Summary

  • Segal’s 1984 pilot asked whether TCP/IP could connect CERN’s established proprietary computers to its newer open systems. It succeeded technically, and he became the laboratory’s first TCP/IP Coordinator in 1985.
  • CERN’s policy still forbade external TCP/IP connections. A contemporaneous 1988 inventory counted 355 internal hosts, showing that adoption inside an administrative domain was not the same as permission to cross its edge.
  • The January 1989 conversion to officially assigned addresses enabled CERN’s first external Internet links. That sequence separates experimentation, operational support, valid addressing and external authority—and keeps the Internet distinct from the Web created later at CERN.

A network with an invisible wall

The most useful number in the early history of TCP/IP at CERN is not a bandwidth figure. It is 355. A network inventory prepared for the Large Electron-Positron Collider era counted that many hosts offering TCP/IP service on 29 March 1988. The list crossed nearly every boundary that made computer networking difficult: IBM, Norsk Data, Apollo, VAX, PC, Sun, Cray and Macintosh machines all appeared in it. The protocols were used heavily for remote login, file transfer and Remote Procedure Call transactions.

This was no laboratory toy. It was an operating service spanning the old central-computing world and the newer distributed one. But the same service could not be used for an external TCP/IP connection. CERN had allowed an Internet protocol to become useful across its site without allowing it onto the Internet.

That apparent contradiction is the centre of Ben Segal’s contribution. The familiar story of networking treats adoption as a contest in which the best protocol spreads until resistance collapses. CERN’s record shows several gates instead. A protocol could be installed without being officially supported, supported without using globally valid addresses, and valuable on hundreds of systems without being authorised to cross an institutional boundary.

The pilot was a test of coexistence

Segal joined CERN in 1971, when data communications consisted of incompatible manufacturer systems, local inventions and incomplete standards. During a 1977 sabbatical in Palo Alto he encountered ARPANET and Unix. Back in Geneva, he worked in an environment where a physics laboratory had to move information among computers that had not been designed to cooperate.

CERN created a Data Communications group in 1983 with a mandate to unify networking. In practice, the field was divided. The new group concentrated on a CERN-wide backbone and placed formal emphasis on ISO standards, with DECnet as a major exception. IBM mainframe networking, electronic mail, news, Unix and workstation networking remained in other hands. Those were precisely the places where TCP/IP was entering, so support for it grew across an organisational seam.

In August 1984 Segal proposed a bounded experiment. TCP/IP would be installed and evaluated on important non-Unix machines, including CERN’s central IBM-VM mainframe and a VAX running VMS. The question was not whether two similar Unix workstations could exchange packets. It was whether one protocol family could bridge newer open systems and established proprietary ones.

The proposal was approved. Segal’s retrospective history says the work established TCP/IP as a promising answer and Berkeley sockets as the recommended programming interface. That second choice mattered. Interoperability was not only a packet format on the wire; developers needed a stable way to build applications above it. A common API lowered the cost of making the network useful on the next system.

A coordinator without permission to connect outward

In early 1985, a formal agreement between the Software and Data Communications groups made Segal CERN’s first TCP/IP Coordinator. The title did not give him sovereignty over external networking. It created responsibility for a service that still sat inside another policy.

The 1988 inventory makes the work concrete. CERN’s routing was static. The coordinator maintained the master host table, routing information and updates; system managers maintained copies on their own machines. Naming was partly centralised and partly distributed. Adding a platform was therefore not a one-time installation. Someone had to preserve a consistent view of addresses and names across systems owned by different teams.

This was coordination in the literal sense: aligning state held in many places. A successful ping could demonstrate protocol compatibility. It could not keep a host table correct after the network changed. Operational adoption required a role, a maintenance rhythm and cooperation from local administrators.

The hard limit was explicit. Data Communications policy restricted the Internet protocols to use on the CERN site. No external TCP/IP connection was to be made. Segal described external networking as remaining under the prevailing ISO, IBM and DECnet order until early 1989. European governments, telecommunications administrations and industry were backing other standards, and the Internet Hall of Fame account notes that promotion of IP could be politically and professionally costly.

The restriction did not stop internal growth. It changed the shape of that growth. Engineers could prove that TCP/IP connected unlike computers and supported real applications, while the institution deferred the political question of whether those protocols should cross its perimeter.

LEP turned a promising stack into infrastructure

The pilot gained a powerful internal customer in November 1985, when management of the 27-kilometre LEP collider selected TCP/IP for its control system. The later choice of Unix-based systems reinforced it. Segal judged the combination important to LEP’s success.

This decision supplied something a protocol demonstration cannot: dependence. Once an expensive scientific facility expected its controls to communicate over TCP/IP, support could no longer rest entirely on enthusiasts. Reliability, platform coverage and responsibility acquired institutional weight.

The March 1988 count captures the result. TCP/IP was not limited to one favoured vendor or workstation culture; it reached machines from the mainframe centre to distributed systems. Its applications were ordinary operational tools, not just networking experiments. At the same time, the inventory named a support transfer from the Software group to the Data Communications organisation. Later that year, the latter accepted TCP/IP support, turning what Segal called a shoestring operation into a staffed activity.

Technical victory, operational adoption and organisational ownership had finally started to converge. External reachability still had not.

The addresses that marked a policy change

An internal network can use addresses that work only inside itself. A participant on the global Internet needs addresses and routing arrangements that other networks can recognise. CERN’s transition joined these technical and administrative facts.

In January 1989, CERN carried out a coordinated “big bang” that replaced its internal IP addresses with officially assigned ones. Its first external Internet connections followed. The renumbering was costly precisely because internal TCP/IP had already succeeded: hundreds of hosts, configuration files, applications and local administrators now depended on the old state.

Calling the change a connection event alone misses the institutional work. CERN had to accept the protocol family for external use, put operational support in the right organisation, adopt globally valid identifiers and coordinate a site-wide migration. The cable or circuit mattered, but it became useful only after those decisions lined up.

The sequence also corrects a common historical compression. TCP/IP was present at CERN before the 1984 pilot; the pilot focused on heterogeneous non-Unix systems. Internal service was extensive before external Internet connectivity. And the Internet was not the World Wide Web.

The Web came later—and should stay separate

Tim Berners-Lee wrote his first proposal for the World Wide Web in March 1989. CERN’s official history dates the first working Web server and browser to the end of 1990. The proximity to CERN’s Internet opening is important, but it does not merge the two inventions.

Segal helped create an environment in which network and socket programming were familiar, and the Internet Hall of Fame says he pointed Berners-Lee toward RFCs and existing services such as FTP and NNTP. Those are legitimate enabling-context claims. They do not make Segal a co-inventor of the Web, nor do they make CERN’s external Internet connection a sufficient cause of it.

Segal himself drew the boundary. He believed the Web might have emerged earlier had CERN connected earlier, but labelled that counterfactual speculation. What he considered certain was that the Internet created an opportunity. That is a more precise form of credit: he helped make a heterogeneous institution Internet-ready, while Berners-Lee designed a new information system on that substrate.

What the pilot actually proved

The pilot did not prove that protocol selection could erase institutional authority. It proved almost the opposite. TCP/IP could win the technical argument on an IBM mainframe, a VAX and hundreds of other machines while the rule at the edge remained unchanged.

Its deeper achievement was to reduce the risk of a later policy decision. By 1989 CERN did not have to debate external adoption in the abstract. It had applications, trained administrators, a recognised API, an accountable coordinator, a major control-system commitment and a record of cross-platform operation. The closed internal network became a body of evidence.

That evidence still did not decide policy by itself. Someone had to accept external use, official addressing and the operational consequences of connection. The story therefore resists both technological determinism and bureaucratic caricature. Policy delayed one form of connectivity, but the bounded pilot accumulated competence that made the eventual opening safer and more valuable.

Sources