In IP, to what extent are the network addresses coordinates?

Question

Routing of data relies on being able to reach the destination address. There are network addressing systems proposed where the addresses are coordinates, Vivaldi from MIT, for example. The routing in these network topologies has the advantage of knowing when a package gets closer to its destination. In the IP system most commonly used as layer 3 in the OSI model, is there any notion of location in how machines are addressed?

Vivaldi: A Decentralized Network Coordinate System, http://pdos.csail.mit.edu/papers/vivaldi:sigcomm/paper.pdf

Netsukuku, http://netsukuku.freaknet.org/doc/main_doc/netsukuku.pdf

PIC: Practical Internet Coordinates for Distance Estimation, http://rowstron.azurewebsites.net/MS/PIC-ICDCS.pdf

Answer 1

That depends very much on what you consider as dimensions.

IP uses a paradigm of network part/prefix + host part that could be regarded as dimensions. The network prefix is used to select the next-hop gateway and on the last hop, the host part is used to select the destination host.

If you visualize all networks on one axis and all hosts in each network on the other axis, that's a kind of coordinate system.

Of course, the division between network address and host part is flexible in IP, as defined by the subnet mask or prefix length - a very clever scheme. The prefix length usually varies along the path, from more general, aggregated routes from afar to more precise, refined routes when closing in towards the host.

For instance, your ISP may advertise the 123.45.0.0/16 route on the Internet. Behind the border, 123.45.64.0/18 indicates the datacenter with your server in it. Inside the datacenter, 123.45.67.0/24 addresses the subsection and 123.45.67.80/28 your allocated address range with your server finally at 123.45.67.89.

Obviously, the IP "coordinate system" is a purely logical one - it as no relation to the actual location. IP routers have to exchange, learn and update their routing tables to be able to route towards any IP address (and not away from it).

The papers you've linked to try to establish a geography-based coordinate system to simplify routing - however, the challenge would be to deploy a rigid scheme that never needs updating, which probably isn't possible in practice. Once people use tunneling (for various reasons) all purely geographical routing fails.

Answer 2

In the IP system ... is there any notion of location in how machines are addressed?

In a word, no. An IP address is not a physical location. In fact, otherwise sequential /32's may be miles apart. For example, my DSL connection has a static /32. One might assume -- from Bellsouth's WHOIS records -- everyone in that netblock are in the same general region, but that's not necessarily the case, as the account for whom that address is assigned can be used to login on any DSL port in the network. (I have personally used my account on devices in 3 states spanning a region of over 1000 miles. And it literally takes seconds to logout in one place, login at another, and *poof* that address is now hundreds of miles from where it just was.)

The same sort of portability exists in data center ("cloud") networks. VXLAN can allow what appears to be a continuous layer-2 network to span buildings, countries, and even continents.

It's up to individual networks to adopt their own "address plan". That plan may, or may not, have some relationship to geography -- it's ultimately a personal preference.

Routing on the internet is done between Autonomous Systems using BGP. BGP's primary path selector is "AS path length" - how many AS will the packet have to traverse. Eg. 1-2-2-3 (4) is longer than 1-2-3 (3) -- 4 vs. 3. 1-4-3 is also 3 hops, so which is chosen in many cases comes down to local preference. (although BGP does have numerous selection criteria, because such "ties" come up often.) This does lead to internet traffic rarely following what might otherwise appear to be the most geographically optimal path.