Last Resort Registries

I guess that most of the issues here are based on the fact that once you
get address space, you keep it. Looks like it shouldn't be the case
anymore, as was outlined in the Internet Draft on address space ownership.

To put it simply:
- if you have private address space, and wish to connect to the Internet
(or some internet or other), you [might] have to renumber
- if you get address space from provider A, and switch to provider B, you
will have to renumber

So, in any case, you can only use address space provided by your provider,
and will have to renumber in case of change.

In fact, one could imagine it as address space belonging to providers,
exactly the same way it works for the telephone: you can't think of
switching phone operator without changing phone numbers, can you? Or even
move from one place in the country to another without renumbering? [I know
there are some initiatives in favor of an "assigned-for-life" phone number,
but its equivalent in the Internet should be reserved for IPv6, I guess].

At 12:51 21/07/95, Kevin Hoadley wrote:
>So dropping the last resort registries does little or nothing to reduce
>non-provider addresses, since organisations can still acquire address space
>from one provider and connectivity from another. The allocation of addresses
>that may not be aggregateable continues, only now by different registries.

Nope. A provider should not route address space coming from another
provider, or even coming from a LR registry.

>  1/ last-resort registries allocate non-aggregateable address space
>     (which in the future may be useless as no one will route it).

No one will route it. That's *the* point.

>  2/ last-resort registries are less necessary than was the case because
>     of the proliferation of provider registries

And anyway, what use is there to have address space without connectivity?
Yep, avoiding a renumber. But that won't be an option anymore.

>  4/ "end-users will be warned better about what they are getting" if
>     they obtain their addresses from provider registries, rather than
>     last resort registries

They'll be warned in any case: if you get address space from A and
connectivity from B, it won't work, you'll have to renumber into B's
address space, whatever A might B (a LRR or another provider).

>(Have I missed anything ?)
>
>Going through these points, I don't believe dropping last-resort registries
>solves #1, as non-aggregateable address will still be allocated albeit by
>different registries. #2 is true, but not necessarily a reason to drop the

Should be refused to be routed.

>LR registries: less demand for them is not the same as no demand. A similar
>argument can be advanced against #3 - the fact that the charging model may

Why is there demand? There isn't a reason to get address space without
connectivity anymore (well, won't be).

>need to be modified to accommodate the LR registries is not by itself
>sufficient reason to drop the registries.

>I'm not at all convinced by #4. Consider the following scenario: a provider
>registry allocates a block of addresses to a private internet, that claims
>it has no intention of connecting to the global Internet. The provider
>registry levies a charge for this (they win). 18 months later the private
>internet then decides it does wish to connect to the global Internet. However
>now they find that the only provider prepared to carry their addresses is
>the one that allocated them in the first place. Hence unless they are
>prepared to renumber they are locked into the first provider (thus the
>provider wins again). The same would have been true had they acquired the

They have to renumber.

>addresses from a last-resort registry, however unlike the last-resort
>registries it is in the provider's interest *NOT* to tell the customer about
>the possible restrictions in the use of the addresses.
>
>I'm happy to see the last-resort registries disappear (we'd be *extremely*
>happy to get shot of the load from our LR registry :-) ) if it is brings
>significant benefit to the community. However on the basis of the arguments
>I've seen here, I personally think that the benefit is not yet proven.

In fact, the whole point is:
- until now, it was considered better to add a route and not renumber.
- from some point in the [near] future, adding a route will not be an
option. You will have to renumber.

Of course, things cannot be that categorical:
- a period of transition should be permitted (without the need for the
client to keep both connections during that time)
- dual-homed networks are another problem
- and *BIG* networks may be hard to renumber, but such *BIG* networks
usually already have *BIG* CIDR blocks, their own AS, multiple connexions
to the Internet, etc, so I guess they won't need to renumber (though they
might be asked to replace a bunch of nets they got here and there over the
course of time with a big CIDR block).

This might be simplistic, but, if my understand is correct, that is the
point of the Internet Draft on the subject, which was discussed at the
IETF, and seems to be the new way to think about it (quite a radical change
from previous assumptions, I agree).

But, well, I'm a young new provider registry, so maybe over time I'll think
about it differently.....

Jacques.

+-------------------------+------------------------+
|Jacques Caron            | Pressimage Telematique |
|jcaron at pressimage.net    | 5/7 rue Raspail        |
|Tel: +33 (1) 49 88 63 56 | 93108 Montreuil Cedex  |
|Fax: +33 (1) 49 88 63 64 | France                 |
+-------------------------+------------------------+