How the internet works in brief
The internet is a network of networks. Thousands of independently run networks, belonging to internet service providers (ISPs), cloud companies, universities and businesses, exchange traffic using the Internet Protocol (IP). Data travels in small packets, each addressed to an IP address, and routers pass every packet from network to network until it arrives.
Loading a web page adds a few protocols on top of IP. DNS turns the site's name into an address, and TCP or QUIC carries data reliably between the two machines. TLS encrypts that data, and HTTP carries the request and the page.
Nobody runs the internet as a whole. Each network decides which neighbors to connect with and which routes to accept, and open standards from the IETF keep them compatible.
Networks of networks: how the pieces connect
Each of these networks is called an autonomous system (AS) and has its own number, an ASN. A broadband provider, a mobile carrier and a large cloud platform are each one or more autonomous systems.
Autonomous systems connect in two main ways. In peering, two networks exchange traffic between their own customers, often at an internet exchange point (IXP) where many networks meet in one building. In transit, a smaller network pays a larger one to carry its traffic to the rest of the internet.
Networks tell each other which address ranges they can reach using BGP, the Border Gateway Protocol. Physically, the links are fiber-optic cables, including cables laid across ocean floors between continents, with copper and radio (Wi-Fi and cellular) covering the last stretch to your device.
Packets, IP addresses and routing
Data crosses the internet in packets. On most links a packet holds at most 1,500 bytes, the standard Ethernet maximum. Every IP packet starts with a header. It holds the source and destination addresses, a hop limit, and a number for the protocol inside, such as TCP or UDP.
Routers forward packets one hop at a time. A router looks up the destination address in its routing table, picks the most specific matching route and sends the packet to the next router on that path. No router knows the whole path, only the next step.
Each router lowers the hop limit (called TTL in IPv4) by one and discards the packet when it reaches zero. That stops packets circling forever when routes form a loop, and the traceroute tool uses it to list the routers along a path.
IP is best effort. Packets can be dropped when a link is congested, arrive out of order or occasionally arrive twice. IP does nothing about it; recovering is the job of the transport protocol above. Routes change when links fail or networks change their policies; routers do not pick a fresh path for each packet.
What happens when you load a web page
Opening https://example.com with nothing cached runs through these steps, from the network up:
- Your device already has an IP address, usually a private one handed out by your router with DHCP. For IPv4, the router translates that address to its own public one with NAT on the way out.
- The browser asks a DNS resolver for example.com's addresses. An A record holds an IPv4 address and an AAAA record holds an IPv6 address.
- The browser opens a connection to that address on port 443. Over TCP, the three-way handshake costs one round trip; if the browser knows the server supports HTTP/3, it uses QUIC over UDP instead.
- TLS checks the server's certificate against the hostname and sets up encryption keys. With TLS 1.3 over TCP this adds one more round trip, while QUIC folds it into its own handshake.
- The browser sends an HTTP request for the page, and the server replies with a status code, headers and the HTML.
- The browser reads the HTML and requests the stylesheets, scripts and images it references, reusing the open connection where it can.
Where the time goes: latency and bandwidth
Two numbers describe a connection. Bandwidth is how much data it can carry per second, and latency is how long a packet takes to make a round trip. Page loads are often limited more by latency than by bandwidth, because each step above waits at least one round trip before the next can start.
Distance sets a floor on latency. Light in optical fiber covers roughly 200 kilometers per millisecond, so every 1,000 kilometers between you and a server adds about 10 milliseconds to each round trip. That is why CDNs keep copies of content close to users.
Latency also climbs when a link is busy. Routers and modems queue packets they cannot send yet, and oversized queues (bufferbloat) can make calls stutter while someone else uploads a large file. Jitter, the variation in latency from one packet to the next, matters most for calls and games.
Internet Speed Checker measures latency both idle and while the line is busy, alongside download speed, upload speed and jitter.