy has undergone a number of revisions but remains in force; a recent version is attached as an appendix to this Report. In March 1992, NSF's Office of General Counsel concluded that "some form of acceptable use policy will continue to be necessary to ensure that NSF funds are used to further the objectives of Section 3(a)(4) of the Act."72 The Scientific and Advanced Technology Act of 1992, introduced by Congressman Boucher and signed into law on 23 October 1992, subtly modified NSF's authority to support computer networks that are not limited to research and education. This statute added the following new subsection to Section 3 of the NSF Act: "(g) In carrying out subsection (a)(4), the Foundation is authorized to **foster and support access by the research and education co--unities** to computer networks which may be used substantially for purposes in addition to research and education in the sciences and engineering, **if the additional uses will tend to ____________________ 71 42 U.S.C. 1862(a)(4) (emphasis added). 72 Memorandum dated 16 March 1992 from NSF Office of General Counsel to DNCRI. 38 |--------------------------------------------------------------------| increase the overall capabilities of the networks to support such research and education activities**."73 Thus, NSF is now authorized to support **research and education access** to networks that are used primarily for co--ercial purposes, **but only if** allowing co--ercial use will enhance the networks' utility **for research and education**. The legislative history of subsection (g) also emphasizes the principle that NSF support must **directly** benefit research and education. In the report accompanying the House version of the bill, the House Co--ittee on Science, Space, and Technology stated: "The Co--ittee intends to provide increased flexibility to NSF in formulating policies for NSFNet and the interim NREN. The Co--ittee does not intend in any way to alter the Foundation's principal role relative to networking activities, which is to develop and use computer networks for support of research and education in the sciences and engineering. . . . In exercising the authority provided by H.R. 5344, NSF is expected to institute policies which result in increased capacity for computer networks supported by NSF **and improved service for network users in the research and education co--unities** at the lowest possible cost to the Federal Government."74 Therefore, in our view an AUP is necessary, now even more than before, to define what constitutes research and education traffic, because NSF is authorized only to provide funding that supports access by that traffic to the networks that NSF chooses to fund.75 ____________________ 73 42 U.S.C. 1862(g) (emphasis added). 74 H.R. Rep. No. 567, 102d Cong., 2d Sess. 4 (1992). 75 By "AUP," we mean a stated policy defining what constitutes research and education traffic, and not necessarily the current version of the NSFNET AUP (set out in the Appendix). In our view the current NSFNET AUP is reasonable, but it can certainly be revised and remain consistent with the NSF Act. 39 |--------------------------------------------------------------------| For policing compliance with the AUP, NSF currently relies on users to avoid sending their co--ercial traffic over NSFNET. It is not practical to monitor network traffic for classification as acceptable or not under the AUP. In the current AUP, DNCRI reserves the right to "resolve any questions about this Policy or its interpretation." While there is no evidence of substantial abuse under the current system, as use of the network grows, so will the likelihood of serious abuse. This reinforces our conclusion that further steps must be taken to ensure that NSF is not exceeding its statutory mandate. The AUP must be applicable to all users of the NSF-funded network. We do not understand how NSF could enforce the AUP against an uncooperative end-user if, for example, the end-user's network also refused to cooperate, because we do not see how the AUP is legally binding on end-users. The AUP is not even part of the award conditions enforceable against Merit, and NSF generally has no direct relationship to other networks connected to NSFNET or to end-users that would facilitate enforcement. In addition, if the AUP is intended to apply to all users of NSFNET, we believe the failure to publish the AUP in the Federal Register and the failure to provide the public with notice and an opportunity for co--ent are inconsistent with the Administrative Procedure Act.76 ____________________ 76 The Administrative Procedure Act requires that each agency publish "substantive rules of general applicability adopted as authorized by law, and statements of general policy or interpretations of general applicability formulated and adopted by the agency . . . ." 5 U.S.C. 552(a)(1)(D). The Supreme Court has noted that one of the purposes of the publication requirements in the Administrative Procedure Act is "to avoid the inherently arbitrary nature of unpublished ad hoc (continued...) 40 |--------------------------------------------------------------------| The Administrative Procedure Act promotes the involvement of the public in government policymaking. Providing the public with notice and an opportunity to co--ent before promulgating the AUP as a regulation will give the affected public -- particularly the networking co--unity -- an opportunity to have its views considered on this important matter of agency policy which may significantly affect their activities. Accordingly, we reco--end that NSF continue to maintain an AUP that will ensure that NSF's support for access to the network is consistent with the NSF Act. We reco--end that the AUP be promulgated as a regulation, after receipt and analysis of co--ents from the affected public.77 Promulgation of the AUP as a regulation will ensure not only that the views of the public are considered in a formal and effective manner, but it will also help NSF take action against abuses, including preventing future access to the network and debarment in extreme cases.78 We also reco--end that the AUP be made a part of the award conditions both for Merit as well as for any organization that later receives NSF funding pursuant to the proposed new ____________________ 76 (...continued) determinations." Morton v. Ruiz, 415 U.S. 199, 232 (1974). For the AUP to be legally applicable to the users of NSFNET, notice and a co--ent period is required by the Administrative Procedure Act. 5 U.S.C. 553. 77 NSF is authorized by section 11(a) of the NSF Act (42 U.S.C. 1870(a)) to promulgate a regulation to codify the AUP. 78 NSF's nonprocurement debarment regulation (at 45 C.F.R. 620.305(b)(3)) permits debarment for "[a] willful violation of a statutory or regulatory provision or requirement applicable to a public agreement or transaction." NSF will be better able to use this enforcement tool if the AUP is adopted as a regulation that specifically states that a willful violation of the AUP is a ground for debarment. 41 |--------------------------------------------------------------------| solicitation. In addition, NSF should ensure that end-users are aware of the provisions in the AUP and the sanctions that may be imposed for violations. Therefore, we reco--end that NSF disseminate the AUP in a manner such that all end-users are aware of its prohibitions. IV. Administrative Issues A. Cooperative Agreement or Contract? The legal funding instrument used by NSF for the creation of the NSFNET backbone was a cooperative agreement, not a contract. Under the Federal Grant and Cooperative Agreement Act of 1977, contracts are the appropriate legal funding instrument to be used by the government "whenever the principal purpose of the instrument is the acquisition . . . of property or services for the direct benefit or use of the Federal Government . . . ."79 In contrast, cooperative agreements are to be used "whenever -- (1) the principal purpose of the relationship is the transfer of money, property, services, or anything of value to the . . . recipient to accomplish a public purpose of support or stimulation authorized by Federal statute, rather than acquisition . . . of property or services for the direct benefit or use of the Federal Government; **and** (2) substantial involvement is anticipated between the executive agency, acting for the Federal Government, and the . . . recipient during performance of the contemplated activity."80 ____________________ 79 41 U.S.C. 503. 80 41 U.S.C. 505 (emphasis added). 42 |--------------------------------------------------------------------| NSF sought an awardee to work with NSF to create a computer network on an unprecedented scale, thereby benefiting the scientific research and education co--unity. Because of the nascent development of large-scale networking, and the anticipated -- but unpredictable -- technical and logistical problems, NSF and Merit expected substantial involvement by NSF in the evolution of the network. In fact, NSF staff and Merit/ANS have maintained close contact throughout the evolution of the network. Thus, in our view the NSFNET program is a paradigm for the use of a cooperative agreement. We also believe the use of a cooperative agreement for NSFNET's next stage of evolution is appropriate.81 B. Conditions of the Award Initially, a "Letter Cooperative Agreement" was entered into by Merit and NSF.82 This letter subjected the agreement to the Grant General Conditions and Cooperative Agreement General Conditions, as well as several special conditions. The first amendment to the Letter Cooperative Agreement consisted of a "Cooperative Agreement," which superseded the Letter Cooperative Agreement.83 This new Cooperative Agreement was also subject to the same General Conditions, as well as special conditions, some of which limited the applicability of the General ____________________ 81 The DOE/NASA award discussed in Section V.C. was a **contract** for provision of a network to connect primarily government research laboratories; thus, those agencies -- unlike NSF with NSFNET -- are "acquir[ing] . . . services for the direct benefit or use of the Federal Government," making a contract more appropriate for that undertaking. 82 Letter from DGC to Merit (19 November 1987). 83 Cooperative Agreement (15 December 1987). 43 |--------------------------------------------------------------------| Conditions. However, in Amendment Number 2 to the Cooperative Agreement, dated 27 March 1989, NSF stated: "Pursuant to NSF Important Notice No. 104, 'Changes in NSF Proposal Processing and Grant Administration Requirements,' dated September 22, 1988, the NSF FL200 Grant General Conditions (10/87) and the NSF FL175 Cooperative Agreement General Conditions (11/87), to which the referenced Cooperative Agreement is subject, have been superseded by the attached FDP-II Federal Demonstration Project General Terms and Conditions (October 1988) and the attached NSF CA-1 Cooperative Agreement General Conditions (October 1988)." This is odd, because Important Notice Number 104 merely stated that all NSF grants would now be subject to the new Grant General Conditions; it did not mention the Federal Demonstration Project General Terms and Conditions or the NSF Cooperative Agreement General Conditions. There is no documentation in the files as to why the General Conditions of the Cooperative Agreement were changed. The reasons for including this Cooperative Agreement in the Federal Demonstration Project ("FDP") are unclear. The Federal Demonstration Project institutions were chosen via a solicitation issued by NSF and other federal agencies participating in the FDP. The solicitation sought organizations wishing to participate in the FDP, which would "eliminate unnecessary administrative burdens on sponsored research thereby enhancing research productivity."84 According to an information sheet provided to us by the Division of Grants and Contracts ("DGC"), the FDP "eliminates most of the current requirements for Federal prior approval . . . as long as pertinent grantee administrative systems are adequate." We were ____________________ 84 53 Fed. Reg. 20,697 (6 June 1988). 44 |--------------------------------------------------------------------| told by DGC that FDP terms should only be applied when the organization is an FDP institution. Merit is not an FDP institution, but the University of Michigan is. Under an agreement between Merit and the University of Michigan, the University "shall provide to Merit: (a) facilities; (b) personnel; and (c) financial and administrative services; as required by Merit to implement and manage NSFNET . . . ." The mere contractual arrangement between Merit and the University of Michigan is not a sufficiently large part of the Cooperative Agreement to justify treating Merit as an FDP institution. Therefore, inclusion of the FDP General Terms and Conditions as the terms of the NSFNET Cooperative Agreement was inappropriate.85 We reco--end that, in the future, NSF apply the Federal Demonstration Project conditions only to awards where the organization receiving the grant is an FDP institution. C. Prior Approval Before the expansion of NSFNET and conversion to T3, in September 1990, Merit, IBM, and MCI formed ANS to take over management and operation of NSFNET. ANS was then assigned all of Merit's rights and responsibilities under the NSFNET Cooperative Agreement. Generally, prior approval for such an assignment is required under the general conditions of the award; however, there is no prior approval requirement in the Federal ____________________ 85 No one who was responsible for the amendment is still employed by the Division of Grants and Contracts. Because we do not believe that NSF was injured by the application of the FDP terms and conditions to the NSFNET Cooperative Agreement (as discussed in Section IV.C., the special conditions of the award still required prior approval of assignments), we did not undertake to locate and interview the individuals who were responsible for including this Cooperative Agreement in the FDP. 45 |--------------------------------------------------------------------| Demonstration Project General Terms and Conditions, to which the Cooperative Agreement was (albeit erroneously) subject.86 On the other hand, the Cooperative Agreement's Special Conditions (in E.3.) do require that, "[p]rior to execution, any agreements between the awardee and other participants for the purpose of activities of mutual benefit must receive the NSF Grants and Contracts Officer's written approval." The term "other participants" is not defined in the agreement; presumably, the "other participants" are MCI and IBM. In our view, since this requirement is included in the special conditions pertaining to this agreement and these special conditions have not been superseded, any documents involved in the creation of ANS by Merit, MCI, and IBM, would be subject to prior approval by the Division of Grants and Contracts ("DGC"). Furthermore, it is our view that ANS, as a creation of Merit, MCI, and IBM, is also a "participant", so that the Cooperative Agreement independently required DGC to approve any agreement entered into between Merit and ANS. There is no record in DGC's files of DGC approval having been sought or obtained for anything having to do with the assignment to ANS; nor is there anything in the program files. In a "final draft of Merit's fourth year NSFNET project final report," Merit said that "NSF approved the subcontracting of backbone services for NSFNET from ANS in a letter dated ____________________ 86 There is a prior approval requirement in the Cooperative Agreement **General** Conditions, to which the Cooperative Agreement was also subject. However, the special conditions deleted the article in the Cooperative Agreement General Conditions that dealt with prior approval. 46 |--------------------------------------------------------------------| 10/17/90 . . . ."87 When Merit was asked to provide a copy of this letter, it was actually dated 10 September 1990. The letter, from the Division Director of DNCRI -- not DGC -- stated: "NSF agrees to MERIT's assigning the existing joint studies with MCI and IBM to [ANS]. "NSF agrees to MERIT's subcontracting services to [ANS]. Although [ANS] may thus become the actual provider of the NSFNET Backbone, NSF will continue to deal with MERIT under the terms of the existing Cooperative Agreement. "**The specific concurrence of NSF's Division of Grants and Contracts has been obtained in the above two paragraphs**."88 Thus, although there is no record of it in either the program's or DGC's files, apparently the program and DGC both approved the creation of, and assignment to, ANS. Important decisions such as this one should have been thoroughly documented by both the program and DGC, in consultation with NSF's Office of General Counsel ("OGC"). We reco--end that NSF ensure that DNCRI and DGC more thoroughly document their decisionmaking processes and ensure that **all** relevant records are included in the program and DGC files.89 ____________________ 87 Final Draft of Merit's Fourth Year NSFNET Project Final Report at 6. 88 Letter from NSF to Merit (10 September 1990) (emphasis added). 89 The Division of Grants and Contracts has recently undertaken to correct the lack of DGC formal, written approval for the Merit/ANS "subcontract". Amendment No. 11 to the Cooperative Agreement (17 November 1992) states: "It has come to our attention that a substantive portion of the effort has been subcontracted without official NSF approval. Article II.E.3 of the Special Conditions for this Cooperative Agreement states the following: (continued...) 47 |--------------------------------------------------------------------| NSF's General Counsel was consulted about the AUP (see Section III.D.); however, the Office of General Counsel ("OGC") was not included in other significant administrative decisions discussed herein. Because these issues can be legally complex, we reco--end that NSF ensure that major program decisions on NSFNET always include documented consultation with the Office of General Counsel, and, when appropriate, legal opinions should be obtained from OGC in writing. We also reco--end that NSF ensure that a prior approval clause is included in any future awards involving the approval of the National Science Board90 that are subject to the Federal Demonstration Project, and that this prior approval clause require written approval from the Office of General Counsel as well as from the Division of Grants and Contracts. ____________________ 89 (...continued) **Prior to execution, any agreements between the awardee and other participants for the purposes of activities of mutual benefit must receive the NSF Grants and Contracts Officer's written approval.** "In accordance with this requirement, your organization is hereby required to submit a formal request for subcontracting by November 30, 1992. The signed request should include the proposed performance statement and budget, a statement indicating the basis for selection of the contractor, and a justification of the proposed arrangement. The request must be submitted to [the] NSF Grants and Contracts Officer and is not considered official until you receive her written approval. Additionally, any further proposals which contain substantive subcontracts must contain subcontracting budgets." (Emphasis in original). 90 In general, the National Science Board approves all awards involving total amounts exceeding six million dollars and all awards involving greater than one and a half million dollars in one year. NSB-91-69. 48 |--------------------------------------------------------------------| D. Documentation In addition to the lack of records on the prior approval of the assignment, reports required by the Cooperative Agreement were not in the program files. In the First Amendment to the Letter Cooperative Agreement, dated 15 December 1987, Merit was required to submit monthly performance reports, quarterly reports, and annual reports. While some of the required reports were included in the program files, no monthly performance reports were included, most of the quarterly reports were missing, and some of the annual reports were omitted from the files. Other documentation was also omitted from the program files. For example, NSF staff frequently co--unicate with Merit/ANS via electronic mail, but no hard copies of such mail were included in the program files.91 Moreover, agreements between NSF and Merit dealing with co--ercial use of the network were not included in the program files. When those items were requested by a co--ercial network provider in October 1991,92 NSF had to obtain them by FAX from Merit; when we sought them in the course of this review, less than a year later, NSF had again misplaced them: we obtained one from the co--ercial network provider that had previously obtained it from NSF, another from a DNCRI staff member's e-mail archive file, and another by FAX from Merit. ____________________ 91 See Section IV.F. 92 See Section III.C.5. 49 |--------------------------------------------------------------------| This sort of disorganization of program files is simply unacceptable. We reco--end that NSF ensure that **all** relevant documents, especially reports required by the Cooperative Agreement and agreements relevant to the award, are included in the program files. E. Compliance with OMB Circular A-110 The agreement between Merit and ANS refers to itself as a "subcontract". If the agreement is a contract or subcontract for the purchase of goods or services, Attachment O of OMB Circular A-110 applies. Attachment O includes two provisions that are particularly relevant to the arrangement between Merit and ANS. Attachment O prohibits conflicts of interests in contracting by federal grantees: ". . . No employee, officer, or agent shall participate in the selection, award, or administration of a contract in which Federal funds are used, where, to his knowledge, he or his . . . organization . . . has a financial interest . . . ."93 Awardees are also required by Attachment O to maximize competition in spending Federal funds: "All procurement transactions shall be conducted in a manner to provide, to the maximum extent practical, open and free competition . . . ."94 There is no problem with Merit, MCI, and IBM creating an independent entity such as ANS. The difficulty arises from the assignment to ANS of the NSFNET networking services, because subcontracts for services under a cooperative agreement subject to the terms of Attachment O must comply with Attachment O's ____________________ 93 Paragraph 3.a. 94 Paragraph 3.b. Numerous procedural requirements are also set out in paragraph 3.c. 50 |--------------------------------------------------------------------| conflict of interest and competition requirements. As described above, the Merit-ANS arrangement would clearly violate those provisions of Attachment O, were they applicable: Merit clearly had an interest in ANS,95 and there was no effort whatever to seek competition. In our view, however, the applicability of Attachment O should be determined by the true nature of the legal relationship between the parties, and not by the label that the parties apply to that relationship. We are convinced that Merit and ANS did not accurately characterize their relationship when they called it a subcontract. Merit did not simply promise to pay money to ANS in exchange for the provision of specifically identified network services. Rather, ANS and Merit were to work very closely together to run, develop, and improve NSFNET, in a manner analogous to the arrangement between NSF and Merit. The legal relationship between NSF and Merit was quite properly a cooperative agreement, because NSFNET was intended for the benefit of the public and was going to be implemented and enhanced with frequent and intimate interaction between the two parties.96 What Merit assigned to ANS was a substantial portion of its responsibilities under the Cooperative Agreement with NSF. Thus, in our view, the relationship between Merit and ANS was really a sub-cooperative agreement rather than a subcontract, and it was therefore not a "contract for the ____________________ 95 Merit, together with IBM and MCI, created ANS, and Merit, IBM, and MCI were each represented on ANS's Board of Directors. 96 See Section IV.A. 51 |--------------------------------------------------------------------| purchase of goods or services" subject to the requirements of Attachment O. F. Accessibility We are especially impressed by the availability of the Division of Networking and Co--unications Research and Infrastructure ("DNCRI") to the public via participation in an electronic bulletin board that is widely read by the co--unity of users of NSFNET.97 NSF and Merit/ANS make announcements of widespread interest on the bulletin board, but more important, network users who have a question or complaint can post it to the bulletin board, **and representatives from Merit or ANS or NSF will usually respond**, often on the same day. Often there is an exchange of views among NSF, the NSFNET provider (Merit/ANS), and the users of the network that is truly remarkable in a federal program. We applaud this accessibility by DNCRI, and encourage its continuation by DNCRI and emulation by other NSF programs.98 On the other hand, none of these electronic bulletin board exchanges was documented in the program files. We reco--end that, to the extent that NSF or NSFNET policy is expressed in postings in electronic or any other media, NSF ensure that hard ____________________ 97 The bulletin board is "com-priv@psi.com", where users post questions and co--ents on the network, its operation, and its uses. 98 But see Section III.C.5. 52 |--------------------------------------------------------------------| copies (and the message(s) to which they are directly responding) are included in the program file.99 G. NSF Funding of the Provider or the Users? We believe that the funding approach taken by NSF -- supporting the creation of the NSFNET backbone directly, rather than giving grants to the users of the network in the hope that the availability of buyers would inspire network providers to enter the market -- was reasonable. When NSFNET came on-line at T1 it was pushing the envelope of large-scale networking technology, and the same was true when it advanced to T3. This is a field that is rediscovering the scope of its usefulness virtually on a daily basis, **as it is used**, which in turn inspires more use. The resultant strain on the NSFNET backbone's capacity culminated in the various expansions and enhancements implemented by NSF and Merit. We are convinced that relying on demand-driven market forces would have advanced neither networking technology nor use as rapidly as both have advanced under the direct funding approach taken by NSF. In its proposed ____________________ 99 While the maintenance of a single electronic mail archive that contains all electronic mail relating to a program is a superficially attractive idea, we do not believe it would work well in practice. The program would have to ensure that each message was referred to at the appropriate place in the program file (program files are generally divided into a few substantive sections, with material arranged chronologically within each section) so that a person reviewing the file would be aware of a related message in the electronic mail archive. Because including a sheet of paper for each message would generally require the same amount of effort and space as including the messages themselves in the program file, while requiring significant additional effort on behalf of a person reviewing the file, we believe the program staff should ensure that hard copies of these messages are included in the program file. In the future, if program files are maintained completely electronically, this requirement will no longer be necessary. 53 |--------------------------------------------------------------------| solicitation for the next phase of NSFNET/NREN, NSF makes it clear that it is going to phase itself out of direct funding, which we also view as a reasonable decision. As the use of networks catches on -- particularly with the uses possible only with T3 speeds and beyond -- demand can take over to sustain and expand the market. The government's focus can then turn to expanding the breadth of the co--unity of users of the network. V. The Future of NSFNET A. The Proposed Solicitation After completing consultation with the co--unity about the continuation of NSFNET backbone services and preparing the Project Development Plan for the implementation of continued services, the NSF staff requested an extension of the Merit Cooperative Agreement for eighteen months so that NSF would have a sufficient amount of time to solicit and evaluate bids and so that the awardee would have a sufficient amount of time to start its operation without an interruption of service. This eighteen month award involved additional spending at current levels, not to exceed $15 million over the eighteen month period of the extension. We believe that the extension was reasonable, particularly since it will provide ample time to analyze the co--ents received in response to the published proposed solicitation.100 ____________________ 100 Merit has administered the Cooperative Agreement for the five year term of the original award, and it will continue to administer the agreement during the eighteen month extension. If NSF determines that any substantial (that is, more than three months) additional extension of the award is necessary, in our view the extension should be made via a new solicitation, allowing free and open competition. 54 |--------------------------------------------------------------------| On 15 June 1992, NSF published a proposed solicitation for a "Network Access Point (NAP) Manager and Routing Authority (RA) organization; and a provider of very high speed Backbone Network Services (vBNS)".101 This proposed solicitation sets forth a new architecture for providing NSFNET backbone services, including regional networks and network service providers.102 The proposed solicitation anticipates total NSF funding of about $10 million per year for 100 mbps on the NAPs and 155 mbps on the vBNS (compared to roughly the same cost per year currently for the T3 (45 mbps) backbone). In the draft solicitation, a NAP is defined as "a high speed network or switch to which a number of routers can be connected for the purpose of traffic exchange and interoperation."103 NAPs are locations where other networks can connect to the vBNS. NAPs must be able to operate at speeds of at least 100 mbps, and networks that attach to NAPs must operate at speeds of at least 1.5 mbps (T1). The NAP Manager/RA will be responsible for arranging for and maintaining the NAPs and routing information for NSFNET. The NAP Manager/RA will also provide for the interconnection of appropriate networks under circumstances not subject to the ____________________ 101 Public Draft at 26,692. 102 "It is anticipated that networks other than the VBNS will connect to the NAPs. Examples of such networks include: other federally sponsored networks; other service providers for research and education; service providers for traffic which is not limited to the support of research and education; and international networks." Public Draft at 26,695. 103 Public Draft at 26,693. 55 |--------------------------------------------------------------------| AUP.104 In essence, the NAP Manager/RA will assume the duties for which Merit and ANS are currently responsible (with contributions from IBM). Networks would pay attachment fees to connect to the NAPs. These attachment fees will consist of an initial and an annual fee and will be determined by the NAP Manager/RA and NSF such that a portion of the costs associated with the NAPs and the NAP Manager/RA will be recovered.105 The vBNS Provider is responsible for establishing and maintaining a vBNS to accept traffic from the NAPs that complies with the AUP. The vBNS Provider is also required to provide for high speed connectivity between regions; facilitate distributed computing applications; facilitate multimedia applications; and promote development and utilization of advanced routing technologies.106 In essence, the vBNS Provider will undertake the provision of the services which MCI currently provides for ANS, at a substantially higher speed. Although vBNS NSFNET/NREN traffic must comply with the AUP, "[t]he vBNS may have connections and customers beyond those specified by NSF as long as the quality and quantity of required services for NSF-specified customers are not affected."107 The NAPs will not be subject to the AUP and "will permit, for example, two attached networks to exchange traffic without violating the AUPs of any other attached network."108 ____________________ 104 Public Draft at 26,692. 105 Public Draft at 26,695. 106 Public Draft at 26,692. 107 Public Draft at 26,694. 108 Public Draft at 26,693. 56 |--------------------------------------------------------------------| Therefore, any co--ercial carrier may attach to a NAP and provide services to others that are attached to that NAP, including other co--ercial users. NSF will support current NSFNET midlevel networks as follows: "NSFNET midlevel networks may connect either to network services providers which are connected to NAPs or directly to NAPs. NSF will support for one year a single such connection fee for midlevel networks which are currently connected to the NSFNET Backbone Network Service. In the following years of the cooperative agreement, NSF support for the connection fee will decrease and the midlevel network will need to increase other sources of support correspondingly. NSF support of this fee will cease after the (first) term of the NAP Manager/RA Cooperative Agreement."109 In short, networks other than the vBNS can attach to the NAPs by paying a fee to the NAP Manager/RA for the provision of the NAP. The vBNS will be attached to each NAP and will provide services in support of research and education. B. Co--ents on the Proposed Solicitation In our view, the draft solicitation was a laudable means of polling the affected public of network users. In a field that is evolving as quickly as this one, seeking the insights of the user co--unity is especially beneficial. In our view, it is reasonable to allow the vBNS Provider to have non-NSF-specified connections and non-NSF-specified customers with traffic that is not subject to the AUP (as long as the quality and quantity of services to the research and education traffic is not decreased), since NSF may reasonably anticipate that by doing so it will receive superior service at ____________________ 109 Public Draft at 26,695. 57 |--------------------------------------------------------------------| substantially less cost to the government.110 The federally-supported vBNS provider may, when the vBNS is running with sufficient reliability, have a competitive advantage over other network providers for customers desiring very high speeds. In order to leverage its funds, the government may provide additional advantages to awardees as an incentive to invest substantial amounts of their private funds. Since the draft solicitation explicitly includes this potential competitive advantage, the public is on notice that the successful offeror will potentially gain this competitive advantage. In these circumstances, the provision of such an advantage is a reasonable means of obtaining substantial benefits for the research and education co--unity at minimal cost to the federal government. However, the draft solicitation provides that the NAPs are not subject to the AUP, and it does not give research and education traffic any priority over co--ercial traffic. Therefore, it is possible that the NAPs could become overwhelmed with co--ercial traffic to the detriment of research and education traffic. It seems to us that permitting co--ercial traffic to use the NAPs in a manner that could be detrimental to the use of the network by research and education traffic may not be consistent with the NSF Act. We reco--end that NSF consider permitting the NAPs to handle co--ercial traffic only under the same condition as the vBNS -- that is, "as long as the quality ____________________ 110 Indeed, it is likely that the only offerors that could hope to win the vBNS award would already have in place a working network of long-distance lines, for primarily (if not exclusively) co--ercial purposes. They will be interested in the vBNS solicitation only if they anticipate having sufficient excess capacity to fill the needs of NSFNET. 58 |--------------------------------------------------------------------| and quantity of required services for NSF-specified customers are not affected."111 If NSF decides to permit co--ercial traffic to use the NAPs without giving the research and education traffic priority (as proposed in the draft solicitation), we reco--end that it obtain a legal opinion from its Office of General Counsel stating that such a decision is consistent with the NSF Act.112 Some regional networks charge lower fees to research and education customers than to co--ercial customers, thereby subsidizing the research and education customers with revenues derived from co--ercial sources.113 In a similar vein, we reco--end that NSF consider whether it is desirable and practical to require the NSFNET provider(s) to charge co--ercial traffic a higher rate than research and education traffic. In this way, NSF may be able to provide additional support to the research and education co--unity -- subsidized rates -- thus leveraging its funds further. In our view, allowing co--ercial traffic on the NSFNET (both the vBNS and the NAPs) only to the extent that it does not interfere with research and education traffic would also simplify ____________________ 111 Public Draft at 26,694. The AUP has always defined the priority in terms of traffic rather than in terms of users, which we believe is appropriate. For example, the same user could send traffic that is subject to the AUP -- e.g., a professor sending a request for material for one of her classes -- and also send packets that are not compatible with the AUP -- e.g., the same professor accessing the Dow Jones service to determine whether her stock portfolio has increased in value. 112 See Section III.D. 113 "The National Research and Education Network Program," p. 17, Report to Congress, Office of Science and Technology Policy (December 1992). 59 |--------------------------------------------------------------------| the AUP issue from NSF's perspective, because the NSFNET provider(s) would have an incentive to ensure that NSF is not subsidizing co--ercial traffic: the provider(s) could charge co--ercial traffic for access to the vBNS that research and education users get free, and could charge co--ercial traffic a higher rate for use of the NAPs. The NSFNET providers would thus have their own financial incentives to distinguish types of traffic. As a result of the requirement that research and education traffic always have priority, and the financial incentives provided by the co--ercial users, the NAP Manager/RA and the vBNS Provider would presumably enhance and expand the network facilities as demand warrants. Of course, if NSF requires the NSFNET provider(s) to charge higher rates for co--ercial traffic, the providers' incentives would clearly be to maximize their income by deciding that any given user's traffic is co--ercial, and thus subject to vBNS charges and higher NAP charges. NSF would thus still have a role to play in mediating disputes that the users and providers are not able to resolve amicably, and would have to reserve the ultimate authority to decide such disputes.114 In addition, we are concerned about continuity. Because the new network will be pushing technology beyond current limits and therefore unforeseeable problems are sure to arise, we reco--end that NSF consider whether it should require that the vBNS Provider and NAP Manager/RA provide an adequate back-up to the vBNS and the NAPs, at least until such time as the new network is proven reliable. In this way, most traffic (other than ____________________ 114 See Section III.D. 60 |--------------------------------------------------------------------| applications requiring very high speeds) will be able to continue with minimal disruption caused by problems with the new technology. C. Interagency Coordination of the NREN 1. The New NASA/DOE Network In February 1991, the National Aeronautics and Space Administration ("NASA") and the Department of Energy ("DOE") issued a request for proposals for provision of a network that will primarily serve to connect laboratories conducting research for those agencies. The network will operate at T3 in 1993, 155 mbps in 1994, and 622 mbps in 1995. In August 1992 NASA/DOE announced their intention to award a contract.115 Under the contract, the network will cost $50 million over five years at T3, with the cost of the higher speeds to be determined later. The first phase of the NASA/DOE contract will have the same speed objective that NSFNET has already obtained; the second phase of the NASA/DOE contract will have the same speed objective as the NSFNET re-solicitation (155 mbps).116 Such redundancy is not necessarily deleterious: government agencies occasionally have awarded similar contracts to two different companies with the expectation that the best product will emerge from two ____________________ 115 NASA/DOE have not yet entered into a contract because the award is currently under protest. 116 The costs of the NASA/DOE and NSF networks are not directly comparable, however, because NASA/DOE are seeking provision of a "Layer 2" network, for which NASA/DOE will themselves be providing routing and administration services. The NSFNET is and will continue to be a "Layer 3" network, with routing and administrative services provided by the awardee(s), not NSF. 61 |--------------------------------------------------------------------| independent product development efforts. The government could take this approach to the federal funding of two national high-speed computer networks. 2. The High Performance Computing Act The High Performance Computing Act of 1991 (the "HPCA")117 was signed into law in December 1991, at a time when NASA/DOE were finalizing the solicitation for their network. One of the stated purposes of the HPCA is "improving the interagency planning and coordination of Federal research and development on high-performance computing and maximizing the effectiveness of the Federal Government's high-performance computing efforts."118 By seeking to improve interagency coordination, the HPCA is designed to ensure against an ad hoc, every-agency-for-itself approach toward high-performance computing and its applications.119 In furtherance of this statutory mandate, the participating NREN agencies "have established a Charter and worked to define their respective roles."120 The HPCA specifically requires the National Science Foundation to "upgrade the National Science Foundation funded network, assist regional networks to upgrade their capabilities, and provide other Federal departments and ____________________ 117 15 U.S.C. 5501-5528. 118 15 U.S.C. 5502(2). 119 S. Rep. No. 57, 102d Cong., 2d Sess. 17 (1991), reprinted in 1991 U.S.C.C.A.N. 1228, 1251. 120 NREN Report to Congress prepared by the Office of Science and Technology Policy (1992) at 35 (hereafter, "NREN Report"). 62 |--------------------------------------------------------------------| agencies the opportunity to connect to the National Science Foundation funded network."121 Accordingly, NSF has been assigned primary responsibility for "coordinat[ing] the Interagency Interim NREN activities, including coordinating the development, deployment, and operations of the Interagency Interim NREN facilities and services . . . ."122 Under this coordination framework, NASA and DOE are expected to upgrade their existing networks, but only "as part of the coordinated Interagency Interim NREN Program"123 -- which NSF has primary responsibility to coordinate.124 The NREN agencies coordinate through the Office of Science and Technology Policy's Federal Coordinating Council for Science, Engineering, and Technology ("FCCSET") and the Federal Networking Council ("FNC").125 ____________________ 121 15 U.S.C. 5521(a)(4). 122 NREN Report at 37. 123 NREN Report at 38. 124 As explained above, the HPCA expressly requires **NSF** to take the lead in the development of a high speed national network for the NREN, to which other agencies may attach; by contrast, neither DOE nor NASA has primary responsibility under the statute for providing a national network. DOE is charged to "perform research and development on, and systems evaluations of, high-performance computing and co--unications systems" and to "provide for networking infrastructure support for energy-related activities." 15 U.S.C. 5523(a)(1), -(4). There is no reference to co--unications or networking activities by NASA in the statute. 15 U.S.C. 5522. The legislative history is also clear on this point: "NSF will coordinate deployment of the NREN, accelerating the harmonizing of multiple agency networks and protocols into a single NREN." S. Rep. No. 57, 102d Cong., 2d Sess. 17 (1991), reprinted in 1991 U.S.C.C.A.N. 1228, 1245 (NASA is not mentioned in the report with regard to the NREN). 125 The Federal Networking Council consists of representatives from interested federal agencies. It coordinates the activities of the NREN agencies while acting as a liaison to persons interested in the program. 63 |--------------------------------------------------------------------| 3. Coordination of the New NASA/DOE Network The HPCA was enacted at a time when preparation of the request for proposals for the NASA/DOE network was very far advanced, and FCCSET finalized its formal scheme for coordination among the NREN agencies only relatively recently. Although NASA/DOE did not formally coordinate the procurement of their new high-speed network through FCCSET and FNC,126 it is our view that the implementation of the NASA/DOE network can still be effectively coordinated with NSFNET.127 We believe the government and the public will benefit if FCCSET and FNC analyze potential cost savings that could result from integrating the NSF and NASA/DOE network undertakings -- and ____________________ 126 Some technical issues related to the overlap of the NASA/DOE network with the next stage of NSFNET were discussed by members of the Federal Engineering Planning Group ("FEPG"), which is a subco--ittee of FNC. The FEPG is composed of people involved in the technical engineering aspects of the implementation of the NREN agencies' currently operating networks. As a result, its meetings are regularly attended by a representative of Merit, but not by anyone from NSF. Thus, since NASA and DOE operate their own networks, attendees intimately familiar with the proposed NASA/DOE network were present, but no one involved in the proposed NSFNET solicitation was present. In addition, a Merit representative would be expected to represent Merit's views on the network but would not necessarily state NSF's views on network issues. We are also aware that the NREN agencies are in frequent contact with each other, through FCCSET as well as through informal interactions that predate the NREN. This frequent contact certainly furthers the goals of the High Performance Computing Act. Because of the NREN agencies' frequent contact, NASA/DOE may have felt no need to discuss the implementation of their network because they were confident that it would not dovetail with NSFNET. 127 Coordination is still possible because the final terms of the NASA/DOE contract have not been finalized, and in any event such a contract could probably be readily amended to acco--odate changes deemed desirable through the FCCSET and FNC process. 64 |--------------------------------------------------------------------| thereby possibly having fewer nodes, simpler routing tables, and/or shared administrative costs.128 In our view, the NREN agencies, through the FCCSET and FNC processes, should: (1) explicitly consider and formally endorse or reject the creation of both NSF and NASA/DOE networks and (2) evaluate possible cost savings associated with integrating the NSF and NASA/DOE network undertakings. The pressure to reduce government spending is only going to get greater as the NREN progresses, so it is incumbent on the NREN agencies to work closely together to achieve maximum efficiency. If the government is going to spend $100 million to create two national networks when there are **potentially** overlapping requirements and duplicative costs, it should do so only after careful consultation among the NREN agencies as coordinated by FCCSET and FNC. 4. A New Paradigm We are well aware that a certain amount of inter-agency competitiveness has traditionally existed, with some agencies trying to "scoop" the others in cutting edge fields such as high-speed computer networking, but it is simply no longer acceptable. The High Performance Computing Act, as well as ____________________ 128 The potential cost savings may in fact not be substantial. As we understand it, the cost of these high speed networks is to a great extent proportional to the number of nodes. The NASA and DOE laboratories need to be able to co--unicate with each other on secure isolated "Virtual Private Networks", which may or may not be able to co--unicate with each other and with which outside parties cannot co--unicate. By contrast, NSFNET's NAPs will be open to all research and education traffic. Although a single provider could serve NASA/DOE's needs as well as NSF's, approximately the same number of network connections may be required even if NASA and DOE combined with NSF to fund a network meeting the needs of all three agencies' user co--unities, without substantial cost savings. 65 |--------------------------------------------------------------------| co--on sense in a period of increasing pressure for budgetary belt-tightening, mandates that every agency's high-performance network undertakings be coordinated with other relevant agencies. The HPCA, as well as the establishment of FCCSET and FNC, reflect a paradigm shift for federal government decisionmaking: government agencies must fundamentally alter their approach to large-budget projects such as the NREN that involve multiple agencies, actively seeking cost savings and efficiencies through coordination and cooperation. Using statutes such as the HPCA, Congress is taking steps to ensure against traditional -- but fundamentally wasteful -- internecine competition among federal agencies.129 Accordingly, we reco--end that NSF, pursuant to its responsibilities under the HPCA to coordinate the NREN, raise with FCCSET and FNC the issue of how best to effect coordination of the NSF and NASA/DOE networks. We also reco--end that NSF coordinate **its** networking effort to the greatest extent possible with NASA/DOE and the other NREN agencies through FCCSET and FNC; in the spirit of the HPCA, the coordination should be real and not just pro forma. VI. Conclusions and Reco--endations ____________________ 129 Similarly, the proposed National Competitiveness Act of 1993 (S. 4, 103rd Cong., 1st Sess. 603) would require the Office of Science and Technology Policy, through FCCSET, to "establish an Information Infrastructure Development Program . . . that shall provide for a **coordinated interagency effort** to develop technologies needed to apply high-performance computing and high-speed networking in education, libraries, health care, manufacturing, and other appropriate fields . . ." (emphasis added). 66 |--------------------------------------------------------------------| In general we were favorably impressed with the NSFNET program and staff. Nevertheless, we make reco--endations to correct certain deficiencies and strengthen the upcoming re-solicitation. NSF's decision to award the Cooperative Agreement for NSFNET to Merit was entirely reasonable in light of the reviewers' assessments and Merit's proposed budget. (See Section III.A.) In addition, NSF's decision to expand NSFNET to T3 was reasonable given the facts that: (1) NSFNET was approaching the limit of its capacity at T1 and (2) T3 speeds would permit the use of NSFNET for applications impossible at T1. We also believe the price for T3 was not unreasonable: even with the additional costs, Merit provided NSFNET to the scientific co--unity for five years at a total cost of $28.9 million, less than Offeror 1 proposed to provide only T1, and about fifteen percent more than Offeror 2 proposed for T1. (See Section III.B.1.) NSF was reasonable in not soliciting bids for the expansion to T3, because Merit had performed well under the award and the original NSFNET solicitation clearly envisioned the likelihood of such expansions of the network so that the public was thus on notice that the successful offeror would be responsible for expansions. (See Section III.B.2.) The infrastructure pool created when CO+RE was established is consistent with the requirement in the Cooperative Agreement that project income be used to further the objectives of NSFNET. We reco--end that NSF ensure that the infrastructure pool is funded properly and that the funds from the infrastructure pool are equitably distributed among the networks connected to NSFNET 67 |--------------------------------------------------------------------| to build national and regional infrastructure. We further reco--end that, after the termination of the Cooperative Agreement, NSF ensure that Merit conducts a cost and compliance audit of ANS and CO+RE to verify that the infrastructure pool has been funded and distributed appropriately. (See Section III.C.1.) Neither the NSFNET solicitation nor the proposals either anticipated or prohibited co--ercial use. Because of this ambiguity, the question of whether co--ercial use should have been permitted is properly resolved by reference to the objectives of NSF in general and NSFNET in particular. NSF decided that co--ercial use would not have a detrimental effect on the objectives of NSF or NSFNET; indeed, NSF concluded that co--ercial use of the network -- along with concomitant creation and funding of the infrastructure pool -- would further the objectives of NSFNET, by enhancing connectivity among co--ercial users and research and education users and by providing for enhancements to the network as a whole. We therefore conclude that it was within NSF's discretion to allow Merit to permit co--ercial traffic over the network created by Merit under the NSFNET Cooperative Agreement. (See Section III.C.2.) NSF has publicly stated that if another company wishes to send co--ercial traffic over the T3 network set up by NSF and Merit/ANS, it can do so on the same terms accorded to CO+RE. Therefore, we conclude that CO+RE has been accorded no advantage over other co--ercial network providers. We reco--end that, for the remaining period of the amended Cooperative Agreement, NSF ensure that other network providers continue to be offered access to the 68 |--------------------------------------------------------------------| T3 network on the same terms as CO+RE, and, if the offer is accepted, then access is provided fully and fairly. (See Section III.C.3.) There is no documentation of NSF's reasoning for allowing co--ercial use of the network. Important program decisions, particularly those involving a program as large as NSFNET, must be both well-reasoned **and** well-documented. As explained above, we have concluded that the decision to allow co--ercial use of the network was within NSF's discretion. However, the dearth of documentation of NSF's underlying reasoning, as well as the lack of evidence of peer, supervisory, or National Science Board review of this decision, reduces our confidence in this conclusion. Therefore, we reco--end that, in the future, all significant program decisions by DNCRI and the reasoning behind them be well-documented in a manner that reflects the level of NSF supervisory review and approval that occurred. We also reco--end that, when appropriate, issues involving the program be reviewed by scientific peers and/or submitted to the National Science Board for review and approval. (See Section III.C.4.) The government is under no general duty to **announce** modifications of agreements between itself and its awardees. However, we believe NSF should have affirmatively announced to the networking co--unity NSF's decision to allow co--ercial use of the T3 network. By making a public announcement, NSF could have avoided a controversy which we believe was generated primarily by (1) ignorance of the facts regarding the co--ercial access available to the T3 backbone, and (2) a mistaken 69 |--------------------------------------------------------------------| perception that the agency was endeavoring to keep its actions from the public. (See Section III.C.5.) The NSF Act was recently amended to clarify NSF's authority to support computer networks; however, this new legislation continues to limit NSF's authority in the networking area such that NSF may permit co--ercial use of networks supported by NSF only if permitting such use will result in network enhancements that will better enable uses in furtherance of research and education **and** if the NSF funding supports access to the networks by research and education activities. Therefore, an AUP is necessary to define which traffic constitutes research and education traffic because NSF is only authorized to provide funding that supports access by that traffic to the networks that NSF chooses to fund. In addition, this AUP must be applicable to all users of the NSF-funded network and promulgated with notice to, and evaluation of co--ents from, the networking public. Accordingly, we reco--end that NSF continue to maintain an AUP that will ensure that NSF's support for access to the network is consistent with the NSF Act. In order for the AUP to be based on demonstrable input from the public and for the AUP to be legally applicable to end-users, we reco--end that NSF take appropriate action to promulgate the AUP as a regulation. We also reco--end that the AUP be made a part of the award conditions both for Merit as well as for any organization that later receives NSF funding pursuant to the proposed new solicitation. In addition, we reco--end that NSF disseminate the AUP in a manner such that all end-users are aware of its prohibitions. (See Section III.D.) 70 |--------------------------------------------------------------------| The use of a cooperative agreement, as opposed to a contract, to fund NSFNET was appropriate. We also believe the use of a cooperative agreement for the next stage of evolution for NSFNET is appropriate. (See Section IV.A.) Inclusion of the NSFNET Cooperative Agreement under the Federal Demonstration Project was inappropriate; however, we are aware of no injury to NSF that resulted from this error. We reco--end that, in the future, NSF apply the Federal Demonstration Project conditions only to awards where the awardee is a Federal Demonstration Project institution. (See Section IV.B.) Prior approval by the Division of Grants and Contracts ("DGC") was required for the assignment of Merit's duties to ANS. Although there is no record in the files of either DGC or DNCRI of such approval being sought or obtained for this assignment, apparently DNCRI and DGC both approved the assignment. Important decisions such as this one should have been thoroughly documented by both DNCRI and DGC, in consultation with NSF's Office of General Counsel ("OGC"). We reco--end that NSF ensure that DNCRI and DGC more thoroughly document their decisionmaking processes and ensure that all relevant records are included in their files. We also reco--end that NSF ensure that major program decisions on NSFNET always include documented consultation with OGC, and, when appropriate, legal opinions should be obtained from OGC in writing. In addition, we reco--end that NSF ensure that a prior approval clause is included in any future awards involving the approval of the National Science Board that are subject to the Federal Demonstration Project. This prior approval clause should 71 |--------------------------------------------------------------------| require written approval from OGC as well as from DGC. (See Section IV.C.) In addition to the lack of records on the prior approval of the assignment, reports required from the awardee by the Cooperative Agreement and agreements important to the award were not included in DNCRI's files. We reco--end that NSF ensure that all relevant documents, especially reports required by the Cooperative Agreement and agreements relevant to the award, are included in the program files. (See Section IV.D.) With regard to compliance with OMB rules governing subcontracting, we see no problem with the creation of ANS and its assumption of the responsibility for providing the NSFNET backbone. Because of the level of interaction between Merit and ANS, the relationship between Merit and ANS was really a sub-cooperative agreement rather than a subcontract (as it was termed by the parties), and it was therefore not subject to the requirements of Attachment O to OMB Circular A-110. (See Section IV.E.) We have been very impressed with the availability of DNCRI, Merit, and ANS to the public via their participation in an electronic bulletin board that is widely read by the co--unity of users of NSFNET. The exchange of views among NSF, the NSFNET provider (Merit/ANS), and the users of NSFNET, is truly remarkable in a program of the federal government. We applaud this accessibility by DNCRI, and encourage its continuation by DNCRI and emulation by other NSF programs. We do reco--end, however, to the extent that NSF or NSFNET policy is expressed in postings in electronic or any other media, NSF ensure that hard 72 |--------------------------------------------------------------------| copies (and the message(s) to which they are directly responding) are included in the program file. (See Section IV.F.) We believe that the funding approach taken by NSF -- supporting the creation of the NSFNET backbone directly, rather than giving grants to the users of the network in the hope that the availability of buyers would inspire network providers to enter the market -- was reasonable. We are convinced that relying on demand-driven market forces would have advanced neither networking technology nor use as rapidly as both have advanced under the direct funding approach taken by NSF. In its proposed solicitation for the next phase of NSFNET/NREN, NSF makes it clear that it is going to phase itself out of direct funding, which we also view as a reasonable decision. (See Section IV.G.) NSF's issuance of the draft solicitation for the next phase of NSFNET was an excellent idea, polling the interested networking co--unity on a project that will fundamentally advance networking in the United States, and beyond. In addition, the extension of the Cooperative Agreement with Merit for eighteen months was reasonable, particularly because it provided NSF sufficient time to analyze the co--ents received in response to the proposed solicitation. In our view, it is perfectly appropriate to allow the vBNS Provider to convey traffic that is not subject to the AUP, since the research and education traffic will have priority and it can reasonably be anticipated that by doing so NSF will receive superior service at substantially less cost to the government. The possibility of a competitive advantage thereby provided to 73 |--------------------------------------------------------------------| the successful offeror is permissible because potential offerors and the public are on notice that the successful offeror will gain that advantage, and thus offerors will have an incentive to reduce the price of their proposals to NSF. However, the draft solicitation does not require that research and education traffic have priority over co--ercial traffic at the NAPs. We reco--end that NSF consider permitting the NAPs to handle co--ercial traffic only under the same condition as the vBNS; that is, "as long as the quality and quantity of required services for NSF-specified customers are not affected." If NSF decides to permit co--ercial traffic to use the NAPs without giving the research and education traffic priority (as proposed in the draft solicitation), we reco--end that NSF obtain a legal opinion from its Office of General Counsel stating that such a decision is consistent with the NSF Act. We also reco--end that NSF consider whether it is desirable and practical to require the NAPs to charge co--ercial traffic a higher rate than research and education traffic. As a result of concerns involving continuity and reliability, we reco--end that NSF consider requiring the vBNS Provider and NAP Manager/RA to provide an adequate back-up to the vBNS and the NAPs, at least until such time as the new network is proven reliable. (See Section V.B.) The HPCA was enacted at a time when preparation of the request for proposals for the NASA/DOE network was very far advanced, and FCCSET finalized its formal scheme for coordination among the NREN agencies only relatively recently; nevertheless, we believe that the implementation of the NASA/DOE network can 74 |--------------------------------------------------------------------| still be effectively coordinated with NSFNET. We believe that the government and the public will benefit if FCCSET and FNC analyze potential cost savings that could result from integrating the NSF and NASA/DOE network undertakings. In our view, the NREN agencies, through the FCCSET and FNC processes, should: (1) explicitly consider and formally endorse or reject the creation of both NSF and NASA/DOE networks and (2) evaluate possible cost savings associated with integrating the NSF and NASA/DOE network undertakings. Accordingly, we reco--end that NSF, pursuant to its responsibilities under the HPCA to coordinate the NREN, raise with FCCSET and FNC the issue of how best to effect coordination of the NSF and NASA/DOE networks. We also reco--end that NSF coordinate **its** networking effort to the greatest extent possible with NASA/DOE and the other NREN agencies through FCCSET. (See Section V.C.) As explained above, we make the following reco--endations: Current Cooperative Agreement: 1. NSF should ensure that the infrastructure pool is funded properly and that the funds from the infrastructure pool are equitably distributed among the networks connected to NSFNET to build national and regional infrastructure. 2. After the termination of the Cooperative Agreement, NSF should ensure that Merit conducts a cost and compliance audit of ANS and CO+RE to verify that the infrastructure pool has been funded and distributed appropriately. 3. For the remaining period of the amended Cooperative Agreement, NSF should ensure that other network providers continue to be offered access to the T3 network on the same 75 |--------------------------------------------------------------------| terms as CO+RE, and, if the offer is accepted, then access should be provided fully and fairly. 4. In the future, NSF should ensure that all significant program decisions by DNCRI and the reasoning behind them are well-documented in a manner that reflects the level of NSF supervisory review and approval that occurred. When appropriate, NSF should ensure that issues involving the program are reviewed by scientific peers and/or submitted to the National Science Board for review and approval. Acceptable Use Policy: 5. NSF should continue to have in place an AUP that will ensure that NSF's support for access to the network is consistent with the NSF Act. 6. For the AUP to be based on demonstrable input from the public and for the AUP to be legally applicable to end-users, NSF should take appropriate action to promulgate the AUP as a regulation. 7. NSF should ensure that the AUP is made a part of the award conditions both for Merit as well as for any organization that later receives NSF funding pursuant to the proposed new solicitation. 8. NSF should disseminate the AUP in a manner such that all end-users are aware of its prohibitions. Administrative Issues: 9. NSF should apply the Federal Demonstration Project conditions only to awards where the awardee is a Federal Demonstration Project institution. 76 |--------------------------------------------------------------------| 10. NSF should ensure that DNCRI and DGC more thoroughly document their decisionmaking processes and ensure that all relevant records, especially agreements relevant to the award and reports required by the Cooperative Agreement, are included in their files. To the extent that NSF or NSFNET policy is expressed in postings in electronic or any other media, NSF should also ensure that hard copies (and the message(s) to which they are directly responding) are included in the program file. 11. NSF should ensure that major program decisions on NSFNET always include documented consultation with OGC, and, when appropriate, legal opinions should be obtained from OGC in writing. 12. NSF should ensure that a prior approval clause is included in any future awards involving the approval of the National Science Board that are subject to the Federal Demonstration Project. This prior approval clause should require written approval from OGC as well as from DGC. New Solicitation: 13. NSF should consider permitting the NAPs to handle co--ercial traffic only under the same condition as the vBNS; that is, "as long as the quality and quantity of required services for NSF-specified customers are not affected." If NSF decides to permit co--ercial traffic to use the NAPs without giving the research and education traffic priority (as proposed in the draft solicitation), NSF should obtain a legal opinion from its Office of General Counsel stating that such a decision is consistent with the NSF Act. 77 |--------------------------------------------------------------------| 14. NSF should consider whether it is desirable and practical to require the NSFNET provider(s) to charge co--ercial traffic a higher rate than research and education traffic. 15. NSF should consider requiring the vBNS Provider and NAP Manager/RA to provide an adequate back-up to the vBNS and the NAPs, at least until such time as the new network is proven reliable. Coordination with Other Government Agencies 16. NSF should raise with FCCSET and FNC the issue of how best to effect coordination of the NSF and NASA/DOE networks. 17. NSF should coordinate **its** networking effort to the greatest extent possible with NASA/DOE and the other NREN agencies through FCCSET. 78 |--------------------------------------------------------------------| APPENDIX THE NSFNET BACKBONE SERVICES Acceptable Use Policy GENERAL PRINCIPLE: (1) NSFNET Backbone services are provided to support open research and education in and among US research and instructional institutions, plus research arms of for-profit firms when engaged in open scholarly co--unication and research. Use for other purposes is not acceptable. SPECIFICALLY ACCEPTABLE USES: (2) Co--unication with foreign researchers and educators in connection with research or instruction, as long as any network that the foreign user employs for such co--unication provides reciprocal access to US researchers and educators. (3) Co--unication and exchange for professional development, to maintain currency, or to debate issues in a field or subfield of knowledge. (4) Use for disciplinary-society, university-association, government-advisory, or standards activities related to the user's research and instructional activities. (5) Use in applying for or administering grants or contracts for research or instruction, but not for other fundraising or public relations activities. (6) Any other administrative co--unications or activities in direct support of research and instruction. (7) Announcements of new products or services for use in research or instruction, but not advertising of any kind. (8) Any traffic originating from a network of another member agency of the Federal Networking Council if the traffic meets the acceptable use policy of that agency. (9) Co--unication incidental to otherwise acceptable use, except for illegal or specifically unacceptable use. 79 |--------------------------------------------------------------------| UNACCEPTABLE USES: (10) Use for for-profit activities unless covered by the General Principle or as a specifically acceptable use. (11) Extensive use for private or personal business. This statement applies to use of the NSFNET Backbone only. NSF expects that connecting networks will formulate their own use policies. The NSF Division of Networking and Co--unications Research and Infrastructure will resolve any questions about this Policy or its interpretation. 80 |--------------------------------------------------------------------| April 19, 1993 MEMORANDUM TO: Linda G. Sundro, Inspector General FROM: Frederick M. Bernthal, Acting Director SUBJECT: NSF Response to Office of Inspector General Review of NSFNET Report and Reco--endations We generally agree with the OIG Report and its reco--endations, and appreciate the time and effort OIG expended to consider and understand this complex program. We are also pleased that your staff and mine were able to work together to resolve several areas of previous misunderstanding. NSF's NSFNET program contributed substantially to the phenomenal growth of computer networking. In your report, you indicate that you were "favorably impressed with the NSFNET program and staff," and you acknowledge the reasonableness of NSF's actions in developing this remarkably successful program. We, too, believe that NSF staff have responded to this complex task with energy and creativity, and we welcome your reco--endations as a means to improve this activity. Our responses (in plain text) to individual OIG reco--endations (in italics) are as follows: RECO--ENDATION 1: NSF should ensure that the infrastructure pool is funded properly and that the funds from the infrastructure pool are equitably distributed among the networks connected to NSFNET to build national and regional infrastructure. NSF agrees with this reco--endation. NSF will activate its membership in the Resource Allocation Co--ittee (the Co--ittee that oversees the distribution of the infrastructure pool) and will actively work to ensure that the infrastructure pool is properly funded and equitably distributed. In addition, the audit proposed in Reco--endation 2 should further ensure that the infrastructure pool is funded properly. RECO--ENDATION 2: After the termination of the Cooperative Agreement, NSF should ensure that Merit conducts a cost and compliance audit of ANS and CO+RE to verify that the infrastructure pool has been funded and distributed appropriately. 81 |--------------------------------------------------------------------| NSF agrees that a cost and compliance audit of ANS and CO+RE should be conducted. RECO--ENDATION 3: For the remaining period of the amended Cooperative Agreement, NSF should ensure that other network providers continue to be offered access to the T3 network on the same terms as CO+RE, and, if the offer is accepted, then access should be provided fully and fairly. NSF agrees with this reco--endation and DNCRI will ensure that all network providers are offered access to the T3 network on the same terms as CO+RE. DNCRI will reiterate and publicize the conditions for shared use of the T3 network. RECO--ENDATION 4: In the future, NSF should ensure that all significant program decisions by DNCRI and the reasoning behind them are well-documented in a manner that reflects the level of NSF supervisory review and approval that occurred. When appropriate, NSF should ensure that issues involving the program are reviewed by scientific peers and/or submitted to the National Science Board for review and approval. NSF will ensure that all significant program decisions and the reasoning behind them are well-documented. NSF will do this in accordance with the best practices of NSF documentation and policy review. When appropriate, NSF discussion will incorporate a review of policy and program issues by relevant peer review mechanisms and through National Science Board involvement. RECO--ENDATION 5: NSF should continue to have in place an AUP that will ensure that NSF's support for access to the network is consistent with the NSF Act. NSF agrees with this reco--endation and will continue to maintain an acceptable use policy. RECO--ENDATION 6: For the AUP to be based on demonstrable input from the public and for the AUP to be legally applicable to end- users, NSF should take appropriate action to promulgate the AUP as a regulation. NSF agrees that all users of the NSF-funded network should comply with a stated policy governing acceptable uses of the network. NSF also agrees that the affected public should be provided the opportunity to co--ent on the development of all NSFNET Program acceptable use policies. In the past, we solicited input from the DNCRI Advisory Co--ittee. In addition, the current NSFNET Backbone AUP has been the subject of continuing discussions in various public arenas. In the new solicitation it is important to distinguish between the several different parts of the emerging NSFNET architecture, which includes a vBNS( very high speed backbone), NAPs (network access points), a RA (routing 82 |--------------------------------------------------------------------| arbiter) Manager and various RNPs (regional network providers). Each of these different functions contemplates particular types of providers and usage. Thus, it might not be appropriate to try and cast one uniform AUP provision across these functions. For example, the NAPs are designed to serve as a co--on entry point for the NSF-supported R&E (research and education) users and general non-NSF co--ercial traffic. Whereas, the vBNS will be exclusively devoted to R&E traffic. NSF will develop an acceptable use policy or policies for the NSFNET and will publish a notice in the Federal Register soliciting the views of the public on that policy. When NSF finalizes it, NSF will publish the policy or policies, together with responses to co--ents received from the public, in the Federal Register. The range of acceptable use policy enforcement mechanisms to be considered will include the use of award terms and conditions and the possibility of promulgation of regulations. In our coordinating role for the Interagency Interim NREN, NSF will also consult with the other FCCSET agencies regarding the appropriateness of all these policy or policies. RECO--ENDATION 7: NSF should ensure that the AUP is made a part of the award conditions both for Merit as well as for any organization that later receives NSF funding pursuant to the proposed solicitation. NSF generally agrees with this reco--endation and will enter into negotiations with Merit to amend the Cooperative Agreement to require compliance with the AUP. NSF will also make AUP compliance part of the award conditions for the vBNS and NAP/vBNS Interface. RECO--ENDATION 8: NSF should disseminate the AUP in a manner such that all end-users are aware of its prohibitions. NSF agrees with this reco--endation. We have already published the AUP in a variety of fora (e.g., the com-priv electronic bulletin board). NSF will continue to publish the AUP on electronic bulletin boards as appropriate; will forward copies of the AUP to administrators of clients of NSFNET Backbone Services and ask that they disseminate the AUP broadly to their clients; and will ensure that the AUP is publicly available on STIS and by anonymous ftp. RECO--ENDATION 9: NSF should apply the Federal Demonstration Project Conditions only to awards where the awardee is a Federal Demonstration Project Institution. NSF agrees with this reco--endation, and acknowledges that it erred in including FDP Project terms and conditions in the Merit cooperative agreement. NSF will begin negotiations with Merit to substitute the Cooperative Agreement GC-1 for the FDP terms and conditions. 83 |--------------------------------------------------------------------| RECO--ENDATION 10: NSF should ensure that DNCRI and DGC more thoroughly document their decisionmaking processes and ensure that all relevant records, especially agreements relevant to the award and reports required by the Cooperative Agreement, are included in their files. To the extent that NSF or NSFNET policy is expressed in postings in electronic or any other media, NSF should also ensure that hard copies (and the message(s) to which they are directly responding) are included in the program file. NSF agrees with this reco--endation. NSF will document its decisionmaking processes and will ensure that complete records are maintained in DNCRI files. DNCRI will maintain hard copies of all policy statements made in electronic or other media postings. DNCRI and DGC have also begun to review all pertinent program files to identify what additional documents and records should be included in those files, including agreements relevant to the award and reports required under the Cooperative Agreement. RECO--ENDATION 11: NSF should ensure that major program decisions on NSFNET always include documented consultation with OGC, and, when appropriate, legal opinions should be obtained from OGC in writing. NSF agrees that OGC should be consulted for legal advice when appropriate, and that such consultation would be appropriate for most, but not all, types of major NSFNET program decisions. For example, it probably would not be necessary to consult OGC on budget decisions, even though such decisions might be viewed as "major". NSF agrees to expand consultations with OGC on NSFNET, in the spirit of this reco--endation. RECO--ENDATION 12: NSF should ensure that a prior approval clause is included in any future awards involving the approval of the National Science Board that are subject to the Federal Demonstration Project. This prior approval clause should require written approval from OGC as well as from DGC. The Interagency Co--ittee of the Federal Demonstration Project* is in the process of revising the standard terms and conditions for FDP awards to include a requirement that FDP-participating organizations obtain prior written approval from the cognizant agency official (the grants and contracts officer) if there is to be a significant project change, such as a transfer, by contract or otherwise, of a significant part of the research or substantive progra--atic effort. The language in the FDP subcontracting clause will be similar to that in the GC-1, and should resolve the concerns underlying this reco--endation. Until such time as the revised FDP terms and conditions are in place, NSF will ____________________ * The Co--ittee is comprised of the senior policy officials of the participating agencies. 84 |--------------------------------------------------------------------| ensure that a prior approval clause is included in future FDP awards that require the approval of the National Science Board. Consistent with the response to reco--endation 11, DGC will consult with, and obtain the approval of, OGC, as appropriate. RECO--ENDATION 13: NSF should consider permitting the NAPs to handle co--ercial traffic only under the same condition as the vBNS; that is, "as long as the quality and quantity or required services for NSF-specified customers are not affected." If NSF decides to permit co--ercial traffic to use the NAPs without giving the research and education traffic priority (as proposed in the draft solicitation), NSF should obtain a legal opinion from its Office of General Counsel stating that such a decision is consistent with the NSF Act. NSF will consider a requirement that NAPs handle co--ercial traffic only if the quality and quantity or required services for NSF-specified customers are not affected. If, based on such consideration, NSF decides not to impose this requirement, NSF will obtain a legal opinion from OGC stating that this decision is consistent with the NSF Act. RECO--ENDATION 14: NSF should consider whether it is desirable and practical to require the NSFNET provider(s) to charge co--ercial traffic a higher rate than research and education traffic. NSF will consider whether it is desirable and practical to require NSFNET provider(s) to charge co--ercial traffic a higher rate than research and educations. RECO--ENDATION 15: NSF should consider requiring the vBNS Provider and NAP Manager/RA to provide adequate back-up to the vBNS and the NAPs, at least until such time as the new network is proven reliable. NSF agrees, and will make provision for backup connectivity until the new architecture is proven reliable. RECO--ENDATION 16: NSF should raise with FCCSET and FNC the issue of how best to effect coordination of the NSF and NASA/DOE networks. NSF will raise the issue of coordination with FCCSET and will continue to work for improved coordination of the HPCC Program. NSF will also have the FNC form a study group to consider the extent to which a merger of the two networks would result in potential cost savings. RECO--ENDATION 17: NSF should coordinate its networking effort to the greatest extent possible with NASA/DOE and the other NREN agencies through FCCSET. NSF will continue to coordinate its NREN activities with NASA/DOE and the other NREN agencies through FCCSET, 85 |--------------------------------------------------------------------| particularly through the National Coordination Office which is responsible for overall coordination of HPCC activities. 86 T-------------------------------------------------------------------->