It is important to realise the difference between having a computer connected to the Internet and a computer which is connected in other ways. (See the "How to" section) The essential aspect of the Internet is that small packets may be sent from any computer to any other computer very quickly. By small, I mean a few hundred characters, and by quickly, I mean to the other side of the world and back within a second. Other networks apart from the Internet can do this but none have grown with such staggering speed to global proportions. Before these "packet switched wide-area networks" were available, computers had to communicate by dialling each other up, or over a dedicated cable between two specific computers. Messages were larger and longer, being stored on disk for some time before it was convenient to send them on.

Another practical facility which has boosted the networked information world is that disk space became sufficiently cheap for interested system managers to keep fairly large quantities of information on their machines, and provide it to the world. In the beginnings of the web, much of the data has been picked up from one source or other, chopped about into a usable form, and left on a server somewhere. Examples of this are the "hacker's dictionary" which has surfaced as hypertext in a number of forms, and various out of copyright works, such as old dictionaries and Roget's Thesaurus. This created a body of information sufficiently large and diverse to indicate what could be done. Soon the serious information followed.

The fact that the web first appeared at CERN, the European Particle Physics Laboratory, is not surprising in the light of these factors. The smaller the particles which physicists investigate, the larger the accelerators they have to build. In order to afford the accelerator and the detectors, groups combine from all over the world. The physicists form a widely dispersed computer-literate group. They also happen to have good networking facilities between their institutes for the exchange of physics data. The particle physics community therefore formed an ideal substrate for the growth of the web.

Fred Brooks, in "The Mythical Man Month", suggests that systems end up with a similar structure to the organizations which design them (rather as dogs seem to some times look like their owners!). CERN's mixture of people from many different institutes was echoed by a diversity in the programming languages, styles of design, and brands of hardware which were in use. Most of the subsystems had been designed by physicists rather than software engineers, giving a wealth of ingenious, if unconventional, techniques.

Of the (aproximately) 10,000 people in the CERN phonebook, only around 3,000 are actually salaried CERN staff. [@@ check] Many of the others have a desk at CERN and also one in their home institute, and they visit every now and again. This makes it imposible for anyone to make global decisions on types of hardware and software to be used. People bring with them machines and customs from their home universities. You just have to do your best to accomodate them. This had already led to a lot of work on mail and file interchange between different types of computer system and different networks.

When I first arrived at CERN in 1980 as a consultant, I found its structure fascinating. I wrote a small rather simple hypertext program ("Enquire") with which to note down all the dependencies between people, programs, and hardware. It was a bit like a card index, but with links between cards. It ran on a plain 24x80 terminal. The links had meanings like "A is part of B", or "A made B". The idea was that when one had put in everything one knew, one could trace through implications, answering questions such as "If I remove this module, what stops working?"

A surprising fact was that Enquire only needed a dozen or so different link relationships.

Looking around CERN, it was clear that if I needed the web, then others, especially the experimental teams, must be desparate for it. In 1989, CERN was in a lull between two large pushes: one the LEP (Large Elecron Positron) accelerator was running and the LHC (Lage Hadron Collider) was a few years off. I suggested an as yet unnamed hypertext project and, after a year of writing memos and proposals which largely fell on stony ground, my boss at the time, Mike Sendall, said (more or less) "why not?".

At the same time, NeXT Inc was producing the first NeXT cubes, which provided an ideal prototyping platform. Again, Mike Sendall suggested we pick one up. When you are trying out the user interface to a hypertext system, the last thing you want is to be involved in great depth with details of writing characters onto screens, font metrics, and window exposure events. The NeXTStep programming environment came with a multifont editable text object which could be fairly simply converted into a hypertext object. The prototype was working as a browser after about a month, and as an editor after another month. This takes us to Christmas 1990. (For comparison, most of the X browsers took around year to be developed. More recently, Omni corp. decided to put together a new W3 browser "OmniWeb" using NeXTStep, and starting from scratch it likewise took them them a month.)

Efficient though it was as a development platform, the NeXT was not installed in any quantity, so any real use needed versions for the PC, Mac, X workstations, and "dumb" terminals. However, to attract any resources I also needed a good visible reason to be doing this at CERN.

CERN had seen many in-house documentation systems rise and fall. It would clearly be a mistake to design the ultimate system and then hawk it around suitable users. No one wiould put any information into it, simply because no one else uses it. Similarly, no one will look into it for data because there isn’t any. The hope was that once the ball was rolling, increasing data would bring a wider user base, which would in turn encourage new information, but an empty system had no appeal for anyone.

Fortunately there was a way to bootstrap the web by making gateways into existing systems. Making gateways into virtual worlds is still a very interesting and useful technique in its own right, but at that stage it was the seed of the web’s existence. The first target data was the CERN phone book, a database sitting on the ageing mainframe and whose maintainer, Bernd Pollermann, was charged with providing to users on their favorite systems. Bernd helped set the old VM machine up as a server to a simple hypertext version of the data. Nicola Pellow, a visiting student, tackled the client side by providing the line mode interface. This was so basic in its assuptions about the terminal type so that no one could claim that there was no client interface available on their system. Unlike the NeXTStep application, the line mode browser (never given any other name) could not edit hypertext: it could only read . It was good enough for the phone book and for the hypertext documentation which we generated for the project. Robert Cailliau, a great enthisiast for accessable no-nonsense technology, joined me and looked at providing a program for his favorite, the Macintosh.

So this was the way the the web was first presented at CERN: as a rather simple user interface which allowed one to look up (mainly) telephone numbers. It was to be two years before it would shed that image. Accepted as a rather quirky information system by people who didn’t see why a simple ad hoc program for getting the phone numbers wouldn’t have done just as well, the W3 may have stayed confined to that role, had it not been for the Internet.

When the basic server and the NeXTStep and line-mode clients were available, I made them available by FTP and mentioned this rather limited software on a usenet newsgroup, alt.hypertext. There was not much to show (unless you had a NeXT), but the great thing about usenet is that there are so many bright and weird people out there, that if you put any idea the right way and in the right place, the chances are fairly high that somone to whom it makes sense will pick it up. The project started getting feedback from people who saw it as a rather simple version of Hypercard(TM), an over complex information system, a frustratingly inconsistent database, or an idea which appealed to some latent excitement about how things could be. From then on, interested people on the Internet provided the feedback, the stimulation, the ideas, source code contributions, and the moral support which it would have been hard to find locally. The people of the internet built the web.

Just as W3 "rescued" the CERN phone book from the prison of a mainframe, Louise Addis, the librarian at the Stanford Linear Accelerator (SLAC) saw it as a godsend for their rather sophisticated but mainframe-bound library system. Under Louise’s encourgement, SLAC became a model site, using interesting W3 techniques, and providing information of real value to physicists.

The next year or two required a certain amount of selling of the idea of the web to the High Energy Physics (HEP) community. Even given that, Robert and I were taunted with the fact that W3’s use in CERN itself was very low. We trod a fine line between dedicating our resources to local user support but neglecting the project as a whole, and pursuing the web’s goal of an interactive seamless information world but being balled out for producing nothing useful.

At conferences on networking, hypertext, and software for HEP, we would point out that we really needed clients for the PC, Macintosh, and X, and that these would make useful student projects. At CERN, I was under a lot of pressure to make an X client. In the end, our strategy of finding volunteers paid off. Marc Andreessen, a student at the University of Illinois at Urbana Champain (UIUC) decided to pck up the X client as his project. Working with Eric Bina at UIUC’s National Center for Supercomputer Applications, Marc put in the work of three men, driven by a vision of what could be done and nourished by large quantities of expresso. The result was "Mosaic", an X application which one could pick up and execute, and which would give an immediate point and click access to the web. NCSA soon started other projects to get the PCs running Windows and Macintoshes onto the web. In September 1993, Mosaic was available on all three platforms. The vision of W3 as a rather crude line mode program was gone, and the world started to know W3 as Mosaic.

While Marc is rightly credited with having broght W3 out into the light, in fact many others produced equally exciting, if not so easy to install nor well deployed, programs. Pei Wei’s "Viola" at Berkeley rivalled "Erwise" from a team of students in Helsinki for the prize as the first X browser, both very crediable programs. "Midas" from Tony Johnson at the Standford Liner Accelerator Laboratory (SLAC) was a later product which is currently in use in SLAC and collaborating labs as an alternative to Mosaic, with a similarly user-friendly interface, and a different set of features.

There are a number of ways to measure the growth of the web, none of them definitive. The earliest figures available to us was the load on our server at CERN, expressed in documents fetched per day. For two years, this showed a steady exponential increase, doubling faster than every four months -- in fact, each time I counted the log for a month, it was roughly 10 times the size of the same month last year. This server, info.cern.ch, held mainly information about the web itself, so the rise in interest in it may have been an indicator of future developments.

If the total use of the web can be expressed as the usage of a average server multiplied by the number of servers, then some statistics on the number of server would clearly be useful. However, one of the factors enabling the growth was the lack of mandatory central registy of servers. Though we do keep an optional list of servers, I do not unfortunately have acuurate history of it. Here are a few points we noted down:

Occasionally I would run a program which would trace two levels of links down from each registered server. The program would typically find just as many unregistered servers as there were registered servers.

@@@

When the global traffic began to become significant, another interesting figure was the port 80 TCP traffic across the NSF backbone. Merit Inc keep such statistics by sampling the traffic through routers. Port 80 is the port number for HTTP servers. Not all servers run on this port, and of course WWW clinets also use other protocols, but the port 80 traffic is some indication of the growth of the web. Rising from 0.1% of the total traffic in March 1993 to 1.0% in September and 2.5% in December, the web was growing at a rate unprecedented even in Internet circles.

Figure. ...@@@

As @@@ put it in an article in March 1994, <quote>.