RE: The state of IPv6 multihoming development

["Tony Hain" <alh-ietf@tndh.net>]

Iljitsch van Beijnum wrote:
> > > Actually holes in PA space is perfectly fine for the 
> network, as you 
> > > can filter the more specifics and still have connectivity.
> 
> > This statement is self contradictory.
> 
> How so?

By definition, when there is filtering, connectivity is being
intentionally broken. 

> 
> > > However, it is not so great for the
> > > organization using the addresses since this doesn't address all 
> > > failure modes.
> 
> > The point I was making about punching holes in PA is that the more 
> > influential organizations will get the full prefix routed globally.
> 
> So then the question becomes: how do we limit the growth in 
> organizational influentialness?  :-)

You can't, so it becomes how do we scale to deal with them?

> 
> This is bad for operators as they have to make complex 
> filters. Also, this "solution" still requires renumbering.

This depends on the mechanism that is selected.

> 
> > > This could be done by having several instances of
> > > this address space and have two places that would otherwise be 
> > > aggregated together sit in a different instance. By rotating the 
> > > axes and limiting everything to (say) from -60 to +60 degrees you 
> > > eliminate the pole problem.
> 
> > I played with rotating the axis, but the only thing that 
> made Europe 
> > contiguous was to put the origin on a 22.5 degree mark.
> 
> That would be shifting. What I mean is have at least three 
> instances, one with the poles at 90 north/south, one with the 
> "poles" at the equator at the 0 and 180 meridians and one 
> with the poles at the equator at the 90 and 270 meridians. 
> Then you only need to go upto 45 degrees and still every part 
> of the world would be covered by two instances. This means 
> 1.5 times more address space used, but 2 times as much usable 
> in the non-polar regions.

What is the real gain?? If it is only 2x, the size of the grid shrinks
from 10x10m to 7x7m. 

> 
> > > Still, there is one unanswered question: what problem is 
> solved by 
> > > having such a close relationship between geography and address 
> > > space?
> 
> > I was not trying to force any particular alignment. The goals were: 
> > Uses current BGP protocol. Aggregate the 
> basement-multi-homer, while 
> > recognizing the CNN's of the world will always get a full prefix 
> > announced. Decouple the scale of aggregation, so it becomes 
> a regional 
> > decision. Make it simple to derive by basing it on a globally 
> > consistent reference.
> 
> You fail to address my question. This scheme uses an 
> incredible amount of address space, 

But you were arguing for using 1.5x the space???

> but still manages to come 
> up short in very densely populated areas. 

This is BS... Show me a space where 128 /64's per vertical meter is
insufficient. 

> The aggregation 
> properties are relatively poor. 

Based on what criterion? For the 200+ existing exchanges it would appear
to reduce the scope of the flat routed parts of the table. Also, the bit
interleave allows for varying aggregation scale based on local need.

> So what is it that makes all 
> these downsides worth it?

Show me a plan that works better and is simple enough to really deploy.

> 
> Basing a geographical address allocation scheme on population 
> (x addresses per person) rather than geography (x addresses per square
> kilometer) makes more sense, as long as you're not going to 
> use IP addresses to aim your microwave antenna or something like that.
> 
PA is an appropriate technology for X addr/person. What we are
discussing here is how we deal with multi-homed entities. What that
requires is a defined way to abstract without loss of utility. 

Tony