Numbering Really Big Networks

[Howard C. Berkowitz]

Perhaps stepping back a bit, and looking at other context, helps. I will
look at two other contexts  in this note, and hope they add perspective.

I like to ask people, "what's the biggest network in the worlds?"  Many
people answer "the Internet."  "The US military" is the most common
alternative answer.

None of these are even close.  It's the international telephone network
(you know, I never looked at it before, but that is a great name for a
conspiracy).  Also important to this discussion is a "sub-conspiracy"
called the North American Numbering Plan.

Look at a telephone number, expanded to include all long-distance and
international prefixes.  It will begin with a national code such as 44 for
the UK.  Subordinate to that are city codes in the UK, and exchange and
line numbers in that city.

North American numbers have a 1 country code, and the familiar area
code/exchange/line structure below that.

Telephone routing (switching) operates at the broad levels of
international/major national (continental)/local.  International carriers
interconnect countries; Interexchange Carriers (IXC) interconnect local
areas, and local providers interconnect exchanges and lines.  There are
different technical requirements at each of these levels, much as there are
different technical requirements for a small ISP that needs to support very
good end user dialup, and a national provider for whom 155 MBPS links (and
certainly routers!) are not fast enough.

Especially due to the advent of new technologies requiring unique telephone
numbers, there has been an extreme demand, especially in large metropolitan
areas, on the telephone number space.  This has led to certain area codes
(e.g., 212, which once covered Manhattan) to fill up.  There may be space
left in Area Code 702 in the deserts of Nevada, but that isn't usable
inside Area Code 212.  Area code 212 has a fixed limit dictated by the
999-9999 phone number structure.

So, there is a practical issue that certain area codes are exhausted.
Telephone  renumbering is a nasty issue for several reasons.

First, if my area code changes, I have to change all my letterheads, etc.

Second, anyone who has stored my number will have to change it, whether or
not their number has changed or not.

Third, there are several technical ways to introduce new area codes.  The
usual way is to split geographic areas, so half the people in them are
affected.  A newer model is "overlay," where additional area codes are
applied to a single geographic area.  The latter model requires 10-digit
dialing for every number inside the local area, but doesn't force people to
change their numbers.  Advantages and disadvantages to both.

How does this relate to ARIN? There has to be a central registry to assign
area codes.  There is a usually-hidden list of provider codes that also
needs to be administered.

When one gets a telephone number in a geographic area, the number
assignment is given to the end user by their serving telephone company.
That telephone company aggregates the subscriber into an exchange, and then
an area code.

If one moves their business to another area code, there is rarely a
presumption that the phone number is portable.  It was allocated to the
aggregating provider.  The telephone switching system has a finite address
space, there is an extremely large capital investment designed to deal with
this, and the system was not designed for portable addressing.

IP wasn't really designed with either portable or nonportable numbers in
mind.  RFC760 started with an 8-bit prefix that soon proved unscalable, and
changes were made.  But  that 8-bit prefix reflected IP was introduced for
small research functions where lots of flexibility was practical, but has
evolved into an environment much more constrained to maintain operational
quality.

An issue that concerns many people on the ARIN list is being "locked in" to
a single upstream provider, assuming the general case is provider-based
allocation.  One concern here is the disincentive for people to go with
small providers, because they may need to renumber as the small provider
grows.

Yes.  This is a true statement. To say it is not -- to say I have provider
independent phone number space -- is saying I get to move from
(703)998-5819 in Virginia to Silicon Valley, and can reasonably expect to
get (408)998-5819, or even keep (703)998-5819 with no call forwarding
expense.

There are two ways to look at this.  One is that renumbering is
anticompetitive and must not happen.  The other says there are technical
reasons to renumber, and probably good ones.  Let's focus on making
renumbering less painful, because it will be a fact of life.  Let's also
realize there are significant real world examples where even large end
users might face renumbering of their Internet interfaces, but would have
to renumber a very small portion of their users.

Making renumbering easier is the focus of the IETF Procedures for
Internet/Enterprise Renumbering (PIER) Working Group.  It has a web page at
http://www.isi.edu/div7/pier/.  Minutes and such are at
http://ds.internic.net/ietf/pier/.

PIER has produced several RFCs, and more are in the works.  There is also a
collection of renumbering case studies, which I _think_ the web page points
to.  With all due humility (yeah, right), two RFCs that might be of use to
ARIN are http://ds.internic.net/rfc/rfc2071.txt and
http://ds.internic.net/rfc/rfc2072.txt.  The first is an overview of the
renumbering problem by Paul Ferguson and myself, the second is an
operational-planning-oriented document I wrote called the Router
Renumbering Guide.