Topologically Sensitive 
                   "Names" for a Very
                      Large Network
                            
                           or
                            
                   Why I Like Variable
                    Length Addresses

                            
                      Noel Chiappa

========================================================

1)    Hierarchy is the only way to
      make routing scale in a large
      network.


2)    Hierarchical names are
      sequences of area names.


3)    Ergo, hierarchical names are
      inherently variable length in a
      non-constrained topology and
      address structure.


4)    Aggregating two things which
      are not topologically related is
      of no use to the routing.

========================================================


           A1

         /     \

       B1       B2

              /      \

            C1        C2  <--- (A1, B2, C2)


========================================================

                   A1

             /      |     \

          B1       B2       B3

        /    \      |        |

      C1      C2    C3      C4

========================================================

1)    In a decentralized network,
      there is variation in tree
      depth.


2)    If naming aggregates are done
      for QOS/policy reasons, there
      are more layers.

========================================================

I don't have a good enough
crystal ball to represent a variable
length hierarchical name in a
fixed length field.


Ergo


We have to look for efficient ways
to use variable length hierarchical
names.

Flows are my best bet, but other
solutions are possible.

See Nimrod IPng Requirements
document.