Re: [lir-wg] IPv6 assignments to RIPE itself

[Alexander Gall <]

[Probably opening a can of worms here...]

On Tue, 14 Jan 2003 14:54:50 +0100, Gert Doering gert@localhost said:

> On Tue, Jan 14, 2003 at 02:17:06PM +0100, Stephane Bortzmeyer wrote:
>> > But the IETF in their infinite wisdom decided that this is not what 
>> > people will want, so you're supposed to give them a /48.  Which, for
>> 
>> RFC 3177 explains in great detail the rationale behind this
>> recommendation. I suggest to read it.

> The recommendation is pretty useless as it doesn't define what a "site" 
> (or a "single edge network") is.

The reasoning in that document is heavily based on the statement that
there is enough address space to give a /48 to everybody and their
dogs.  With this assumption (which, I believe, is fairly sound by
itself), you don't need a precise definition of a site (``when in
doubt, assign a /48'').  I think that RFC3177 is self-consistent in
that respect.  But see below.

[snip]

> If we accept the argument that "there is no shortage of /48s" and hand
> out /48s freely to all individual employees of a customer, or to each
> individual student of a university, then this will deplete the LIRs /32
> pretty quickly (a university with 35.000 students might easily use up 
> a /32 on their own).  Which is not what people envisioned when the
> current IPv6 allocation policy was made, obviously.

I agree.  The numerics in section 4 of RFC3177 assume that the top 45
bits in 2000::/3 can be utilized with an H ratio of 0.25 (giving on
the order of 10^11 /48).  IMHO, the problem with the current
allocation policy is that it is a lot more conservative than RFC3177
suggests while still holding on to the /48-for-everybody rule.  The
relatively small LIR allocations create a level of scarcity in the
number of /48's, which is enough to make people believe that giving a
student as much address space as her entire University is just crazy.
However, the whole point of RFC3177 was that this should be completely
irrelevant.

> Also, this practice would be violating RFC 3177:

>       -  Very large subscribers could receive a /47 or slightly shorter
>          prefix, or multiple /48's.

> - which makes it pretty clear that an enterprise should not receive a
> /42 just because their employees want a /48 each.

> Now how are you going to solve that riddle?

I think this contradiction exists mainly within the current allocation
policy.  RFC3177 seems to be flexible (or vague ;-) enough.  With a
policy in which such an enterprise can act as a (sub-)LIR as well as a
site, they could get a large block from which they can assign a /48 to
each employee as well as to their own network.  Thus, their enterprise
network would be as much a site as the homes of the employees, even
though it provides transit to them.

But this is already done today, isn't it?  We (SWITCH) hold a /32 as
LIR but our network (which I consider to be a site) is a /48 out of
that block.

> (A viable solution would be to give every LIR a /20.  But last time I 
> proposed that, people were Not Amused, accused me of wasting address 
> space and flatly refused to accept that as a new policy...)

That seems to be the dilemma: on the one hand, we should be a lot more
liberal if we want to implement the recommendations of RFC3177.  On
the other hand, we are very reluctant to do that as long as we don't
have very strong arguments that such a policy will actually work in
the long run.

Alex
--
 __________ SWITCH - The Swiss Education and Research Network __________
 Alexander Gall,  SWITCH,  Limmatquai 138,  CH-8001 Zurich,  Switzerland
 gall@localhost         Tel: +41 1 268 1522          Fax: +41 1 268 1568