We actually saw OTL how this kind of thing played out: back before the IP stack was ubiquitous, there were many networks. FidoNet (and other BBSes), Usenet / UUCP, ARPA-ish things like NSFnet, JANET in the UK. Some of these (like JANET) were institutional and lived in the minicomputer world, while others were anarchic clusters of micros like FidoNet. Some would even argue that FidoNet or even UUCPnet were "internetworks", as they connected together various networking domains with different local protocols. But they lacked a global and uniform address space, leading to kludges like UUCP bang paths and BITNET-to-ARPA gateways. This is why FidoNet, despite having had the best part of a decade as the de facto standard for microcomputer networking, never made the kind of mass-market breakout that the (TCP/IP) Internet managed in the early '90s. If you think that it's business use, rather than consumer use, that drives these things, in which case it's UUCP on Unix minis, rather than FidoNet on a million incompatible micros, that's relevant; but UUCP also never became a true internetwork, and for the same reason.
The very name "Internet Protocol" hearkens to the goal of connecting together _everyone's_ network into one big system, where not only can any node send a message to any other node, but (crucially)
all the nodes agree on what a given node is called. Of course, the domain
naming system that's overlaid on this doesn't have to be uniform, but the addressing system really does. (Of course, IOTL there
are multiple Internets; what do you think
RFC1918 is for?)
So, in a TL where Jon Postel gets hit by a bus, all these various networks will soldier on, and will probably continue to grow gateways joining them together. But with addressing being highly source- and path-dependent (having to route across FidoNet to the UUCP gateway, then a series of bang-path hops to arrive at a JANET host, will give a very different-looking address to the one used to reach that same host via an ARPA gateway elsewhere in the JANET), there is no way a global namespace can emerge - which means you may never even form the idea of separating addresses from names. It will be impossible to create something like the Web, because —
no hyperlinks. How can I link to someone else's site, when I don't even know
how to address it from the reader's machine? The nearest equivalent you can get is hyperlinks within one networking provider's customers; far more of a walled garden than AOL or Compuserve ever created. Even other providers on the same network can't crosslink, because the address will be different. There will be various technical attempts to solve this problem in the client software — an automated version of "...{{bigsite,otherbigsite}!foovax,barvax}!utzoo!henry" UUCP addresses (an inverted tree of routes from well-known sites to the user's local machine) — but they will be fragile and flaky, and every time the network topology changes for technical or political reasons (University of Foo notices their CS department's huge phone bill and shuts off foovax's phoneline) old hyperlinks cease to be followable. A further problem is that many of these networks are store-and-forward message-routed systems, piggybacked on the telephone network not just for the last mile but for the backbone too. The latency this produces (Usenet articles commonly took multiple
days to propagate) means the kind of interactive browsing we take for granted today would rely heavily on caching proxies which would only be practical for a small number of popular *websites; most content distribution would follow the flood-fill method as used by Usenet.
With such limited applicability for Web-like technologies, probably no-one will put much effort into trying to develop them; in this world Gopher, far from being an historical footnote, is a technophile's plaything — something many people
wish they could have, if only they weren't stuck on a multi-hop store-and-forward network, or just a horrendously underprovisioned walled garden. ("Those lucky Californians, with ARPANET nodes on every campus, I hear they're even managing to share
video over Gopher now.") No-one bothers with a massive build-out of backbone networking capacity because — without the Web, no killer app. Large companies will have their own cross-country network links, but those probably won't even be gatewayed to any of the 'public' networks (every sysadmin has seen WarGames, and besides, our network is for Serious Business Business, not for programmers in Palo Alto to read alt.*).
(Acknowledgement: This dystopian timeline is partially inspired by Eric Raymond's
World Without Web.)
Would the idea of the global address emerge? Eventually, probably; some of the Unix hackers at Bell Labs were
complaining in 1985 about the networking namespace botches and comparing it to the elegant uniformity of the Unix file system namespace (trees, rooted on mounts, with a single simple syntax). But even they appear to have been advocating simply a more regular and standard way of expressing gateway-hopping (in the filesystem analogy, gateways are mounts). On the other hand, I wouldn't like to bet against Unix hackers, and it seems they would have acquired sufficient clue sooner than others. If internetworking came out of a Bell Labs summer project, rather than being an external invention grafted on by the tasteless folks at UC Berkeley, we might even have a proper Unixy sockets interface (where everything really is a file). But it seems a small consolation for an Internet delayed by perhaps as much as ten years.