01
16 min read · 5 briefings

How the Internet Really Works

No cloud, no magic — just packets, glass, and a planet-sized game of pass-the-parcel.

01 Packets, not pipes

The single most important idea in networking is that the internet does not send your data as a continuous stream down a dedicated wire. It chops everything — a photo, a video call, this sentence — into small chunks called packets, and sends each one independently.

A packet is a self-contained envelope. It carries a payload (a slice of your data) wrapped in headers that say, among other things, where it came from, where it's going, and how the pieces fit back together. A typical packet on the public internet maxes out around 1,500 bytes (the Ethernet MTU, or maximum transmission unit), so a single web page is often hundreds or thousands of packets.

This design is called packet switching, and it was a radical break from the telephone network's circuit switching, where a physical path was reserved end to end for the duration of a call. Packet switching means no single link is dedicated to you — thousands of conversations share the same fiber, each packet slotting in wherever there's room.

Insight Because each packet is routed independently, two packets from the same download can take entirely different physical paths across the world and arrive out of order. Reassembling them correctly is a job for a higher layer (TCP), not the network itself.

02 Routers and the road trip of a packet

Once a packet leaves your device, its whole life is a series of hops between routers — specialized computers whose only job is to read a packet's destination IP address and forward it one step closer. No single router knows the full path. Each just makes a local decision: "for that destination, the next best hop is out this interface."

Your packet's journey usually goes: your device to your home router, out to your ISP (internet service provider), up through the ISP's regional and core routers, across one or more internet exchange points (IXPs) where networks physically interconnect, and finally down into the destination network. You can watch this yourself: traceroute (or tracert on Windows) prints every router that handles your packet, hop by hop.

Each hop adds a little delay, and each router has finite capacity. When a router's queues overflow, it simply drops packets — there's no apology, no retry at that level. Detecting the loss and resending is, again, the job of TCP up at the endpoints.

Pro tip If a traceroute shows a hop timing out with asterisks but later hops respond, that hop is usually just declining to answer ICMP — not necessarily broken. Read the whole path, not one line.

03 Autonomous Systems and BGP

Zoom out far enough and the internet is not one network — it's roughly 75,000 independent networks agreeing to carry each other's traffic. Each is an Autonomous System (AS): a big ISP, a cloud provider, a university, a bank. Every AS has a number (an ASN), and the protocol they use to tell each other which destinations they can reach is the Border Gateway Protocol (BGP).

BGP is a path-vector protocol. Networks announce, in effect, "I can reach this block of IP addresses, and here's the list of AS numbers a packet would traverse to get there." Routers pick the best available path — often the shortest AS-path, shaped heavily by business relationships (who pays whom for transit).

Here's the unsettling part: BGP runs largely on trust. If an AS announces that it owns address space it does not actually own, neighbors may believe it and route the world's traffic to the wrong place. This is BGP hijacking, and it has caused real outages and interceptions.

Watch out BGP has no built-in authentication of who owns what. Defenses like RPKI (Resource Public Key Infrastructure) and route filtering are being deployed, but adoption is still incomplete — the routing core of the internet is more fragile than most people realize.

04 The physical internet: cables and the last mile

The "cloud" is, physically, mostly glass on the ocean floor. Somewhere around 99% of intercontinental internet traffic travels through submarine fiber-optic cables — a few hundred of them, each often no thicker than a garden hose, carrying data as pulses of light. Satellites and newer constellations handle a small (growing) slice, especially for remote areas, but the backbone is undersea.

Data moves through this fiber as light, and light in glass travels at roughly two-thirds the speed of light in a vacuum — about 200,000 km/s. That's fast, but not instant: it sets a hard floor on how quickly a signal can cross an ocean, no matter how much money you spend.

The last mile — the link from the backbone to your actual home or office — is where technology varies wildly: fiber to the home, coaxial cable, DSL over old phone lines, cellular (4G/5G), or fixed wireless. The last mile is usually the slowest and least reliable part of the whole path.

Insight Cables get cut — by ship anchors, earthquakes, fishing trawlers, and occasionally sabotage. The internet survives because of redundancy: traffic reroutes over other cables via BGP, often within seconds, though latency and capacity can suffer.

05 Internet vs. Web, and latency vs. bandwidth

Two distinctions separate people who understand networks from those who just use them.

First: the Internet is not the Web. The Internet is the global network of networks — the packet-switched plumbing. The World Wide Web is just one application that runs on top of it, built from HTTP, HTML, and URLs, invented by Tim Berners-Lee around 1989. Email, video calls, gaming, DNS, and streaming are all separate applications riding the same Internet. Saying "the internet is down" when a website won't load is like blaming the entire road system because one shop is closed.

Second: bandwidth is not latency. Bandwidth is how much data you can move per second (your "100 Mbps" plan). Latency is how long a single packet takes to make the trip, usually measured as round-trip time (RTT) in milliseconds. A satellite link can have enormous bandwidth but terrible latency; a fiber line can feel snappy at modest bandwidth.

PropertyBandwidthLatency
MeasuresCapacity (bits/sec)Delay (milliseconds)
Hurts mostLarge downloads, streamingGaming, video calls, trading
Fixed byBigger pipeShorter distance, fewer hops

Field Glossary

Packet
A self-contained unit of data with headers (source, destination, sequencing) and a payload. The internet moves everything as independently routed packets.
Packet switching
The design in which data is split into packets that share network links and are routed independently, rather than reserving a dedicated circuit end to end.
Router
A device that reads a packet's destination IP address and forwards it one hop closer, making a local best-next-hop decision without knowing the full path.
Autonomous System (AS)
An independently administered network (ISP, cloud, university) identified by an AS number, which exchanges reachability information with other ASes via BGP.
BGP
Border Gateway Protocol — the path-vector protocol that lets autonomous systems advertise which IP ranges they can reach. It relies heavily on trust and is vulnerable to hijacking.
Latency vs. bandwidth
Latency is the delay for one packet's round trip (ms); bandwidth is how much data moves per second (bits/sec). They are independent and fixed by different means.
MTU
Maximum Transmission Unit — the largest packet size a link will carry, typically about 1,500 bytes on Ethernet. Larger data is fragmented into multiple packets.

Knowledge Check

Field Assessment

0 / 3

01 Why can two packets from the same file download arrive out of order?

02 What is the core weakness that makes BGP hijacking possible?

03 A satellite link advertises very high bandwidth but feels sluggish for video calls. Why?

ESC
↑↓ navigate jack in