Re: SIP Addressing Limitations

[William Allen Simpson <[e-mail]>]

> Date: Sun, 23 May 1993 23:32:53 -0700
> From: Tony Li <[e-mail]>
>    > It amuses me greatly that there's anyone who still seriously
>    > believes that network topology, and geography, are even remotely
>    > related when it comes to international connections.
>
>    Apparently I am in good company.  With Rehkter, Li, and the CIDR luminaries.
>
> Excuse me, but you have misread all of CIDR if you believe that to be
> true.
>
I was discombobulated and enervated to see this line from Tony.  The
plan is largely based on his previous writings.  Apparently, having
convinced me, he has repudiated his own previous thoughts.

(Apologies to Rekhter for the typographical error.)


In CIDR:

9.  Recommendations

   The NIC should begin to hand out large blocks of class C addresses to
   network service providers.  Each block must fall on bit boundaries
   and should be large enough to serve the provider for two years.
   Further, the NIC should distribute very large blocks to continental
   and national network service organizations to allow additional levels
   of aggregation to take place at the major backbone networks.

Clearly says "continental and national".  Of course, it advocates
provider-based within that.  My plan supports provider-based allocation,
in addition to others.


In TAP (p 2):

   As the network evolves, the topology, and necessarily the addressing,
   will also evolve.  Thus, the addressing plan must also allow the
   flexibility to evolve.  Further, political considerations also affect
   the addressing plan.  For example, a single administration may
   control a certain aggregate and may wish to delegate sub-aggregates
   according to some particular scheme.  Another administration
   performing the identical function may wish to use an entirely
   different scheme.  The addressing plan must be able to support both.

Thus my plan's requirement to support metropolitan, provider, and
end-point assignment simultaneously, and follow political boundaries.


In TAP (p 2):

   Aggregation in the addressing plan must be efficient and result in a
   small number of routing table entries within the hosts and routers on
   the network.  ...  Thus, an addressing plan
   should result in at most hundreds of aggregates in the common case,
   for all levels in the hierarchy combined.

The basic division into countries meets the "hundreds" criteria, as well
as the political criteria above.


In TAP (p 4):

   Assigning the top level aggregates to large land masses aids the
   aggregation process in two ways.  Since the top level aggregates are
   large, they describe significant portions of the address space.
   Entities which are topologically distant from a top level aggregate
   can perform aggregation to reduce routing table sizes as the distance
   from the top level increases.  Performing aggregation where possible
   is very effective at localizing routing information to parts of the
   topology where it is most relevant.  Since the natural divisions
   between large land masses constrain topology, they also provide
   natural locations to perform such aggregation.  Experience with CIDR
   already has shown that continental aggregation is effective in
   containing routing table entropy [6,7,8].

Here is some of the argument for continental and regional aggregation.
Note the reference to experience with CIDR.


In TAP (p 5):

   As with other routing schemes that allow advertisement of aggregates,
   this scheme allows a single hierarchical level to be disconnected and
   to still retain connectivity by passing lower level routing
   information through the higher levels in the hierarchy.  This
   introduces "noise" into the routing system and should be avoided
   where possible, but is sometimes unavoidable.  The ability to
   advertise routes which are not aggregated is also sometimes useful
   when the topology

As you can see, the draft peters out, so I'm not entirely sure where it
is leading (probably an editting error when he submitted the draft).

Note that my plan clearly allows for "noise".  And has an explicit
criterion that it continue to operate effectively when the noise is up
to 10 times the base!  As well as an intermediate "cluster" level
between continents and countries to reduce noise at a regional level.


I put in a lot of time reworking Steve Deering's straight list of
countries (the original SIP plan) to meet Tony's ideas on continents,
political contraints, and noise reduction.  In designing those clusters,
I researched current undersea cables, tropospheric scattering for
telephone communications, regional participation in satellite
television, etc.  If you want to quibble about specifics, go ahead --
and cite your data.

In the meantime, I'm pretty disgusted with a group that flames without
reading the documents, and is revisionist about its own previous writing.

[e-mail]