The internet seems immaterial to us, and yet it is made of cables, computers and shared rules. Having some understanding of these technologies may seem pointless. Yet it seems to me that, without trying to become a specialist, this understanding helps us make far better choices. In October 2025, this is where I began the series of webinars I ran for associations on behalf of the Fondation Afnic: how data travels, how HTML structures a page. Here I go back over these basics in more depth, taking as my example the journey of a page from this site, from the server in Clermont-Ferrand where it is hosted to your screen. I end with a few guidelines for choosing an organization’s tools.
Why know what the internet is made of
When we publish a text, a photo or a video on the internet, we see neither where it is stored nor which way it travels to reach someone else’s screen. We even say that our files are “in the cloud”. In 2023, in Digital is not immaterial, I spoke in this connection of a “magical world of data”: it is very convenient, but it leads us to rely on the people who seem to know, for instance the contractor who finds the site old-fashioned and advises rebuilding it. Does that mean we all have to become computer scientists? Of course not. In my view, it is enough to know what the internet is made of, how data moves through it and who sets its rules, in order to ask precise questions and judge for ourselves, especially when it comes to what has to last.
In 2012, almost all web documentaries relied on Flash, an add-on to browsers (the programs we use to view the web, such as Firefox, Chrome or Safari) owned by a company, Adobe. At the time I wrote, in The technology behind web-documentary, that they would become unreadable when Flash disappeared. Flash finally disappeared on December 31, 2020, and with it all those sites, those web documentaries, those interactive works, which became unreadable because they depended, as the saying goes, on a proprietary system, that of a software publisher. The very first web page, created by Tim Berners-Lee and put back online in 2013 by CERN, the laboratory where the web was invented, still opens in every browser today. This is where we can see the strength of open and free technical standards, what are called “digital commons”. The problem is that commercial offerings, precisely because they are commercial, know how to charm: thanks to marketing, they look more convenient, and sometimes more widespread. But they are bound to be much less durable.
Computers linked by cables and radio waves
A network is a set of computers linked together to exchange data. The internet is a network of networks. The page you are reading right now, for example, leaves a server belonging to the hosting company o2switch, in Clermont-Ferrand, crosses that host’s network, then the network of your internet service provider (Orange, Free, SFR, Bouygues Telecom, etc., the company you pay for your subscription), which is connected to hundreds of other networks at a great many points. Hence its name, inter-net, “between networks”.
This network is made of physical objects, as I explained in Digital is not immaterial. They fall into three kinds:
- Cables and radio waves carry the signals. Optical fibers and copper cables under the streets, cables laid on the seabed between continents, mobile phone antennas (4G, 5G), wifi in our homes. According to TeleGeography, a research firm specializing in telecommunications, 607 submarine cable systems were in service in September 2026, and 102 more were planned.
- Devices are used to view content. Computers, phones, tablets, televisions, etc.
- Computers store content and send it to whoever asks for it. These are called servers. They are usually installed at hosting companies, whose business is keeping computers running day and night, in large buildings called data centers. This site is hosted by o2switch, a French company that owns its own data centers in Clermont-Ferrand. The Seine-Saint-Denis department alone had twenty-seven of them in the summer of 2026, as I reported in Data Centers : We Fill Them Ourselves.
Like any physical object, they can be damaged. On March 14, 2024, four submarine cables (ACE, SAT-3, WACS and MainOne) were cut off the coast of Côte d’Ivoire, most likely by an underwater landslide; according to the Internet Society, an international association devoted to the development of the internet, access was disrupted in thirteen countries, and almost entirely cut off in Côte d’Ivoire.
An IP address for every machine
Your phone, your internet box and the o2switch server come out of different factories, often on different continents. How do these machines manage to exchange data? Through shared rules, called “protocols”. Since April 7, 1969, the date of the first of them, these rules have been published in numbered documents, the RFCs (Requests for Comments), which anyone can read online for free, and hardware manufacturers and programmers all over the world apply them in the same way.
The first of these rules is that every machine connected to the network must have a unique address. It is called the IP address, for Internet Protocol. Perhaps you have already seen one in your internet box’s settings? In the original version, IPv4, described in September 1981 in RFC 791, it is written as four numbers from 0 to 255 separated by dots: the address of benoitlabourdette.com, for example, is 109.234.164.81. That makes a little over four billion possible addresses, which seemed almost infinite in 1981, when only a few hundred computers were connected to the network. Yet the organization that distributes them in Europe, the RIPE NCC, allocated its last ones on November 25, 2019. The new version, IPv6, with much longer addresses, has existed since 1995 and is gradually spreading: in France, according to the telecoms regulator Arcep, its usage rate exceeded 62% in 2023, which put the country in second place worldwide. Why has it taken so many years? Because all the equipment that carries data from one point of the network to another has to be updated (the routers, which I come to below), and two machines that do not speak the same version cannot communicate directly. It seems very slow, but this slowness is the price of reliability.
Data travels in packets
Everything that circulates on the internet (a text, a photo, a sound) is first turned into a sequence of 0s and 1s, as I explain step by step in Digital is not immaterial. To send it, the computer cuts this sequence of 0s and 1s into “packets”, pieces of no more than 12,000 bits, that is, 12,000 digits, each a 0 or a 1. A 3-megabyte photo taken with your phone is thus split into about two thousand packets. Each packet carries the IP address of the machine it is intended for, a little like the address we write on an envelope. If you send the photo to your grandmother through a messaging app, that machine is first the messaging service’s server, which will pass it on to her phone.
The packets then go through “routers”, computers installed at internet service providers and at the points where networks meet. Your internet box, for example, is a small router. Based on the address written on each packet, each router passes it on to the next router, closer to the destination. So the two thousand packets of the photo do not necessarily take the same route, and the computer that receives them puts them back in order. And if a cable is cut, as off the coast of Côte d’Ivoire in 2024? The packets go through other cables, if there are any left. A second protocol, TCP, checks that all the packets have arrived and asks again for the missing ones, hence the name “TCP/IP” often given to the two together.
The first message sent over a network of this kind dates from October 29, 1969. At the University of California, Los Angeles, the student Charley Kline tried to send the word “login” to a computer at the Stanford Research Institute, which crashed after two letters: the first message was “lo”. This network, ARPANET, was funded by the research agency of the US Department of Defense, ARPA, and built with universities. In 1964, Paul Baran, an engineer at the RAND Corporation, had published in On Distributed Communications the plan for a network able to survive destruction. ARPANET’s designers always said they had not built their network to withstand a nuclear war, but they took up his principles; I told this story in A specialist in your pocket. In May 1974, Vinton Cerf and Robert Kahn published a paper describing a protocol that allowed different networks to exchange packets. This protocol was split in two in 1978, TCP and IP, and ARPA made TCP/IP mandatory on all ARPANET machines on January 1, 1983.
Domain names, so we don’t have to remember addresses
Who could remember 109.234.164.81? So domain names were created, such as benoitlabourdette.com. When you type this name and press Enter, or confirm it on your phone, it is sent to the DNS (Domain Name System). The DNS is a directory, duplicated on thousands of servers around the world, which simply converts the domain name into an IP address, that of the server where the site is hosted. In the 1970s, in the days of ARPANET, a file called hosts.txt, also copied onto every machine, already listed all the connected computers and their IP addresses. It was in 1983, when a little over five hundred computers were connected to this network worldwide, that the computer scientist Paul Mockapetris proposed creating the DNS directory.
So the domain name and the hosting are two separate things. You could perfectly well register a .fr name and, with the right software, host your site on your own computer. For my part, to delegate maintenance and gain reliability, I use a hosting company, o2switch, which takes care of both my domain name and the hosting of my site.
We can also see that a domain name is not bought: it is rented, usually by the year, from a “registrar”. For .fr names, it is Afnic that has managed the extension since 1997, on the technical, legal and financial levels. But you do not register your .fr directly with Afnic: you go through a registrar, such as Gandi or OVH, or through your hosting company, which work with Afnic. That is why these institutions are so little known: we are never in direct contact with them.
I learned this with filmerlesmots.fr, the site of “Filming the words”, the participatory creation project I launched in 2011 with the Délégation générale à la langue française et aux langues de France, the French Ministry of Culture’s department for the French language. I did not renew the rental in 2016, and someone else immediately registered the name: Afnic’s registry, which anyone can look up online, shows a new registration on December 2, 2016, by an anonymous holder. I now have to watch for the moment this person stops using it and does not renew the rental, so that I can take it back, or else buy it from them; but domain names are then negotiated almost as if at auction, and it can be expensive. In the meantime, the links that my articles from 2012 to 2015 make to this site lead nowhere. I go into these questions in more detail in Domain name, hosting, online platform: who owns your association’s website?.
Clients and servers
When you sit down in a café, you are the customer: from your table, you call the waiter and place your order, then he goes to fetch your coffee, or your chocolate profiterole, and comes to serve you. Computing has taken up this image, in a very technical sense (in French, the same words, “client” and “serveur”, are used for the café and for computers). Your computer, when it views a site, is the “client”: it asks another computer, identified by its address, to send it a page, for example the site’s home page. That other computer is the “server”, which sends it the page. For this article, the server is o2switch’s, in Clermont-Ferrand.
In reality, every computer is both client and server: to receive the page, yours had to send its request. The client gives something in order to receive something else in return. Benjamin Bayart, then president of the non-profit internet service provider French Data Network, made this point in July 2007 in Amiens, in his talk “Free internet or Minitel 2.0?”: the internet is “a network of computers that are all both clients and servers”.
NAS devices, which I recommended to small organizations in 2023 in Digital is not immaterial, are connected hard drives, switched on permanently at home or at the office and accessible over the internet. They let you have your own cloud, without going through a hosting company, or through Google or another of the big tech platforms, and also, if you wish, host your own website. But you then become technically responsible for it: if there is a power cut at home, for example, the site is no longer accessible at all. This is what happened, on a completely different scale, when a data center belonging to OVH, the leading European hosting company, burned down in Strasbourg on the night of March 9 to 10, 2021: according to the firm Netcraft, 3.6 million websites became inaccessible from one moment to the next. And the people who had not made their own backups, who had no copy of what was at OVH, lost their site forever.
Protocols, rules stacked in layers
IP, TCP and DNS already make three protocols. The web adds a fourth, HTTP, which Tim Berners-Lee defined at the same time as HTML: it is through HTTP that your browser asks the o2switch server for a page and that the server sends it back. You have probably noticed that site addresses usually begin with “https”: the “s” means that the exchanges are encrypted, that is, made unreadable to anyone who might intercept them along the way, thanks to another protocol, TLS. Email uses yet others, SMTP to send a message and IMAP to read it, which you may have come across in your email settings.
These protocols are stacked on top of one another, each relying on the one below. Computer scientists describe seven levels in their reference model; I suggest a simpler breakdown, into five layers, which is enough to find one’s way around:
| Layer | What it does | Examples | Where you come across it |
|---|---|---|---|
| Physical | carry the signals | fiber, copper, wifi, 4G, 5G | the internet box, the cable, the phone |
| Network | give each machine an address and route the packets | IP (IPv4, IPv6) | a server’s IP address |
| Transport | split into packets, check, put back in order | TCP | rarely visible |
| Encryption | make exchanges unreadable along the way | TLS | the browser padlock, the “s” in https |
| Application | provide a specific service | HTTP and HTTPS for the web, SMTP and IMAP for email, DNS for names, SFTP for uploading files to a server | a site’s address, the settings of an email account |
What is the use of knowing all this? To read error messages and quotes, for example. An “expired certificate” concerns encryption: the certificate is the site’s identity card, which makes it possible to set up the encrypted connection, and when its validity date has passed, the browser displays a warning. A “DNS problem” concerns the directory of names: the site exists, but its name no longer leads to it. “SFTP access” is a way of uploading files to the server. And since email and the web are two separate uses, which rely on the same underlying layers, you can change the host of your website without changing your email, and vice versa.
The web, one use of the internet among others
We often say “the internet” when we mean the web. But when you send an email from your phone’s mail app, are you on the web? No: you are on the internet, and your message goes out through another protocol, SMTP. The same goes for video calls, television through the internet box, many apps, etc. The web is just one use among others: pages, which we view with a browser and which are connected to one another by links.
It was invented by Tim Berners-Lee, a British computer scientist at CERN, the European particle physics laboratory near Geneva. In March 1989, he proposed to his management a system so that researchers could share their documents; he revised his proposal in May 1990, and in December 1990 the first web server and the first website were running on his computer. On August 6, 1991, he presented the project publicly on an internet discussion forum, alt.hypertext.
His idea was to apply hypertext to the internet: a text in which certain words are links, which you click to reach another page, on the same site or on a site at the other end of the world. To write these pages, he defined a language, HTML (HyperText Markup Language), in which small marks, tags, tell the browser that this sentence is a heading, that this word is a link. A web page is therefore a simple text file. Berners-Lee’s first program, WorldWideWeb, was used both to read and to write pages; a team gathered at CERN in February 2019 rebuilt it so that it can be tried online.
What happens when you type an address?
With these notions, we can follow what happens, in less than a second, when you type an address such as https://www.benoitlabourdette.com into your browser:
- The browser asks the DNS for the IP address that corresponds to the domain name.
- It opens a connection with the server that has this address (TCP protocol).
- The browser and the server exchange a key to encrypt what follows (TLS protocol); the padlock appears.
- The browser requests the page (HTTPS protocol).
- The server sends back the page, a text file written in HTML, split into packets.
- This file also contains the address of the images, of the style sheet (the file that describes the colors and the layout) and of the fonts, which the browser then requests, sometimes from other servers.
- The browser puts it all together and displays the page.
Each of these steps can fail, and when you know which one, you know whom to turn to: whoever manages the domain name if the DNS cannot find the address, the hosting company if the server does not respond. You can see the HTML file from step 5 for yourself: right-click on an empty area of the page and choose “View page source”. The keyboard shortcut is Ctrl+U, and Cmd+Option+U on a Mac (in Safari, you first have to enable the Develop menu in the settings). I show how to read it in Ctrl+U: viewing a web page’s source code to see it as search engines receive it.
A common good, placed in the public domain in 1993
On April 30, 1993, CERN’s management signed a document placing the web software in the public domain: anyone can use it, copy it, modify it, without paying anything or asking anyone. Robert Cailliau, the CERN engineer who carried the project with Berners-Lee, recounted this choice twenty years later:
“Finally, as we were more interested in the excitement of making something useful than in getting rich, we decided to use the traditional CERN model for technology spin-off: make it freely available.”
Robert Cailliau, “Twenty years of a free and open www”, 2013.
The web then spread very quickly: Matthew Gray, then a student at MIT, counted 130 sites in June 1993, and about 100,000 in January 1996. In October 1994, Berners-Lee founded the World Wide Web Consortium (W3C), which brings together companies, universities and associations to agree on the shared rules of the web, what are called standards. Since February 2004, participants in its work have committed to granting royalty-free licenses on the patents needed to implement these standards. As for HTML itself, another group has published its reference version, freely available, since an agreement with the W3C in 2019: the WHATWG, founded in 2004 by browser makers (Apple, Mozilla, Opera). So anyone can build a browser, a website or website-building software without paying anyone, and a page written in HTML can be moved from one host to another and read on any device.
Six guidelines for choosing your tools
For an organization that has a website, I suggest six guidelines:
- Prefer open standards to the technologies of a single company. The first web page, written in HTML, still opens more than thirty years later, whereas web documentaries made for Flash no longer open.
- Be the registered holder of your domain name yourself. It is the address everyone knows, the one used for the organization’s emails, and the one that links and search engine results point to.
- Know where the server is. Ask your contractor which hosting company your site is with, in which country, and who makes the backups; the OVH fire showed what it costs not to have a copy, and I come back to it in Digital is not immaterial.
- Check that the site uses https. The padlock shows that exchanges are encrypted. Since July 2018, Chrome has displayed “Not secure” next to the address of a site that does not use it, which rightly worries visitors. Yet since 2016, the non-profit Let’s Encrypt has been issuing certificates for free, and most hosting companies install them automatically.
- Tell the web and platforms apart. Facebook and Instagram belong to Meta, which sets their rules and can change them whenever it likes, as in January 2025, when the company ended its fact-checking program in the United States. This site, on the other hand, could leave o2switch tomorrow without changing its address, and it can be read with any browser. Social networks remain useful for making an organization known, but I believe that what has to last belongs on its own website.
- Think long term. Before adopting a tool, ask the person who suggests it two questions: will what we put into it still be readable in ten years? Will we be able to get it back and install it elsewhere?
