Skip to content
Tutorials Center
My learning

Topics

Lesson 7 of 79 ยท Getting Started with Networks

The Structure of the Internet

See how the Internet is built from ISPs in tiers, exchange points, content delivery networks and autonomous systems, and who sets the standards (IETF, IEEE, ICANN).

No single company owns the Internet. It is thousands of independently run networks that agree to exchange traffic using common protocols. Your home router connects to a local provider, which connects to bigger providers, which connect to each other and to the companies that serve most of the content you read and watch. Understanding that structure explains why some sites load faster than others, why an outage at one company can break half the web, and why "the cloud" is physical machines in particular buildings.

This lesson walks from your home outward: the last mile, ISP tiers, exchange points, content delivery networks, autonomous systems, and finally the organizations that write the rules. It extends the picture from What Is a Computer Network? and uses the delay ideas from Bandwidth, Latency, Throughput and Loss.

The last mile

The last mile (or access network) is the link between your home or office and your Internet service provider (ISP). Its technology decides the speed you can buy:

  • DSL reuses telephone copper lines, with speeds that fall as the distance to the exchange grows.
  • Cable (DOCSIS) shares coaxial cable with television, so neighbors share capacity.
  • Fiber to the home (FTTH) brings glass all the way to your door, with symmetric gigabit speeds common.
  • Fixed wireless and satellite serve places where cables are costly; satellites in high orbit add large propagation delay.
  • Mobile networks (4G and 5G) connect phones and hotspots through radio cells.

Access links are usually the slowest and most expensive part of a path, so they are often the bottleneck. Many access links are also asymmetric: much faster for download than for upload, because most households download more than they send.

ISPs and their tiers

An ISP sells connectivity. Because no ISP can run a cable to everywhere, ISPs connect to one another. The classic picture arranges them in tiers:

  • Tier 1 networks are very large global backbones. By definition, a Tier 1 network can reach every other network on the Internet through agreements that need no payment for transit (they only exchange traffic with other Tier 1s as equals). There are fewer than twenty of them.
  • Tier 2 (regional or national) ISPs have their own networks but buy transit from Tier 1 networks for the parts of the Internet they cannot reach directly.
  • Tier 3 (local or access) ISPs mostly serve end customers and buy transit from the tiers above.

Money flows through two kinds of relationship:

  • Transit (customer-provider): the smaller network pays the larger one to carry its traffic to and from the rest of the Internet.
  • Peering: two networks of similar size agree to exchange traffic between their own customers, usually without paying each other, because both gain.
flowchart TB
  T1a["Tier 1<br/>backbone A"] --- T1b["Tier 1<br/>backbone B"]
  T1a --> R1["Regional ISP 1"]
  T1b --> R2["Regional ISP 2"]
  R1 --- R2
  R1 --> L1["Local ISP"]
  R2 --> L2["Local ISP"]
  L1 --> H1["Home and office"]
  L2 --> H2["Home and office"]

Customers pay providers above them (arrows); equals peer (plain lines). Real networks have many more links than this tree.

The tier model is a useful simplification, not a precise description. Large content companies now run their own global networks, and most traffic never climbs to the top of the pyramid, because networks connect sideways at exchange points.

Internet exchange points

An Internet exchange point (IXP) is a physical location, often one or more data centers, where many networks connect to a shared switch fabric so they can peer with each other cheaply. Instead of running a separate cable to each partner, a network buys one port at the exchange and agrees to exchange traffic with many others there.

IXPs shorten paths. Without one, two small ISPs in the same city might send their traffic hundreds of kilometers up to a Tier 1 backbone and back down. With one, traffic crosses town directly. That lowers delay and the provider's transit bill, which is why large IXPs exist in cities such as Frankfurt, Amsterdam and London, and many smaller countries have their own national exchange.

Content delivery networks

Most traffic today is video, images and web pages from a small number of companies. A content delivery network (CDN) puts copies of popular content on servers close to users, inside ISP networks or at exchange points. When you open a video, the data usually comes from a nearby edge server, not from the company's origin data center on another continent.

CDNs help in three ways:

  1. Lower latency. Propagation delay shrinks with distance, so a server 50 km away beats one 5,000 km away. This is the "move the server closer" remedy from the performance lesson.
  2. Less load on the origin. The origin serves each file once to the CDN, and the CDN serves it a million times.
  3. Less traffic on long links. ISPs prefer it, since the data does not cross their expensive upstream connections.

A user is sent to a nearby edge in one of two main ways: DNS answers that depend on where the question comes from, or anycast, where many servers share one IP address and routers send each user to the closest. Both ideas return in Proxies, Load Balancers and CDNs.

Python Why a nearby server wins

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The numbers are lower bounds, because cables never run in straight lines. They still show the point: a request that needs four round trips takes about 360 ms of pure travel time to a far-away origin and about 1 ms to a server in the same city.

Autonomous systems

The Internet is organized into autonomous systems (AS): a network or group of networks under one administration that presents a single, consistent routing policy to the outside world. Examples are an ISP, a university, a cloud provider or a large company. Each AS has a unique number (an ASN), such as AS64500 (a number reserved for documentation).

Inside an AS, the operator chooses its own routing protocol, tuned for speed and simplicity. Between ASes, a single protocol, the Border Gateway Protocol (BGP), exchanges which address blocks each AS can reach, and each AS applies its own business policy: prefer the cheapest route, prefer a customer over a provider, refuse to carry traffic for a rival. This two-level design is what lets the Internet scale: the interior of each AS is hidden, and the rest of the world only deals with a list of ASes. The details are in BGP and Inter-Domain Routing.

The next example models a small Internet as a graph of ASes and finds the path with the fewest AS hops, which is a simplified version of how BGP compares routes.

Python A tiny Internet of autonomous systems

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Through the exchange point, HomeISP reaches the CDN in 3 hops (HomeISP, RegionalA, IXP-City, CDN) and the campus network in 4. When the exchange fails, traffic to the campus still gets there, but it climbs through the Tier 1 networks and takes 5 hops, and the CDN, which connected only at the exchange, becomes unreachable (None). Real BGP picks routes by policy first and path length later, but the effect of redundancy is the same.

What is real here? The tiers, exchanges, CDNs and ASes describe how the Internet is actually organized. The graph in the example is invented to be small, and breadth-first search is a simplification of BGP's route selection. Real captures of routing tables and tools like traceroute are covered later as read-only material.

A short history: from ARPANET to the web

The Internet did not appear all at once.

  • 1960s: packet switching. Researchers in the United States and the United Kingdom independently worked out that chopping messages into packets and routing each one separately would be more robust and efficient than keeping circuits open (see how data travels).
  • 1969: ARPANET. ARPANET (Advanced Research Projects Agency Network) was funded by a US defense research agency. It first linked four university computers using dedicated minicomputers called Interface Message Processors (IMPs), which were the first routers. Its characteristics were packet switching, decentralized control (no single node had to keep working for the others to talk), resource sharing between research sites, and a layered set of protocols. Its advantage was reliable communication that survived link failures; its limits were that it was small, restricted to approved institutions, and had no security design, since users were assumed to be trustworthy colleagues.
  • 1971-1973: email and Ethernet. Email quickly became the most used service on ARPANET, and Ethernet was invented for local networks.
  • 1983: the TCP/IP switch. On January 1, ARPANET changed from its early protocol (NCP) to TCP/IP. This date is often called the birthday of the Internet, because separate networks could now be joined into one.
  • 1980s: DNS and NSFNET. Names replaced host files (see DNS), and the US National Science Foundation built a backbone for universities.
  • 1989-1991: the World Wide Web. At CERN, Tim Berners-Lee proposed linked documents, then built HTTP, HTML and the first browser and server. The first site described the web project itself.
  • 1990s: commercial Internet. ARPANET was retired in 1990, restrictions on commercial traffic were lifted, graphical browsers arrived, and ISPs appeared.
  • 2000s onward: broadband, Wi-Fi, smartphones, the cloud, and IPv6 deployment.

The World Wide Web and web pages

The World Wide Web is a system of linked documents and resources identified by URLs and fetched with HTTP. It is a service that runs on the Internet, not the Internet itself, just as email is. A web page is a document, written in HTML and styled with CSS, that a browser downloads, while a website is a set of related pages. A static page is the same for every visitor and is served from a file. A dynamic page is built on the server (or in the browser) for each request, using data such as the visitor's account.

Typical pages of a site include a homepage (the front door and navigation hub), information pages such as "About" or "Contact", a registration or login page for accounts, and content pages such as articles and product pages. Good pages work on every device (responsive design) and keep their links accessible, meaning link text that reads well to a screen reader and that can be reached with the keyboard.

The Web is often described in stages: Web 1.0 was mostly read-only pages, Web 2.0 added user-generated content and interaction (blogs, social media, web apps), and Web 3.0 is a name for visions of decentralized and semantic data. Challenges of the Web include misinformation, privacy and tracking, security threats, online harassment, and unequal access. How a page really gets to you is the topic of what happens when you type a URL.

Who makes the rules

Nobody governs the Internet from the top, but several organizations coordinate the shared parts.

  • IETF (Internet Engineering Task Force) is an open community that develops protocols such as IP, TCP, HTTP and TLS. Its documents are RFCs (Requests for Comments), numbered and public, for example RFC 791 (IPv4) and RFC 9293 (TCP). Anyone can read them and join the mailing lists. Work moves through drafts until the community reaches "rough consensus and running code".
  • IEEE (Institute of Electrical and Electronics Engineers) writes standards for links and local networks: 802.3 is Ethernet and 802.11 is Wi-Fi.
  • ICANN (Internet Corporation for Assigned Names and Numbers) coordinates the Internet's unique identifiers: the DNS root, top-level domains and, through the IANA function, address blocks and protocol parameters such as port numbers.
  • Regional Internet Registries (RIRs: ARIN, RIPE NCC, APNIC, LACNIC, AFRINIC) hand out IP address blocks and AS numbers in their part of the world.
  • W3C standardizes web technologies such as HTML and CSS, together with the WHATWG; ITU sets international telecommunication standards.

Standards matter because they let devices from different makers work together: a laptop built in one country talks to a server built in another because both follow the same RFCs and IEEE specifications.

Check your understanding

Quick check ยท 5 questions

of 5 answered You scored / 5 ยท perfect!

  1. 1. What is the main reason two ISPs peer at an Internet exchange point?

    Peering lets two networks exchange traffic between their own customers directly, which shortens paths and avoids transit fees.
  2. 2. What is the difference between transit and peering?

    Transit is a paid customer-provider relationship that gives access to the rest of the Internet. Peering is a mutual deal between similar networks to exchange only their own customers' traffic.
  3. 3. Why does a CDN reduce page load time?

    Propagation delay grows with distance, and many protocols need several round trips, so a nearby edge server saves a lot of time.
  4. 4. What is an autonomous system?

    An AS, identified by an ASN, is the unit of routing between networks. BGP exchanges reachability information between ASes.
  5. 5. Which organization publishes the RFC documents that define protocols such as TCP and HTTP?

    The IETF develops Internet protocols and publishes them as RFCs. IEEE writes link-layer standards such as Ethernet and Wi-Fi, and ICANN coordinates names and numbers.

Summary

  • The Internet is thousands of independent networks joined by agreements and common protocols.
  • Your last mile (DSL, cable, fiber, wireless, mobile) is often the slowest hop on a path.
  • ISPs form tiers; money flows by transit (customer pays provider) or peering (equals swap traffic).
  • IXPs let many networks peer in one place, shortening paths; CDNs place copies of content near users to cut latency and origin load.
  • Autonomous systems are the units of routing between networks, joined by BGP.
  • IETF (RFCs), IEEE, ICANN with the regional registries, W3C and ITU coordinate standards, names and numbers; none of them owns the Internet.

Next: Signals, Bandwidth and Transmission Media.