BGP & Internet Routing — Junior¶

The internet is not one big network. It is thousands of separate networks — Google, your ISP, a university, a bank — each run by different people, stitched together at the edges. When you load a page, your packets somehow cross several of these networks to reach a server that may be on the other side of the planet. Nobody sits in the middle holding a complete map. This page explains how packets still find their way, and the protocol that makes it happen: BGP.

Contents¶

1. The internet is a network of networks
2. Autonomous Systems and AS numbers
3. Why we need a routing protocol between networks
4. What BGP actually says: "reach these prefixes through me"
5. Routes, paths, and choosing among them
6. Following a packet across the internet
7. Inside vs between networks: IGP vs BGP
8. Anycast: one IP, many locations
9. Summary

1. The internet is a network of networks¶

The word internet literally means "between networks." It is a collection of independently owned networks that agree to carry each other's traffic. Your home ISP is one network. The company hosting a website is on another. Between them may be several large "backbone" networks that carry traffic across cities and continents.

Each of these networks is managed separately: different owners, different budgets, different equipment, different rules about who they will and will not carry traffic for. No single organization controls the whole thing. That independence is the internet's greatest strength — and the reason routing across it is hard.

A packet leaving your laptop for a distant server usually crosses several of these networks in sequence. The problem: how does each network know which neighbor to hand the packet to, so it keeps moving toward its destination rather than in circles?

2. Autonomous Systems and AS numbers¶

Each independently managed network is called an Autonomous System (AS). "Autonomous" means it makes its own routing decisions; it is not told what to do by a central authority.

Every AS on the public internet is given a unique Autonomous System Number (ASN), handed out by regional registries. Think of an ASN as the network's identity card. Some well-known examples:

• Your ISP is an AS with its own number.
• A big cloud or content provider is an AS with its own number.
• A national research network is an AS with its own number.

An AS is also responsible for a set of IP prefixes — blocks of IP addresses it owns and can be reached at. A prefix is written like 203.0.113.0/24, which is shorthand for "this block of 256 addresses." When an AS says "I own this prefix," it is claiming: traffic for any address in this block should end up here.

So the map of the internet, at the level BGP cares about, is not millions of individual machines — it is thousands of ASes, each announcing which prefixes live behind it.

3. Why we need a routing protocol between networks¶

Nobody has the whole map. There is no master file listing every network and how to reach it. So the ASes have to tell each other.

Two independent networks that physically connect (via a cable, an exchange point, or a leased link) become neighbors. Neighbors talk to each other and share reachability information: "Here are the prefixes you can reach through me." Each AS collects these announcements from all its neighbors and builds up a picture of which neighbor leads toward which part of the internet.

This information-sharing has to be:

• Automatic — the internet is too big to configure by hand for every destination.
• Continuous — links fail, new networks appear, routes change; the map must update.
• Between independent parties — it must work even though ASes don't trust each other and don't share management.

The protocol that does this job is the Border Gateway Protocol (BGP). It runs at the borders between Autonomous Systems, which is where its name comes from. Because it routes between separate networks, it is called an Exterior Gateway Protocol (EGP). (BGP is defined in RFC 4271.)

4. What BGP actually says: "reach these prefixes through me"¶

Strip away the detail and BGP is one simple idea repeated a billion times:

An AS announces to its neighbors: "You can reach these IP prefixes through me."

That is an advertisement, or a route announcement. When AS 100 tells its neighbor AS 200 "you can reach 203.0.113.0/24 through me," AS 200 now knows: any packet for an address in that block should be handed to AS 100.

AS 200 can then re-announce this to its own neighbors: "you can reach 203.0.113.0/24 through me" — because it can now reach it through AS 100. This is how reachability spreads outward, hop by hop, until every corner of the internet has learned a way to reach that prefix.

Each time an announcement passes through an AS, that AS stamps its own AS number onto the route. So a route doesn't just say which prefix — it also records the list of ASes the announcement travelled through to get here. That list is the path, and it is what lets everyone reach everyone, and lets each AS choose sensibly among options.

5. Routes, paths, and choosing among them¶

A route is the pairing of a destination prefix with a way to reach it. The "way" includes the path: the sequence of ASes to traverse.

For example, a distant AS might learn two different routes to 203.0.113.0/24:

• via path [AS 400 → AS 300 → AS 100] (goes through three networks)
• via path [AS 400 → AS 200 → AS 100] (goes through three networks)

Both eventually arrive at AS 100 (the owner of the prefix). When an AS learns more than one route to the same prefix, it has to pick one to actually use for forwarding.

At the junior level, the intuition is enough: BGP prefers routes that are, roughly, shorter and preferred by the network's own policy. A shorter path (fewer ASes to cross) is a reasonable default, but real networks also choose based on business relationships — for example, preferring a neighbor they don't have to pay. The key idea:

• Multiple routes to the same place can exist at once.
• Each AS selects one best route per prefix to use for forwarding.
• The path list also prevents loops: if an AS sees its own number already in a path, it rejects that route, so packets can't circle forever.

The precise rules BGP uses to rank routes are covered in the Middle tier. For now: learn many, pick one, avoid loops.

6. Following a packet across the internet¶

Let's watch reachability spread, then a packet follow it home. AS 100 owns 203.0.113.0/24 and announces it. The announcement propagates outward. A user sitting in AS 400 wants to reach a server in that prefix.

AS 400 has now learned two routes to the prefix and picks one best route — say, via AS 200. A packet for 203.0.113.5 then hops from network to network:

Notice what each AS does: it doesn't need the full end-to-end map. It only needs to know which neighbor leads toward the destination prefix — the next hop. BGP gives every AS exactly that, one prefix at a time. The packet is forwarded neighbor to neighbor, each step getting it closer, until it lands in the owning network.

7. Inside vs between networks: IGP vs BGP¶

Routing inside a single network is a different job from routing between networks. Inside your ISP, all the routers are owned by the same company, trust each other, and share a complete internal map. That job is done by an Interior Gateway Protocol (IGP) — protocols like OSPF or IS-IS. Between separate companies, none of that trust or shared map exists, so a different kind of protocol is needed: BGP, an Exterior Gateway Protocol (EGP).

The short version: an IGP finds the best path within a network; BGP finds a path across the whole internet by having independent networks advertise prefixes to each other.

8. Anycast: one IP, many locations¶

Normally an IP address lives in one place — this is unicast. But BGP enables a powerful trick: the same IP prefix can be announced by many locations at once. This is called anycast.

If a company runs identical servers in New York, London, and Tokyo, and all three announce the same prefix into BGP, then every network on the internet learns a route to that prefix — and each picks the nearest one. A user in Europe gets routed to London; a user in Asia gets routed to Tokyo. Same IP address, different physical destination, chosen automatically by routing.

Anycast is how large DNS resolvers and content networks put a "copy" of a service physically close to every user, using nothing more than the routing system itself. Compare the two addressing styles:

Because "nearest" is decided by BGP path selection, anycast automatically sends users to a close site and routes around a site that goes down — if London stops announcing, European users simply get pulled to the next-nearest one.

9. Summary¶

• The internet is a network of networks: thousands of independently run Autonomous Systems (ASes), each with a unique AS number and a set of IP prefixes it owns.
• No one has the whole map, so ASes must tell each other how to reach things. That's the job of a routing protocol between networks.
• BGP is that protocol. Its core message: "you can reach these prefixes through me." Announcements spread hop by hop, and each AS stamps its number onto the path.
• An AS can learn many routes to the same prefix and selects one best route, using path length and its own policy; the recorded AS path also prevents loops.
• A packet is forwarded neighbor to neighbor, each AS only needing to know the next hop toward the destination.
• IGP routes inside one network (trusted, shared map); BGP routes between networks (untrusting, no shared map).
• Anycast announces one prefix from many sites, so BGP steers each user to the nearest copy — the foundation of global DNS and CDNs.

Next step: BGP & Internet Routing — Middle