Summary
- US federal programmes influenced early number administration through funded backbones, contractor-run registries and access policies, but these mechanisms operated through separate desks and did not establish universal authority over Internet identifiers.
- The surviving record shows practical international consequences but lacks a complete award-to-allocation dataset and representative non-US request files, so its reach cannot responsibly be converted into a count of assignments controlled by Washington.
On 1 April 1993, the administrative destination for a large class of Internet registration requests changed. Non-Defense Data Network users were directed away from the DDN Network Information Center and towards the Registration Services component of the new InterNIC. DDN users remained with the DDN-NIC. New templates removed DDN-specific references, separate WHOIS services divided the resulting records, and Network Solutions became the named recipient for non-DDN requests.
This was not simply a change of postal address. It moved an applicant-facing service that handled IP-address and autonomous-system-number work from a centre rooted in the military network environment to Network Solutions as the Registration Services recipient under a National Science Foundation cooperative agreement for the non-military Internet. It did so without moving every related function to the same organisation. USC’s Information Sciences Institute retained the coordinating work associated with the Internet Assigned Numbers Authority. The DDN-NIC continued serving its military constituency. The other InterNIC participating organisations supplied information and directory services. Backbone operation and routing remained elsewhere.
The transition exposes the distinctive character of federal influence over early number administration. The United States did not exercise that influence through a single office possessing every relevant power. It financed research, made cooperative-agreement awards, sponsored backbones, contracted for other work and imposed rules on facilities under particular programmes. Those interventions could make one registration desk, one route or one information service exceptionally important. Their reach depended, however, on the instrument, the service population and the network path involved.
The same separation is necessary when reading the celebrated growth figures for NSFNET. Merit’s final report says that the expanded T1 backbone went online in July 1988 with only the United States, France and Canada represented among its international connections. The report is internally inconsistent about the network population at that moment. Its overview, on page 4, gives 217 connected networks; its operational narrative, on page 24, says more than 170. It does not reconcile the two figures. Neither should therefore be used as an exact baseline for a growth calculation.
At the other end of the programme, the report presents a country-coded table dated 30 April 1995 containing 50,766 networks and says that 93 countries had been announced to the NSFNET backbone service. That later population is routing-related. It is not a count of grants, applicants, physical circuits, recipient institutions, people or assignments made by a federal agency. The three figures—217, more than 170, and 50,766—describe populations whose counting relationships are not documented well enough to form one precise time series.
The administrative transition of 1993 and the routing reach observed in 1995 nevertheless belong to the same institutional history. Federal programmes helped establish both an important registration corridor and a highly valuable connectivity environment. They affected the conditions under which numbers were requested, recorded, announced and made useful. But assignment, registration, connection and route acceptance remained different acts. Recovering that difference is the key to understanding both the strength and the limit of US funding.
A sequence of roles, not a single chain of command
The institutional sequence began before the name “IANA” had acquired its later significance. According to the US Government Accountability Office’s 2016 legal reconstruction, the identifier-coordination work led by Jon Postel moved from the University of California, Los Angeles, to USC’s Information Sciences Institute in 1977. It continued there through Defense Department-funded research projects. The GAO was examining property implications of a much later transition, not writing an allocation history, but its chronology is useful for locating the contractual relationship.
The surviving contractual record is incomplete. The GAO could not obtain copies of the 1970s-1990s DARPA contracts under which the IANA functions were considered to have been developed and performed. That absence matters. It prevents a clause-by-clause account of the government’s rights, USC’s duties and the precise performance language governing early number coordination. The existence of Defense-funded research projects supports a sponsorship and contractor relationship; it does not supply missing terms about every identifier or applicant.
A better-preserved instrument appears only at the edge of the period. DARPA’s final USC contract for this work was the Tera-node Network Technology contract, DABT63-95-C-0095. Task 4 covered “network infrastructure activities,” including acting as the Internet Assigned Numbers Authority, and ran from July 1995 to July 1999. Those dates are essential. The task confirms an explicit contractual duty beginning in July 1995, but its language cannot be projected backwards as if it were the text of every Defense contract from 1977 onwards.
The operational division visible in August 1990 came from another source. RFC 1174, an informational statement of the Internet Activities Board’s recommended policy to the Federal Networking Council, said that USC/ISI performed the central IANA function. It also said that IANA had lodged responsibility for numeric network and autonomous-system identifiers with an Internet Registry operated by SRI International at the DDN-NIC. The RFC proves what the IAB described at that date. It is not SRI’s contract, and the underlying funding instrument for SRI’s Internet Registry work is not established by the RFC.
The agency terminology also changed during this history. The Defense Communications Agency was reorganised and renamed the Defense Information Systems Agency on 25 June 1991, according to DISA’s official institutional history. Contemporary documents reflect the transition. RFC 1174 belongs to the August 1990 DDN environment. By March 1993, the forms retained for DDN users were described as DISA-approved. Extending the DCA label across both dates would conceal an institutional change; using DISA for the 1990 record would be equally anachronistic.
A parallel civil programme was developing under the National Science Foundation. NSF issued Solicitation 87-37 for management and operation of the expanded NSFNET backbone on 15 June 1987. Merit Network managed and coordinated the resulting project under cooperative Agreement NCR 8720904, working with IBM, MCI and the State of Michigan. Merit was the cooperative-agreement award recipient and programme manager. IBM supplied engineering and equipment. MCI supplied telecommunications facilities and expertise. NSF sponsored the programme and set its public purpose. None of these roles made Merit or NSF the general issuer of every network number carried by the backbone.
The operating arrangement changed again in September 1990 when the partners announced Advanced Network & Services. The Merit final report, written by a programme participant and published as a retrospective account in 1995, says that ANS subsequently provided the NSFNET backbone service as Merit’s subcontractor and took over the Network Operations Center. That is evidence of a service and NOC role under the Merit arrangement. It is not evidence that ANS assumed IANA coordination or Internet Registry duties.
The report says IBM and MCI initially contributed $4 million to ANS and that each pledged an additional $1 million, along with personnel and equipment. The initial contribution and the two later pledges must remain separate. A pledge is not proof of receipt or expenditure, and these corporate amounts do not share an accounting basis with NSF’s reported support under NCR 8720904.
By 1994, another bounded function appeared in the NSFNET successor architecture. NSF Solicitation 93-52 divided the transition among a very-high-speed research backbone, Network Access Points, a Routing Arbiter and regional-provider support. Merit and USC/ISI formed the team selected for the Routing Arbiter. The Merit report identifies the team and the 1994 announcement but does not supply an award number that can safely be reproduced here. The Routing Arbiter handled route information for providers connecting through the emerging architecture. Its database did not become the source of the network numbers represented in those routes.
This sequence is more revealing than a consolidated institutional label. USC/ISI coordinated and delegated identifier work under incompletely preserved Defense contracts. SRI operated the Internet Registry described in August 1990, although that RFC was not its funding instrument. Merit managed the NSFNET cooperative agreement. ANS later ran the backbone service and NOC under subcontract to Merit. InterNIC’s Registration Services took over the non-DDN request population in 1993. The Routing Arbiter coordinated route information in the successor environment. Each role was consequential; none encompassed all the others.
What RFC 1174 actually recorded
RFC 1174 is unusually valuable because it captures a system under strain. It is also easy to overread. The document was issued in August 1990 as an IAB recommendation to the Federal Networking Council. It did not specify an Internet standard, enact a statute or report comprehensive implementation results.
Its historical account described an Internet that had outgrown an original client base centred on military, government and government-sponsored research organisations. In the earlier research environment, it said, numbers had gone to participating organisations. As the infrastructure expanded, other organisations obtained numbers and were allowed to interconnect when they were part of the US government or had a government sponsor. The DDN-NIC at SRI supplied the single Internet Registry intended to maintain global uniqueness.
Commercial use disrupted that arrangement. Organisations were installing private networks that needed unique network numbers even though they were not intended to connect to the federally sponsored system. RFC 1174 reported that the Internet Registry had adopted a policy of assigning network numbers to those who requested them while distinguishing networks sanctioned for interconnection through “connected” status. This is significant counterevidence to any claim that federal connection was invariably required for assignment.
Connected status did have practical administrative effects. The RFC associated it with US government sponsorship and said the concept appeared in Internet Registry forms and databases. Its DNS attachment reported a past policy under which inclusion in a key DNS database was tied to approval of Internet interconnection by a US government agency. These are specific claims about reported practice. They should not be widened into a conclusion that connected status uniformly controlled every kind of registration, assignment or route.
The IAB proposed three immediate changes. It recommended removing connected-status references from forms and databases for network and autonomous-system registration. It recommended collecting short statements about acceptable use, access and transit policy. It also recommended allowing any registered network into the DNS without regard to connected status. The attached DNS recommendation called for name and reverse-address registration to be separated from routing and said that no US government sponsor should be required for that registration.
Those were recommendations. The proposed policy database was not proof that every registry and operator later implemented such a database uniformly. Nor did the RFC demonstrate that every connected-status reference disappeared immediately, that every foreign delegation occurred, or that every route operator used the proposed policy information. The document establishes the IAB’s view that the old coupling had become untenable and records the practices to which the proposals responded.
The proposed distribution of identifier work was similarly cautious. RFC 1174 recommended retaining central IANA and Internet Registry functions. The Registry would allocate blocks of network and autonomous-system numbers to organisations approved through the Coordinating Committee for Intercontinental Research Networking, allowing those organisations to make further assignments. The existing Registry would remain the default where no delegated authority had been identified. Aggregate databases would be shared, but updates would initially remain centralised.
That was not a proposal for unrestricted competition among interchangeable registries. It envisaged delegated blocks within a coordinated hierarchy, with central updates preserved during the initial period. Its architecture tried to widen participation without abandoning consistency.
The RFC also drew a firm territorial and operational boundary. It said that the growing population of non-US networks should not be required to follow US access and use criteria merely because they belonged to the global Internet. Federal criteria could apply to traffic sent through federally sponsored networks. The distinction was between using a funded facility and possessing a globally unique identifier.
That boundary did not guarantee perfect observance. RFC 1174 did not audit every operator, request or route. Yet it shows that leading participants rejected the equation of global registration with universal US acceptable-use authority. It also located enforcement with network administrators rather than the Registry. The Registry was to administer number space and collect information; operators were to decide which traffic they would carry.
The result was a more plural conception of authority. IANA coordinated and delegated. The Internet Registry processed and recorded identifiers. DNS operators maintained naming data. Backbone sponsors defined conditions for their facilities. Network operators accepted or rejected routes and traffic. These functions interacted, but the recommendation’s purpose was to stop their interaction from being mistaken for one global permission.
NSFNET money and the meaning of a backbone award
NSFNET supplied the most visible infrastructure in this history, but its financial record needs careful treatment. The Merit report identifies NCR 8720904 as the cooperative agreement under which Merit managed and coordinated the NSFNET Backbone Network Project. On page 13, it says NSF provided funds in the amount of $57.9 million for creation of the expanded backbone over an agreement originally awarded for five years and later extended to seven and one-half years.
The figure is nominal as reported. The source does not state a price year or provide a fiscal schedule. More importantly, the participant retrospective is not an award ledger. It does not establish whether $57.9 million was the initial award value, the cumulative amount obligated through amendments or the final amount spent. It is best understood as the report’s retrospective figure for funds provided to create the expanded backbone over the extended agreement.
That boundary matters because the programme drew on several financial sources. IBM and MCI provided equipment, staff and services; the State of Michigan also contributed support. ANS later received the reported corporate contribution and pledges. Regional networks assembled money from universities, states, companies and users. Adding these figures together would imply a common accounting frame that the sources do not supply.
Nor was the award a payment for number assignments. It supported a backbone service: nodes, circuits, routers, engineering, operation, technical coordination and transition. Its institutional importance arose from what the service enabled. A network that could exchange routes with the backbone gained access to a large research and education community. A network number visible through that system became operationally useful across a widening set of destinations.
The launch statistics illustrate both the scale and the evidentiary problem. Page 4 of the Merit report says the service grew from 217 connected networks in July 1988. Page 24 says the new backbone went online that month with more than 170 connected networks. The difference could reflect timing, counting rules, editorial error or inclusion of different downstream populations, but the report does not say. Selecting either value as the exact start of a percentage-growth series would manufacture precision.
The three-tier structure further complicates the unit. Campus networks generally reached the national backbone through regional or midlevel networks. A connected network was not necessarily a direct circuit, an award recipient or a single institution. The relationship between attachment, announced network and downstream user changed as routing and aggregation practices developed.
A separate reach claim comes from the GAO’s 2016 chronology. It says NSFNET connected networks serving more than 4,000 research and educational institutions throughout the United States. That is a later retrospective statement about institutions served through connected networks. The GAO does not give an exact observation date that would allow the figure to be aligned with the 30 April 1995 country table. One institution could operate multiple networks, while one regional network could serve multiple institutions. The two populations cannot be divided into each other.
NSF’s acceptable-use conditions still gave the award practical force. Traffic over the backbone was expected to support research and education. Intermediate networks might carry broader traffic on their own facilities, while their use of NSFNET remained subject to programme policy. RFC 1174 noted the resulting difficulty: a commercial organisation could generate both research-related traffic and unrelated commercial traffic, while policy-based routing operated at the coarse level of whole networks.
This was a real mechanism of influence, but it concerned carriage. A route could be eligible or ineligible for a funded backbone without changing the validity of the underlying number. An assigned number did not compel NSFNET to announce or carry it. Conversely, appearance in NSFNET routing information did not show that NSF had assigned it.
The backbone agreement therefore created conditional operating dependence rather than a merged administrative power. NSF governed the cooperative-agreement funds and the facilities supported by them. Merit managed the cooperative agreement; ANS later operated the backbone service under subcontract to Merit. Regional and network operators managed attachments and routes. Identifier coordination and applicant-facing registration remained with other institutions.
The DDN-NIC to InterNIC corridor
The most concrete link between a federal funding agreement and number administration is the 1993 transfer of non-DDN registration services. Unlike a hypothetical applicant story, the surviving documents identify the affected class, the operating change, its date and its exception.
The GAO’s later reconstruction says that NSF’s cooperative agreement with Network Solutions became effective on 1 January 1993. The underlying agreement text is not reproduced in the cited record, so its detailed clauses should not be inferred from the GAO summary. The effective date nevertheless belongs to the award chronology.
The operational transition is documented directly in RFC 1400, issued in March 1993 by an author at Network Solutions. It said that, as a result of the NSF NREN Network Information Services award, non-DDN registration services were being transferred from the DDN-NIC to the Internet Registration Service within InterNIC. References to “InterNIC” in that document applied only to Registration Services, not to every InterNIC component.
During March, requests sent to either site could still be processed. From 1 April 1993, new non-DDN Internet registration requests were to use the InterNIC template and the new address at Network Solutions. The templates had been modified for automated processing and to remove DDN-specific references. This is specific implementation evidence following the broader 1990 recommendation: the operational form and destination for non-DDN requests changed.
The exception was explicit. DDN users continued to receive full registration support from the DDN-NIC. They were told to keep using the relevant military addresses and DISA-approved forms for IP, .mil, reverse-address and user-registration work. The service transfer was therefore divided by constituency, not a wholesale closure of the older centre.
The information services were divided as well. Effective 1 April, the DDN-NIC WHOIS service was to contain only DDN information, while the InterNIC WHOIS service would contain information on IP addresses, domains and autonomous-system numbers. Root-zone and host-file arrangements changed on their own schedules. RFC 1400 also described a new automated system that parsed templates, checked verifiable fields, returned errors and required requesters to confirm the interpreted information before final processing.
The document does not provide a representative set of applications or demonstrate how often the new system accepted, delayed or refused technically valid requests. It does not establish a causal effect on allocation volume. What it establishes is narrower and stronger: non-DDN applicants were directed to a new registration service on a defined date, using changed forms and an automated workflow, while DDN users remained with the old service.
A year later, RFC 1594 described the operating division in greater detail. The InterNIC was a five-year project partially supported by NSF and had begun operations in April 1993. It consisted of three organisations. General Atomics provided Information Services. AT&T provided Directory and Database Services. Network Solutions provided Registration Services.
RFC 1594 expressly included IP-address allocation and autonomous-system-number assignment among Network Solutions’ registration functions, alongside domain registration and maintenance of contact information. It separately described the Registration Services component as the Internet Registry then responsible for assigning IP network and autonomous-system numbers, while noting delegations to regional registries. USC/ISI remained the location of IANA and the coordinating point for other protocol parameters.
This division prevents several common collapses. InterNIC was not synonymous with Network Solutions. Network Solutions’ registration work was not the same as General Atomics’ reference service or AT&T’s directories. Domain registration was not the whole of number administration. The Registration Service was not the NSFNET NOC. Maintaining IP and ASN records did not give the award recipient authority to decide which routes a backbone accepted.
The corridor nevertheless shows how a federal programme could redirect administrative dependence. Before the transfer, a non-DDN requester used a service whose institutional identity and forms remained rooted in the Defense network environment. After 1 April, that requester used Network Solutions’ InterNIC Registration Service under the NSF NIS cooperative-agreement arrangement. The registration records and request workflow moved even though the requester’s physical carrier, route, sponsor and funding source might not have changed.
The outcome was a clearer separation between military and non-military constituencies. It also shifted a globally important registration function to the recipient responsible for Registration Services under a civil networking cooperative agreement. NSF’s leverage over that recipient arose from the cooperative agreement. It did not automatically extend to every operator carrying a registered route or every foreign institution using the identifier.
The available sources do not supply the complete award, all amendments or enough applicant files to measure performance before and after the transfer. They do, however, support the central institutional claim: the federal cooperative-agreement award altered where non-DDN number-registration requests were processed, by whom and under which service architecture.
Registration and routing diverged further after 1993
The registration transition coincided with a larger reorganisation of connectivity. Commercial providers were expanding, NSF was preparing to retire its general backbone, and route coordination was becoming a distinct operational problem.
Merit introduced an expanded Policy Routing Database in 1993 as part of procedures used on the NSFNET backbone. Route-policy records could tell operators which networks another network intended to announce and under what conditions. They did not allocate the number being announced. Their value lay in making already assigned number space manageable across multiple administrative domains.
NSF Solicitation 93-52 made that distinction more visible. The successor architecture had four components: a very-high-speed research backbone, Network Access Points, a Routing Arbiter and regional-provider awards. The Network Access Points were intended as interconnection locations for commercial providers. The Routing Arbiter was intended to manage route tables and databases used by those providers. Regional and midlevel networks received phased support to purchase commercial Internet service, with providers required to connect at the exchange points.
The awards announced in 1994 distributed the roles. MCI received the very-high-speed backbone award. Sprint and MFS Datanet received Network Access Point management roles, while Bellcore was involved in two NAPs. Merit and USC/ISI formed the Routing Arbiter team. Seventeen regional and midlevel networks received interregional-connectivity awards. These are counts within one NSF transition programme, not totals for the Internet’s providers, registries or connected networks.
The Routing Arbiter Database succeeded Merit’s Policy Routing Database and became an important source of routing and networking information. Calling it a registry is accurate in the ordinary sense of a structured operational record, but it was not the Internet Registry responsible for allocating number space. A route-policy entry expressed how an existing identifier might be announced. It did not create the identifier.
This mattered when the NSFNET backbone was decommissioned on 30 April 1995. The numbers carried over it did not expire. Regional networks moved towards commercial providers; routes were exchanged through the new interconnection structure; coordination records migrated into the emerging Internet Routing Registry environment. The service dependency changed while the identifiers persisted.
That continuity is stronger evidence than a broad claim about ownership. If a network’s number had derived its validity solely from the NSFNET programme, terminating the backbone would have invalidated it. Instead, the number could remain registered and reachable through new routes, subject to the policies and contracts of new operators.
Federal influence did continue. NSF selected the initial components, funded transition awards and imposed interconnection requirements on providers receiving programme money. The Routing Arbiter’s early position benefited from the award. A regional network’s purchasing choices were shaped by the terms and phase-out schedule of federal support. These were material interventions in the emerging market.
But the successor design dispersed operating control. Providers made their own service arrangements and exchanged traffic at shared points. A regional network could purchase connectivity rather than depend on a single general NSF backbone. Route acceptance remained an operator decision, even where the Routing Arbiter supplied information that made the decision easier.
The distinction between registration and routing had therefore become institutional as well as conceptual. Network Solutions processed non-DDN IP and ASN requests. USC/ISI retained coordinating responsibilities. Merit and USC/ISI handled Routing Arbiter work. Commercial carriers operated paths. Each depended on reliable information from the others, but none could replace the others’ function by administrative assertion alone.
International reach without an assignment count
NSFNET’s final country table captures the international scale of this operating environment. It should be read as a dated routing population, not as a census of federal authority.
The table is titled “NSFNET Networks by Country” and dated 30 April 1995. Its total is 50,766. The accompanying text says that, at the end of the project, 93 countries had been announced to the backbone service. The United States row contains 28,470 entries. Canada contains 4,795; France 2,003; Germany 1,750; and the United Kingdom 1,436.
Subtracting the US row from the total yields 22,296 entries associated with the other country rows. As author calculations from the 50,766-entry population, the US share is approximately 56.1 per cent and the combined non-US share approximately 43.9 per cent. These percentages describe the report’s country-coded NSFNET population on one date. They are not shares of global allocations, federal funding, applicants, traffic, physical connections or exclusive dependence.
The historical contrast with July 1988 is striking but cannot be rendered as a clean growth rate. The report says only three countries—the United States, France and Canada—were connected when the T1 service came online. It also provides the unreconciled 217 and “more than 170” network counts. By 1995, the table referred to networks announced to the backbone, a population likely to include extensive downstream reach. Without a stable counting method, exact multiplication from either 1988 figure would be spurious.
France demonstrates what the table can and cannot show. Its 2,003 entries establish substantial representation in the final country-coded routing population. They do not identify the registries that allocated the numbers, the institutions operating them, the funding behind them or the availability of other transit. No complete French request file has been located that follows an applicant through assignment, registration, sponsorship, connection, route acceptance and any alternative service.
A French network could depend heavily on NSFNET for access to American research institutions while obtaining its number through another administrative path. It could have more than one route, or it could face alternatives that were technically possible but commercially unattractive. The table cannot decide among those possibilities. It records reach into one important service.
The non-US boundary in RFC 1174 helps interpret that reach. Traffic sent through federally sponsored networks could be required to conform to federal criteria. A foreign network’s use of a number did not, by itself, subject all of its traffic or internal policy to those criteria. The practical question was which path the traffic used and what alternatives existed.
Alternative paths were present but uneven. RFC 1174 observed that intermediate networks willing to carry arbitrary traffic could interconnect directly even when purely commercial traffic could not cross federally sponsored backbones. Commercial providers were developing unrestricted services. Foreign research systems and regional arrangements could supply other routes. The sources do not establish their price, geographic availability or equivalence for a representative set of non-US users.
The final table therefore measures the scale at which NSFNET policy could matter, not the number of assignment decisions governed by NSF. It shows that a US-funded backbone had consequences well beyond the United States. It cannot convert those consequences into a general jurisdiction over the identifiers visible through it.
Where the dependence broke
Several forms of counterevidence define the edge of federal leverage.
RFC 1174 itself records two decisive breaks. Private organisations received globally unique numbers for networks not intended to connect to the federally sponsored system, with connected status treated separately from assignment. Networks with compatible policies could also interconnect directly. Route acceptance arose from technical configuration and operator policy, not automatically from an identifier’s presence in a registry; preserving uniqueness did not compel a carrier to announce or transport a route.
Federal support was not one policy environment. NSFNET, MILNET, NASA networks, Energy Department facilities and other systems had different missions and access rules, which is why RFC 1174 found that one connected-status field could no longer describe eligibility across them. Its non-US boundary was equally important: the IAB rejected applying US access criteria to all foreign networks and confined the relevant requirements to traffic traversing federally sponsored facilities. Imperfect implementation would not turn that policy record into a claim of unlimited territorial authority.
Commercial alternatives added another boundary. By the early 1990s, providers could sell Internet service under private contracts, although the Merit retrospective’s account of controversy around subsidised infrastructure, ANS and its competitors shows that those alternatives were neither frictionless nor politically neutral. When the NSFNET backbone was eventually decommissioned, the identifiers and their registration records survived and new routing arrangements preserved their usefulness. Federal carriage was no longer the only operating model, and coordination did not depend on the continued existence of one funded backbone.
None of these limits makes federal influence trivial. A theoretically available route may have been expensive, slower or geographically inaccessible. Direct interconnection might reach only a subset of destinations. An applicant dependent on the dominant registration service could face delay even if another network path existed. The record does not contain representative price, availability or applicant-outcome data with which to measure those burdens.
The point is that the dependencies were separable. A network might rely on a federally supported registration service but not on a federal backbone for transit. Another might receive a number through delegated authority yet rely on NSFNET for useful reach. A third might obtain commercial connectivity while its routing policy was recorded in a database established through an NSF award.
That variation is why institutional scope matters. A funding instrument could bind its recipient. A programme rule could govern funded traffic. An operator could control its routes. A registry could allocate or record identifiers within delegated authority. None of those powers, without an additional instrument or dependence, proved the others.
Could registration have been separated earlier?
A plausible alternative would have separated basic registration from backbone eligibility before the 1993 transfer. Organisations needing globally unique network numbers could have approached a registration service without presenting eligibility for a particular federal route. Backbone sponsors would have maintained separate access records, while operators would have decided which routes and traffic to carry.
Part of this alternative was already contained in RFC 1174. The IAB recommended removing connected status from number-registration forms and databases, admitting registered networks to DNS without regard to connection approval, and leaving policy enforcement to network operators. It also proposed international delegation of blocks while retaining central IANA and Internet Registry functions.
The proposal did not establish a free market in interchangeable registries. Initially centralised updates were intended to protect consistency. The existing Internet Registry would remain the default where no delegate existed. Candidate registries would be approved and would operate within allocated blocks.
A more plural service-provider structure might have offered applicants another administrative path when one service was slow or difficult to reach. It might also have made appeals or conflict referral more explicit. But no period instrument examined here establishes an enforceable right to switch providers or appeal a technically valid refusal. Those are analytical possibilities, not observed remedies.
Synchronisation would have been a serious constraint. Multiple centres assigning from overlapping pools could create duplicates. Complete copies of the registration database would need timely distribution over infrastructure that was less mature and less uniformly connected than today’s Internet. Conflict resolution would require a recognised coordinating authority.
Address aggregation created another problem. Independent assignment without disciplined block delegation could scatter addresses and enlarge routing tables. By the early 1990s, routing growth was already a central operational concern. The costs of a fragmented allocation pattern would fall on routers and operators far beyond the registry processing the request.
Administrative capacity was uneven internationally. A distributed model could move decisions closer to applicants, but some centres might lack staff, reliable connectivity or experience. A central expert service could offer consistency and a global view, particularly when the number of qualified operators was small.
Cost also cut in both directions. Federal sponsorship could make registration available without charging each applicant the full cost of the service. Multiple service providers might increase capacity, or they might introduce fees and uneven service. The available record does not support a welfare comparison.
There would also be disputes over policy and hierarchy. A delegate might interpret need differently, allocate too broadly or resist central correction. Operators might distrust records from a weak centre. A conflict mechanism would need authority to quarantine duplicates, correct errors and preserve the integrity of the common database.
The period alternative was therefore bounded delegation, not unrestricted competition. Registration could be separated from connection approval while assignment remained coordinated. DNS inclusion could be separated from routing. Policy information could be collected without turning the registry into an enforcement body. Backbone operators could apply their rules only to the facilities they controlled.
The 1993 InterNIC transition implemented part of that separation institutionally. Non-DDN registration moved to the recipient responsible for Registration Services under a civil cooperative agreement; DDN service remained distinct; the other InterNIC participating organisations handled other information functions. It did not eliminate central coordination, dependence on one registration service or federal involvement. It changed their configuration.
What the record ultimately supports
The dated record shows several distinct forms of federal support between 1977 and 1995. Defense-funded research sustained coordination at USC/ISI, while SRI’s DDN-NIC performed the Internet Registry role described in August 1990. NSF supported the far-reaching backbone through cooperative Agreement NCR 8720904, placed non-DDN registration with Network Solutions through the NREN NIS cooperative agreement, and funded route coordination and the transition to commercial interconnection through its successor architecture.
The documentary evidence is strongest where it identifies instruments, roles and service changes. RFC 1400 gives the 1 April 1993 destination for non-DDN requests and preserves the DDN exception. RFC 1594 separates the three InterNIC organisations and assigns IP-address allocation and ASN work to Network Solutions’ Registration Services. The Merit report identifies the NSFNET cooperative agreement, Merit’s programme role, ANS’s later subcontracted backbone operations and the final routing population. GAO reconstructs the Defense contract chronology while acknowledging the missing earlier instruments.
The evidence remains incomplete at the level of individual outcomes. There is no complete award-to-allocation crosswalk, representative non-US request files are absent, and the cited materials do not show how many applicants were delayed or refused, how many relied exclusively on one route, or how consistently the 1990 recommendations were implemented. Those gaps prevent the observed administrative and operating mechanisms from being converted into a causal total.
The quiet reach of US funding was therefore neither mythical nor absolute. It was strongest where contracts, backbone dependence, route policy and registry practice aligned. It weakened where assignment separated from connection, where foreign or commercial networks found other paths, and where operators asserted their own traffic policies. The surviving record supports a history of conditional leverage—not a history of universal command.