RE: Provider Independent addressing format drafts

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

Iljitsch van Beijnum wrote:

> Ok, so it can work. But does it have any real advantages? I don't see
> them. If we want geo addressing, why not apply some brain
> power and come
> up with something that works better.

I have no problem with yielding to a better approach, but so far I
haven't seen one that addresses anything more than the service provider
technical perspective. This is a multi-facetted problem space, and we
need a broad perspective approach.

> The problem with a
> direct translation
> between location and address is that there are locations that
> need very
> few addresses and there are locations that need very many.

This is only a problem if one tries to absolutely optimize the
allocation to the need. As long as those with a large requirement can
get what they need, there is no problem. One of the things that keeps
coming up when I talk to site network managers is the fact that a single
/48 is functionally equivalent to a /8 in IPv4. There are very few
organizations that need more than a single /8 today, so one could
extrapolate that very few organizations will need multiple /48's anytime
soon.

> Also, the
> boundaries will be in impractical places. For instance, half
> of London is
> in the western hemisphere, the other half in the eastern
> hemisphere. So
> should there be a LINX East and a LINX West then?

If one chose to implement it that way, yes that might happen. The point
is both prefixes would be exchanged at the current LINX until growth
caused that to be inadaquate. There is no reason to make this harder
than it needs to be.

> On the other hand: direct mapping between addresses and
> locations could be
> _very_ interesting if we had geo routing. For instance, a satellite or
> UMTS base station could decide which antenna to use to
> transmit a packet
> based on the address. But this doesn't seem to be happening.

Is this a chicken and egg problem? We don't currently have a decent
mapping that would allow a satellite operator to build a service like
this, so is lack of service simply a function of lack of an address
model?

> I still believe is some form of geo addressing, but it has one huge
> disadvantage: you need to be fully interconnected within the
> region.

Only for those who are not expressing an explicit policy to the DFZ (ie:
the ones who really don't need to be known outside the region anyway).
The sites that are looking for explicit policy will show up in the DFZ
no matter what we do.

> With
> current multihoming, two networks could lose all their
> interconnects in
> (for instance) the US and there would still be some level of
> reachability
> through interconnects in other parts of the world. With geo
> addressing,
> this isn't possible.

Why? You appear to be assuming that there is magic around one of these
prefixes which would prevent it from being announced via an alternate
path. I would argue that these prefixes would recover automatically
where the alternate announcements for PA space would require manual
intervention.

> I'm starting to think we should give large address blocks to
> commercial
> organizations who will then negotiate what kind of
> announcements will be
> accepted within that block. This way, someone who wants their
> /24 (v4) or
> /48 (v6) in the DMZ can do this by paying a large sum of
> money to one of
> these address brokers, who will make sure it happens by paying the
> networks to accept them. Obviously, this will not stop the
> routing table
> growth but at least it will introduce the laws of supply and demand so
> those who supply can buy bigger routers with the money from those who
> demand.

To a large degree allocating large blocks is what this proposal is
about. The current rules for PA space need to persist intact if there is
to be any structure. If we don't hold the line on the current rules, we
are simply recreating the IPv4 mess (which people are comfortable with
so they seem to like that level of pain). So this means we need an
alternate space which is provider independent to work with, and it needs
some level of structure to scale globally.

One way to implement your suggestion would be to have the registries act
as the brokers, and have the explicit value item be the AS number rather
than any specific prefix. If any organization wants to have an
independent origin AS in the DFZ money would flow. The number of
prefixes a single AS could be origin for could be self regulated through
financial feedback, or I was suggesting we set a hard limit which would
cap table growth at a small multiple of AS growth. I really don't care,
and maybe it gets implemented as one price for up to N prefixes, then a
significantly higher rate for each one over N.

Tony