Re: comments on draft-py-multi6-gapi-00.txt

[Brian E Carpenter <brian@hursley.ibm.com>]

below...

Iljitsch van Beijnum wrote:
> 
> On Wed, 5 Feb 2003, Joe Abley wrote:
> 
> > > Without pre-allocation there can't be any successful aggregation. We
> > > can see this in IPv4 today: the RIRs (pre-) allocate blocks that are too
> > > small to ISPs so these ISPs end up with lots of relatively small
> > > blocks.
> 
> > Organisations move, merge and split apart. Physical and layer-2
> > networks can change radically with no impact on layer 3. There are
> > operators today selling wide-area layer-2 transport services in which
> > single subnets span continents. Layer-3 topologies (both intra-AS and
> > inter-AS) change on a daily basis.
> 
> People keep themselves busy. So?
> 
> > Given such a turbulent soup of connectedness, what criteria for
> > pre-allocation of routing names stands a chance of not being
> > out-of-date as soon as it is published?
> 
> Let's take the speed of light for this. You can change many things about
> your network, but not the fact that making packets take a 10000 km
> detour adds a 50 ms delay. Also, if you're going to physically move your
> network renumbering is only a minor extra task in addition to everything
> else that's going on.
> 
> > [There was no pre-allocation with IPv4, and there is certainly *some*
> > successful aggregation today, despite what you say, and even given the
> > allocation issues you mention.]
> 
> ISPs get /19s, /20s and /16s because it is assumed they'll need more in
> the future than the single /24 they're requesting today. This is what I
> mean by pre-allocation.

Yes, but that is in a CIDR context with *topological* aggregation.
You seem to imagine that geographical allocation will somehow
match up with topology and therefore cause aggregation. All
historical evidence suggests otherwise.

If we could pass a law that every local government in the world
had to set up a monopoly IXP for its area, we might be able to
obtain enough congruence between geography and topology. But I
don't expect that any time soon.

Can't we just accept that only a two-layer solution that separates
identfier-addresses from locator-addresses can solve this puzzle,
and move on to figure out the two-layer solution?

   Brian