In TCP/IP routing, a host will know how to reach other hosts on the network to which it’s attached, because they’re all on the same network sharing the same network address pool. Beyond that, it needs to have specific routing instructions as to how to get to every other IP address. To get to every IP address in the world, you’d have to have direct connections to all of the networks in the world with unique IP subnets. Then, no matter what IP address you were presented with, you’d always know where to send the packets (that is, via what specific exit interface on which specific router in your network).
But that’s impossible. And the smaller you are, the more impossible it is. If you look at the routing on your local computer (Windows: open command window, type “route print”; OS X/Linux: in terminal type “netstat -nr”), you’ll see routing to the LAN you’re sitting on (likely one of the designated private networks – 10.0.0.0/8, 172.16.0.0/12, or 192.168.0.0/16), and then you’ll see a route for Default. That’s a specific device on your network that – in theory – knows how to get to everything else.
So, how does it know how to get to everything else? Well, it learns that from whatever it’s connected to. That usually comes when you buy service from someone bigger. They give you routing that says “point default at this location”, and take it from there. You could be going to a location on their network, or you could be using their network to get to the network that has the desired location. That’s all opaque to you. You just know that you pay them, they deliver you traffic.
That’s what we mean when we say “transit” – you’re buying the service of some ISP sending your packets across their network (transiting their network) to the real ultimate destination.
In theory, it’s possible to build a network that doesn’t “point default” at anyone else. If your network is big enough, and you can arrange a peering agreement with enough other networks that are also big enough, you discover that for every legal IP address, you actually know the route to it.
This is important – it’s part of the definition of Tier 1 network.
A Tier 1 network is one which doesn’t point default at ANYONE. It knows how to get to everyone. That may be because they themselves have the connection, or it may be because they are connected via peering agreements to other networks which are so connected that in composite, no one has to “point default” at anyone else.
In the late 1990s, there was a land rush to purchase Tier 1 networks. I missed buying the last two – Geonet and NetRail – while I was at Enron Broadband. That boat has sailed, and it’s not really possible for anyone to re-build that these days because of the realization that Dorn Hetzel made at Level 3 in the late 1990s causing a re-thinking of how peering should be handled, and the current state of play being that making money off of transit is critically important for the operation of current backbone networks.
Today, the best that you can arrange is something called “settlement free peering”. That is typically an agreement that says “you send me traffic, I send you traffic, at the end of the month we tally up, and if it’s about equal, no one pays anyone anything”. To get settlement free peering with a major network would require that you have something they valued enough in your network that they would be willing to forego payment for your traffic.
Even the “giants” like Amazon and Facebook buy transit. They have to. It’s just the way it works.
So, Tier 1 networks ARE “The Internet”. If you’re hosted on a Tier 1 network, you know that their route engineering literally can get you to any address that’s actually reachable (note: not all addresses ARE reachable; try getting to something on SIPRNetsometime, from the outside. On second thought, don’t. People routinely do big prison stretches for that sort of thing)