This is an unedited version - Megan 04/20/92

BGP/CIDR BOF
Wednesday, 18 March, 1992, 1:30-3:30pm.
San Diego IETF
Chaired by Yakov Rekhter (IBM Watson) and Peer Ford (LANL)

The BGP working group met jointly with people interested in 
Classless InterDomain  Routing (CIDR) in a BOF to discuss the  
development of an addressing plan which can be used for IP.
CIDR  would allow collapsing adjacent network addresses into a single prefix, 
and that prefix would be passed within the routing system   as the 
route to all the "collapsed" networks.  CIDR is proposed to mitigate
the scaling problems in the Internet's routing system which are due to 
``flat routing'' and the fact that the Internet will shortly  (1-3 years) 
run out of class B addresses.  When the Internet runs out of class B addresses,
the current available option is to allocate class C network addresses which 
will require networks which have more than 255 end systems to  advertise 
multiple network addresses to the global Internet routing system.
The purpose of the BOF was to discuss various schemes for assigning and 
collapsing addresses, including collapsing along a multi-level
hierarchy, what the hierarchies would look like (size and placement),
what the mapping between network providers and collapsed prefixes would
look like.  There was a significant turnout of interested people and
the discussion was quite spirited.

Yakov Rekhter led off the discussion with a brief overview of CIDR and 
an explanation of the goals of the BOF.  He then presented a proposal for 
Address Assignment Authorities (AAAs).  Following are notes from his 
slides:

{beginning of slides}
Goal: "Recommended Guidelines for IP Address Assignment."
	To achieve:
		consistency
		efficiency 
		ease of management and coordination

"Address Assignment Authority" (AAA)

	Distributed way of managing address space
	Promote routing information efficiency 
	Recursive ==> delegation of AAA


Need to extend CIDR

	Pure class "C" supernetting provides _limited_ extension with
		respect to the IP address space

	Large portions of A & B are still unused!

	"AAA"  concept needs to be applied to the _whole_ IP address space.

How  to carve address space?

	Top-Down to ensure feasible routing (wrt scaling)
	How many levels
	Branching Factor at each level
		(deep trees vrs bushy trees)
	Need to determine number of top level AAAs.

CIDR & "NSAP Address Guidelines"

	Attempt to solve the same problem
	May benefit from coordination between NSAP address assignment and
		IP address assignment
	
	Single AAA --> IP & NSAP

	IP and NSAP topology is likely to be congruent
	Address administration boundaries are likely to be congruent:
		a service provider provides both IP and CLNP services
		the same geographical area provides both IP and CLNP services

{end of Slides}


Yakov went on to propose a possible allocation of AAA's, which was to 
assume a top down allocation of 1000 AAAs which would require 
coding top level AAA coding  of 10 bits.  Within the  class c 
address space this would imply that each AAA would have a 
maximum of 1000 class C network addrs.
This  was a good starting point for discussing network topology 
issues, and "who would be candidates for being AAAs?".

There were several people disagreeing with Yakov's proposal for picking 
a fixed size breakout for top level AAAs.  Several people 
proposed  an allocation of top level AAAs which was scaled by the 
size of the community one was trying to serve, perhaps 
using the population size or the size of the telephone networks as
scaling factors.  It was noted that using a Kampei style address
assignment scheme might be a good thing to do here.

There was concern expressed for deploying CIDR too soon, before 
a sufficient technology base was deployed for aggregating multiple 
class C network addresses.    Several people noted that 
this may have severe impacts on intra-domain routing protocols since
an aggregated prefix would have to be exploded to its constituent 
class C networks if the routing protocols did not handle 
aggregation correctly (RIP and EGP).

There was significant discussion of how to carve up class A and class B
network addresses effectively.  There was general concurrence that
for the time being class A's should not be allocated.  This would be until 
there is a technology base which can be used with carved up A's.  It was 
noted that this would be feasible once  most routers "knew  how to 
do variable length subnets".

There was discussion on how Class C# (Solensky and Kastenholz) could 
coexist with CIDR.

Jon Postel gave a short description of what the IANA does and how it 
decides who gets what network addresses.

The discussion clearly overran the time allotment and future 
discussion of this issue was proposed to continue on Email 
using the big-internet@munnari.oz.au mailing list.  Yakov Rekhter
agreed to discuss with the IESG about forming a working group to 
work on an IP addressing plan.