The internet is not one machine but an agreement between millions of networks. This guide shows how packets, addresses, DNS, routers and undersea cables work together, with diagrams, worked examples and the numbers behind them.
The internet is not one machine, one company or one place. It is an agreement. Millions of independently owned networks around the world have agreed to follow the same technical rules, so a message from a phone in Mumbai can reach a server in Oregon in a fraction of a second. No single computer runs it and no single organisation owns it. This guide opens up the machinery behind that agreement: how data is cut into packets, how devices get addresses, how names are turned into numbers, how routers pass data along, and how it crosses oceans through cables on the sea floor. You will also work through the numbers behind it.
People often use the two words as if they meant the same, but they are different layers. The internet is the worldwide network of networks: the cables, radio links, routers and rules that let devices exchange data. The World Wide Web is just one service that runs on top of it, a system of linked pages that you view in a browser. Email, video calls, online games, messaging apps and file downloads also travel over the internet, and many are not part of the web at all.
The word "internet" comes from inter-network: a network joining many networks. Your home Wi-Fi is a small local network. It connects to the network of an Internet Service Provider (ISP), which connects to larger regional and national networks run by other companies, which connect across countries. Where networks meet and swap traffic, they often do so at buildings called internet exchange points. Each network is run by its own owner, such as an ISP, a university, a company or a government agency, yet all of them can exchange data because they follow common rules called protocols.
Who is in charge? Nobody controls the whole thing, but several bodies coordinate pieces of it:
Most activity on the internet follows the client-server pattern. A client, such as your phone or laptop, asks for something. A server, a computer built to answer requests all day and night, usually in a data centre, finds it and sends it back. (Some services, such as certain file-sharing and calling systems, connect devices more directly to each other, which is called peer-to-peer.)
A request does not teleport. It hops from one device to the next. Each device that passes a packet closer to its destination is one hop.
When you send a photo or load a page, the data is not sent as one block. It is cut into small pieces called packets. Each packet has two parts: a header, which carries labels such as the sender's address, the destination address and the packet's position in the sequence, and a payload, which carries a fragment of the actual data. Packets from the same message may take different routes, and may even arrive out of order. The receiver uses the sequence labels to put them back in the right order.
Splitting data into packets is efficient. The millions of messages sharing a cable do not each need a private wire. Packets from many different conversations take turns on the same links, a bit like many cars sharing a road network. It is also resilient, because if one link fails, later packets can go another way.
The rules for handling packets are called protocols, and two common transport protocols make different trade-offs. TCP is careful: the receiver confirms what it has received, missing packets are sent again and the data is delivered in order. It is used for web pages, email and downloads, where every byte must arrive. UDP is lighter and faster but makes no guarantees. It is often used for live video, voice calls and games, where a late packet is worse than a lost one.
A typical network link carries packets of up to 1,500 bytes. About 40 bytes of each is used by headers (a common figure for IPv4 plus TCP), leaving about 1,460 bytes for data. How many packets does a 3 MB photo need? Take 3 MB as 3,000,000 bytes.
Step 1: Divide the size by the data per packet: 3,000,000 ÷ 1,460 ≈ 2,054.8.
Step 2: A packet cannot be part full and still counted as a fraction, so round up: 2,055 packets.
Step 3: What does the labelling cost? 2,055 × 40 bytes = 82,200 bytes of headers. Compared with the file, 82,200 ÷ 3,000,000 ≈ 2.7%.
Meaning: one ordinary photo becomes thousands of small, independently routed pieces, and the labels add only a few percent. The numbers vary with the exact protocols and link, so treat this as a typical case.
Every device on the internet needs an IP address so that packets can be delivered to it, like a postal address. The older system, IPv4, writes an address as four numbers from 0 to 255 separated by dots, such as 192.168.1.10. Each number is stored in 8 bits, so an IPv4 address is 32 bits in total. The newer IPv6 uses 128 bits, written in hexadecimal, to provide far more addresses.
IPv4: 32 bits give 232 = 4,294,967,296 addresses, about 4.3 billion.
Compare: in 2025 about 6 billion people used the internet (ITU), and many own several devices. So there are fewer IPv4 addresses than people online, never mind devices. The central pool of unallocated IPv4 addresses ran out in 2011.
IPv6: 128 bits give 2128 ≈ 3.4 × 1038 addresses. If you share that over 8 billion people, each person could have roughly 4 × 1028 addresses.
A byte in binary: the first number of 192.168.1.10 is 192 = 128 + 64, which is 11000000 in 8 bits.
How the shortage is managed: home and office devices usually get private addresses, such as those starting 192.168, which work only inside the local network. The router shares one public address among them, using a technique called network address translation (NAT). IPv6 is gradually being adopted so that every device can have its own.
People prefer names such as www.example.org to numbers. The Domain Name System (DNS) translates names into IP addresses. It is a worldwide, distributed system rather than one big list. Your device normally asks a resolver (often run by your ISP). If the resolver does not already know the answer, it works down a hierarchy: the root servers point to the servers for the ending (the top-level domain, such as .org, .edu or a country ending), which point to the servers that hold the records for the specific domain, which give the address. Answers are remembered (cached) for a time, so most lookups are instant.
Take www.mail.example.org.
Step 1: The part on the far right, org, is the top-level domain.
Step 2: Moving left, example is the name registered by an organisation. Together, example.org is the domain that someone owns.
Step 3: Anything further left, mail and www, are subdomains, chosen freely by the owner of example.org.
Why it matters: the owner controls everything to the left of the domain they have registered. A scam link such as www.mybank.example.verify-login.example looks like a bank at first glance, but the real domain is the last two parts, verify-login.example (see Cybersecurity Basics).
A router is a device with several connections. When a packet arrives, the router reads the destination address, checks its routing table to find the best next hop, and forwards it. No router knows the whole path. Each one only decides the next step. Routers exchange information with their neighbours, so they can learn about changes and avoid broken links. Between large networks, the protocol used for this is called BGP (Border Gateway Protocol), which is sometimes described as the glue that holds the internet's networks together.
You can see the hops yourself. On many computers the command traceroute (on Windows, tracert) lists the routers a packet passes through on the way to a site, with the time taken for each. Some routers do not reply, so the list may have gaps.
All this data travels through physical equipment. Your device may connect by Wi-Fi, a cable or a mobile network (4G, 5G). Behind that are ISP networks, with fibre-optic cables, copper lines, cable systems and radio masts. Fibre-optic cables carry data as pulses of light through hair-thin strands of glass and are the backbone of long-distance links. Between continents, undersea cables do the work. The ITU and the International Cable Protection Committee report that submarine cables carry more than 99% of intercontinental data traffic, and satellites carry only a small remainder. Cables are damaged by anchors, fishing and natural events, with an estimated 150 to 200 faults every year, so ships are regularly used to repair them and networks are designed with alternative routes.
Data centres, exchange points and the cables between them are all connected in this way, which is why the internet keeps working even when single links fail.
Light in an optical fibre travels at roughly two-thirds of its speed in empty space, about 200,000 km per second. Suppose a cable route between two cities is 10,000 km long. This distance is for illustration.
Step 1: One-way time. 10,000 km ÷ 200,000 km/s = 0.05 s = 50 ms.
Step 2: Round trip. A request goes out and a reply comes back, so the minimum is 2 × 50 = 100 ms.
Meaning: even with perfect equipment, a very distant server cannot respond in under about 100 ms. Real delays are longer, because cables do not run in straight lines, and routers add small delays at every hop. This is why services place copies of popular content closer to users (see Cloud Computing), and why delay, called latency, is a different thing from speed in megabits per second.
You download a 500 MB file on a connection that delivers 50 Mbps.
Step 1: Convert bytes to bits. 500 MB × 8 = 4,000 megabits.
Step 2: Divide by the speed. 4,000 ÷ 50 = 80 seconds.
Step 3: Check against reality. If the connection really delivers only 40 Mbps at that moment, the time is 4,000 ÷ 40 = 100 seconds.
Meaning: the speed advertised for a connection is a maximum. Shared Wi-Fi, busy networks and slow servers can all reduce what you actually get. Bandwidth (how much data per second) and latency (how long a trip takes) both affect how fast something feels.
The rules of the internet are organised in layers, each doing one job and relying on the one below. This is often called the TCP/IP model. When your device sends data, each layer wraps it with its own header, like putting a letter into a series of envelopes. At the other end, each layer unwraps its own envelope.
Web pages are requested using HTTP, and HTTPS adds encryption so people on the path cannot read the content. The padlock shows the connection is encrypted, but it does not prove the site is honest (see Search Engines & Web Browsers). Email uses other application protocols, such as SMTP for sending and IMAP for reading. They are all different "languages" riding on the same transport and internet layers.
For more on how this fits into the wider story of computers, see The History of Computing.
Wi-Fi is only the wireless link between your device and your router. Being connected to Wi-Fi does not guarantee you can reach the internet. Use what you know about the layers:
| Symptom | What it suggests | Sensible check |
|---|---|---|
| No Wi-Fi networks, or the connection drops | A problem at the local link | Move closer to the router, switch Wi-Fi off and on, and restart the device. |
| Connected to Wi-Fi, but no sites load on any device | The router's connection to the ISP, or an ISP outage | Check the router's lights, restart the router, and check whether the provider reports a fault. |
| Works on one device but not another | A problem with that device's settings | Restart it, check its network settings and test on mobile data. |
| Sites by number open, but names do not | A DNS problem | Restart the router, or ask an adult to check DNS settings. |
| One website fails but others work | A problem at that server or site | Wait and try again, and check whether others report the same. |
| Everything works but feels slow | Low bandwidth, high latency or a busy network | Pause big downloads, move closer to the router and test at another time. |
Try each question first, then tap it to reveal the answer. A scored multiple-choice quiz follows below.
The internet is the global network of networks. The web is a system of linked pages that runs on it. Email, calls and games also use the internet but are not the web.
So many conversations can share the same links, packets can take different routes if a link fails, and only damaged or missing pieces need to be sent again.
Packets can arrive out of order, so the receiver uses the numbers to put them back in the right order.
It translates a domain name such as www.example.org into the IP address of the server, using a worldwide hierarchy of servers and caching.
2 MB × 8 = 16 megabits. 16 ÷ 4 = 4 seconds.
One way: 6,000 ÷ 200,000 = 0.03 s = 30 ms. Round trip: 2 × 30 = 60 ms.
Human packet network (no devices needed): write a short sentence, split it into numbered word cards, and send the cards through "routers" (students) to a "receiver" who must reassemble the sentence. Let some cards take a different route, and ask what happens if a card is lost.
Traceroute with an adult: on a home or school computer, run tracert (Windows) or traceroute (macOS and Linux) to a well-known educational website, and count the hops and their times. Talk about why the numbers differ between sites.
Calculation practice: repeat Worked Examples 1, 4 and 5 with different file sizes, speeds and distances.
Safety note: only run network tools on devices and networks you have permission to use, and only against ordinary public websites. Never scan other people's devices or networks.
This article is original EDUSAMBAM writing. The user figures come from the ITU, and the submarine-cable figures come from statements by the ITU and the International Cable Protection Committee. The explanations of packets, addressing, DNS, routing and layers follow the standard internet specifications, called RFCs, including RFC 791 (Internet Protocol), RFC 793 (TCP), RFCs 1034 and 1035 (domain names) and RFC 8200 (IPv6). These are freely available from the IETF. The diagrams are original EDUSAMBAM schematics, and the calculations use rounded figures for teaching.
Statistics are revised and technology changes, so check the latest figures from the ITU and your own provider's information. This article does not recommend any particular provider.
10 questions. Select an answer for each, then submit to see your score instantly.