Geographic v. topological address allocation

Mon Nov 17 15:25:04 UTC 1997

Sean Donelan <SEAN at SDG.DRA.COM> writes:

> What I was trying to postulate, unsuccessfully, there is no such
> thing as an universal, optimal hierarchical addressing scheme.  I
> thought I had chosen examples from the opposite ends of the spectrum.
> I guess I wasn't extreme enough in my examples.  Perhaps I should
> have used the ISBN hierarchy, a combination of language group, country
> and publisher prefix.  I'm going to publish a million books, so I
> should get a 'big' publishers prefix.

You are confusing transport address (describing a location
in a topology) with an object name (a book, a computer, a
process running on a computer).

An object name like an ISBN does not need to be
hierarchical because they do not describe discrete
locations.  Introducing hierarchy improves managability
and efficiency of databasing.  Many hierarchies can be
imposed on object names.

IPv4 addresses and anything like them need to be
hierarchical in sufficiently large networks because they
do describe the location of one or more objects, and
because flat addresses are known not to scale in large
networks.  Hierarchical routing is the only known means of
scaling IP addresses as they exist now, and therefore the
only hierarchy that can be imposed on IP addresses is
strictly topological.

The canonical object-describing database that is the
roughly the analogue of the ISBN database is the DNS.
(I hate analogies I hate analogies I hate analogies please
don't use them).  The DNS is also manifestly hiearchical,
and that hierarchy introduced efficiency compared to the
former flat ARPA namespace.

Note that the DNS works with suffixes rather than
prefixes, which is a cosmetic difference unless one is
interested in doing binary sorts or tree-based searces,
and that the DNS is variable-length, which is not a
cosmetic difference from the ISBN.

If you are a big organization and plan to have lots of
objects you need a sufficient swathe of DNS names to
describe them all.  These in turn should resolve into
LOCATORS which describe where in the Internet topology (as
opposed to the corporate topology or the geographical
topology) the objects can be reached.

For example, one hierarchy of Internet object names is
clock.org.  Although cesium.clock.org and solar.clock.org
are siblings in that hierarchy, they have very different
IP addresses because they are located in different parts
of the Internet topology.   They should NOT have the same
IP prefix as any attempt at that would introduce
unnecessary inefficiency into the routing system.

Remember: at each level of naming there can be different
and completely disjoint hierarchies.  There are
scalability implications in all of them, most notably when
the names used have size limits or are distributed
non-hierarchically (like ethernet addresses or the COM
domain).  However, the important thing is that when a name
is used as a LOCATOR in a topology, in order to be
scalable that name must be related to that topology and in
large networks must lend itself to aggregation in order to
reduce the amount of information needed to have that
locator be used throughout the network.

	Sean.