[ppml] Policy Proposal 2002-3

Lea Roberts lea.roberts at stanford.edu
Tue Oct 7 02:45:54 EDT 2003

Trevor -

the basic problem is that old model for everyone to get direct assignments
from the InterNIC (or now the RIRs :-) just doesn't scale.

in the early 1990s, it was asserted that IPv4 addresses were being
consumed too rapidly.  Back then addresses were assigned in "classes". If
a site needed more than a /24 (Class C) they would get a /16 (Class B) and
class B assignments were being made at an increasingly rapid rate.  In
response to this, Classless InterDomain Routing (CIDR) was born and
addresses could be assigned more efficiently with variable "prefix"
lengths. (see RFC 1519, September 1993)

then, as Owen has said, there were problems with the explosion of routing
table size approaching the limits of the hardware then running in the
core.  the solution was to encourage hierarchical address assignment, also
known as provider assigned addresses, so that an ISP could aggregate a
number of its sites into one advertisement into the global routing table.
thus CIDR came to the rescure for the routing table problem as well.
it's taken most of these ten years for the paradigm shift to solidify and
I would hate to see a rush back to where individual assignments from an
RIR would be the norm rather than a special case.

note there is another dichotomy.  when ARIN makes address assignments, it
cannot guarantee the routability of the addresses.  there are other
players in the scene, e.g. the network operators.  they can negate the
effect of a prefix length change by refusing to accept the routes.  an
unroutable assignment is not worth much!  :-)

since ARIN was formed, one of the challenges has been to establish
policies that preserve the stability of the internet while allowing it to
continue to expand.  the struggle over micro-assignments is ongoing, many
of us who were around 10 years ago still feel that it is wise to be
cautious in changing the size of assignments.  there are ongoing studies
on the next potential for network instability: the convergence time after
route flaps, which is also related to the routing table size.

so while router hardware has improved, there may need to be protocol and
other software improvements in the backbone before advocating the end of
hierachical address assignments makes sense.  As the "swamp"  of original
address assignments reminds us and as you express as a wise concern, the
assignment of addresses is almost completely a one way street.  once the
boundary is moved, we won't be able to undo the assigments should the next
performance barrier be reached.  thus some of us on the Advisory Council
are suggesting that these changes should be gradual and their effects
monitored carefully before moving the boundary again.

		there's never a dull moment in networking land,  /Lea

On Mon, 6 Oct 2003, Trevor Paquette wrote:

> Folks used to be able to receive direct assignments from ARIN (and other registries) in the past. This ability was revoked for some reason, and it seems that no-one can remember why...
> I'd hate for us to go back down the same path and encounter the same problems as did our predecessors only to later go 'DOH!.. no wonder they revoked this ability'.
> Why was this ability revoked and the current policy put in place?? Can anyone explain that?

More information about the ARIN-PPML mailing list