Report on Class InterDomain Routing (CIDR)

                        Peter S. Ford 
                    MCI Telecommunications
                        pford@mci.net
                       +1 703 715 7150

                         20 July 1995
                          33rd IETF
                      Stockholm, Sweden

=====

Opinions in this talk do not necessarily represent the opinions or
policies of MCI

Acknowledgments:

IETF CIDRD and IDR Working Groups & Mailing Lists

Tony Li
Yakov Rekhter
Phill Gross
Erik-Jan Bos

Los Alamos National Laboratory
U.S. National Science Foundation
MCI

=====

Outline of Talk

History 
CIDR Basics
Aggregator Based Addressing
Renumbering

=====

IP Addressing Principles Over Time

Used to be 1-byte network number

Then we got classes: A, B, C

Then fixed length subnets

Then variable length subnets

Then supernets/CIDR

Then ...

Observation:

Addressing principles are still evolving

=====

Global Internet Addressing Requirements

Support routing of packets
Scalable assignment
Identify end systems

Pre-CIDR
     Inefficient:  3 sizes fit all -- not!
     Not enough class A and B network addresses
     InterDomain routing system used network addresses
          NO InterDomain hierarchy
               Flat routing
          Potential of >2M global routing entries


     Old System was not going to scale

=====

Classful InterDomain Routing (CIDR)

Extends IP subnet model to entire IP address in two ways

     Eliminates fixed class boundaries
          Network addresses become CIDR prefixes with lengths

               Class A 10.0.0.0 => 10.0.0.0/8
               Class B 128.165,0,0 => 128.165/16

     Multiple levels of hierarchy (longest match)

=====

Number of BGP Entries In Amsterdam1.dante.net

Graph

=====

Current Growth In Addresses Allocated Per Year

Graph

=====

Major Benefits from CIDR

CIDR dramatically decreases the size of global routing tables.

CIDR dramatically increases the efficiency of IP address space
utilization.

Both of these dramatically increase the useful life of IPv4.

And we can do even better ... 

=====

Figure

Prefixes in the global routing system could be:
     Providers A and B
     Sites C and D
     Perhaps Site B (dual homed)

Aggregator networks efficiently represent global routing
information.

Site D exemplifies the waste the global routing system currently
endures.

=====

Provider Based Addressing

Bad terminology

Should be renamed to reflect the role of both public and private
networks:

     Aggregator based addressing.

The goal:
     Only aggregators of routing information should inject prefixes
     into the global system.

Need guidelines for:

Routing efficiency

hosts/prefix

=====

Instances of Renumbering

Dynamic host addressing where host gets IP address when attaching
to an IP network

     Common case for dial up hosts DHCP with directory support
     Critical for Plug and Play

Changing addresses
     New site topology
     Changing providers

Current state of auto-configuration and renumbering technology for
IP IS inadequate

Support for renumbering should be a MUST for IP[46]

Can be added to IPv4 to lengthen the utility of IPv4

=====

Renumbering

This needs to be a core technology, since there are many changes
in topology that occur:

     IP subnet re-engineering

     Subscribers moving btw providers.

Topology changes are not flash cuts.  They will be transitions.

     Longest match supports transition

     Hosts should gracefully accommodate multiple simultaneous
     addresses per interface.

     DHCP notion of Address Lifetime can be exploited

=====

A Pragmatic Note on Renumbering

From the Lottor Survey, 4000 to 5000 class C networks have less
than 10 hosts!
     Crunch down global routing
     Conserve IPv4 Address Space.

Renumbering of these networks "by Hand" into aggregator based
addressing is not a killer.

People do not wait for the absolute cure to a disease prior to
applying late 20th century medicine.

Observation:
The Internet needs to figure out how to get pragmatic to address
our global problems.  Governance is a feature of most civilized
societies.

=====

Name Based Internet

Most customers don't care about permanent IP addresses.  They want
simplicity of operation and assurance of global connectivity.

Telephony analogies do not hold water:

     Telephone numbers are a name.
     There is a database lookup to route a call
     Better analogy is a DNS NAME
          Names should be for forever (relatively)
          Like telco numbers they get put on stationary, business
          cards,  etc.

     Let's focus on using IP addresses for routing!

=====

Conclusion

IP's addressing architecture is still evolving.

CIDR is a simple evolutionary step to use hierarchy to maintain a
routable global Internet.

Renumbering is the next evolutionary step.  Tools and technology
are limiting factor. 

(Keep Plug and Play in mind).