Summary

  • NCFC was commissioned as a domestic scientific-computing and resource-sharing facility linking the Chinese Academy of Sciences, Peking University and Tsinghua University. Its international Internet connection was an unbudgeted extension, not the project’s original deliverable.
  • Before the 64K Sprint line began full-function service on 20 April 1994, NCFC needed both domestic authority to negotiate on China’s behalf and agreement at the US National Science Foundation boundary. Qiheng Hu helped assemble those mandates; engineers and institutions had already assembled the network.
  • Reachability did not finish the transition. Moving the .cn name server from Karlsruhe to China on 21 May separated national naming operations from an external hosting dependency, while the later creation of CNNIC gave the function a more durable institutional home.

A network designed to share a computer

The National Computer and Network Facility of China did not begin as a national Internet-access project. The World Bank appraisal describes a prototype scientific computing facility in Zhongguancun: a mini-supercomputer and a demonstration network connecting the computer centres of Beijing University, Qinghua University and the Chinese Academy of Sciences. The intended output was resource sharing among researchers who otherwise could not reach the same computational capacity.

That original scope matters because it separates two decisions that retrospectives often compress. First came a domestic network joining institutions. Only later did participants decide that this network should connect to the global Internet. The World Bank plan put local-area networks first and the NCFC backbone second. It also divided governance among a Management Committee, an academic committee and a Network Center. Qiheng Hu, then a vice president of the Chinese Academy of Sciences, chaired the Management Committee; she did not personally operate the links or write the routing software.

By December 1993, high-speed optical links and routers connected the three academic networks. Getting there had required a protocol choice across incompatible local systems. TCP/IP, DECnet, OSI and X.25 were all present in the environment. NCFC made TCP/IP its principal architecture while accommodating installed systems, a decision that made full Internet interconnection technically possible without pretending the domestic network was homogeneous.

The physical network was equally untidy. Fibre served a metropolitan backbone, microwave and leased facilities filled gaps, X.25 carried long-distance traffic, and dial-up reached individuals. Engineers dealt with broadcast storms and difficulty importing routers by building and deploying PC-based routing equipment. The eventual 64K international circuit was therefore the last visible segment of a much larger operational assembly.

An international line already existed

Calling 20 April 1994 the moment “China first connected” erases another line. The Institute of High Energy Physics had put a 64K satellite circuit to SLAC into operation on 2 March 1993 using DECnet. Institutional histories describe it as a partial Internet connection through which scientists could exchange email. IHEP later ran TCP/IP over that path in April 1994.

The distinction between partial and full-function service is not a consolation prize. It identifies different scopes. IHEP’s line served a scientific collaboration and a bounded user community. NCFC joined several academic networks and sought general TCP/IP connectivity to the US backbone. A circuit can be international without exposing the same protocols, routes, users or services as another circuit.

This is why the word “first” performs badly here. It asks one chronological question where operators need at least six: Was the line live? Which protocol crossed it? Which destinations were routed? Who could use it? Which applications worked? Who had authority to change the arrangement? The answers changed at different dates.

Ready routers, absent consent

The Chinese Academy of Sciences retrospective says that NCFC’s international connection was a project of choice, not an item in the original assessment plan. Hu recalled it as a bottom-up, volunteer-driven effort outside the NCFC budget, supported by participants and institutional donors. The CAS account also records supplementary funding from the State Planning Commission. These need not be competing origin stories: an unplanned project can combine personal initiative, institutional contributions and later public support.

By the end of 1993, the account says, the technical problems had been solved but the network still did not pass traffic internationally. Domestic telecommunications rules required support for an international leased line. More fundamentally, NCFC needed formal authority to represent China in negotiations. Hu tried to establish a national scientific Internet association; when that route did not advance, CAS sought authorization from the State Council. The retrospective records endorsements from several senior officials.

An authorization on one side did not configure the other. CNNIC’s official timeline traces repeated approaches to the US National Science Foundation: Qian Hualin raised the issue at INET’92, NCFC participants renewed it around INET’93, and the CCIRN discussion drew broad support. In early April 1994, before a bilateral science and technology meeting in Washington, Hu restated the request to NSF on China’s behalf. In her own account, she met Neal Lane and Steve Wolff and reached agreement on a direct NCFC connection to the US Internet backbone.

The careful word is agreement, not admission by an Internet sovereign. NSF managed a crucial backbone boundary but did not own the global Internet. Hu could carry an authorized request but did not command Sprint, NSF or every Chinese institution. The connection became possible because several authorities accepted a common operating state.

What changed on 20 April

On 20 April 1994, the 64K line from the CAS Computer Network Center through Sprint entered operation with TCP/IP. Chinese institutional records call this a full-function Internet connection. Capacity was modest even by the day’s standards, but bandwidth is not the most revealing measurement. The important change was that the circuit, the domestic routes, the protocol choice and the two-sided institutional agreement were finally true at the same time.

That formulation also explains what did not happen. The line did not instantly provide public service across China. It did not create later commercial networks, produce subsequent user growth or settle national Internet governance. Nor did it erase IHEP’s earlier email and research traffic. It changed the state of one interconnected academic facility.

Some institutional accounts call China the seventy-first country to join the Internet; others say the seventy-seventh. The conflict is a warning against treating a coloured map or ordinal rank as the event. Network participation has no single global membership desk. The observable evidence is narrower and stronger: on a dated circuit, a defined network began exchanging full TCP/IP service through an identified counterparty.

Reachability still depended on a name server abroad

The transition had another boundary. The .cn country-code domain had been registered on 28 November 1990, but because China lacked full-function connectivity its name server was hosted at the University of Karlsruhe. That arrangement gave the country an Internet identity while leaving a critical operational function abroad.

On 21 May 1994, a month after the NCFC link went live, the CAS Computer Network Information Center established the .cn server in China with the assistance of Qian Tianbai and Karlsruhe. Qian Tianbai and Qian Hualin served as the administrative and technical contacts. The CAS account also places Hu in the coordination work, inviting Qian Tianbai into the NCFC effort and meeting Werner Zorn about the move. None of that makes the server her personal achievement; it shows how governance work connects people who hold different operational permissions.

RFC 1591, published that March, supplied an unusually apt vocabulary. A country-code manager was a trustee for both the national and global Internet communities, with administrative and technical contacts and a duty of service. The document rejected ownership language. Moving the server was therefore not a territorial trophy. It shifted operational custody while preserving an obligation to a distributed naming system.

CNNIC’s establishment in 1997 provided a later institutional home for national Internet information-centre functions. Hu helped found it and led its steering structure, coordinating government, companies and scholars. That succession did not flow automatically from one 64K line. It answered the next question: how could functions first held together by a project and personal negotiation persist as traffic, users and institutional stakes expanded?

The work at the edge of the cable

Hu’s role is easiest to miss if infrastructure history is told through equipment inventories. She did not replace the engineers who selected protocols, improvised routers, provisioned links and operated servers. Her contribution was to make their technical state legible across boundaries that spoke in different terms: project mandate, national authorization, scientific cooperation, backbone policy and public responsibility.

That is not soft work surrounding a hard network. A route that one party is not authorized to announce, a line that another party will not accept, and a name server whose operator cannot change the delegation are all technical limitations expressed through institutions. Authority is part of the data plane’s precondition even when it is absent from the packet header.

The 64K link became historical because the ends aligned. One end had a network prepared to speak TCP/IP and an authorized representative able to state its purpose. The other had a backbone operator prepared to accept the connection. Between them were carrier facilities, money, routing and staff. Afterward, naming custody and a permanent information centre reduced the dependence on the temporary coalition. The event was not a switch flipped by one pioneer. It was a sequence in which permission became operation and operation became institution.

Sources