Re: [GLOBAL-V6] Comments on AP Consensus

[Thomas Narten <narten@us.ibm.com>]

Carlos Friacas <cfriacas@fccn.pt> writes:
> > - Organizations requesting address space must be an LIR. (see Section
> >    2.6).
> >
> > - Organizations requesting address space must not be end sites.

> These two "rules" imply that Provider Independent networks shall not
> exist ???  IMHO, this is a good thing!

Actually, I would say this differently. First, we don't know how to
scale routing using anything other than aggregatable addressing. Thus,
we need for the RIRs to adopt policies that encourage good aggregation
as much as possible and discourage (to the degree possible) schemes
that have the potential to produce routing table explosion.

Second, *this* global IPv6 policy document does not support portable
addressing per se. That does not preclude other documents/policies
that relax this in carefully thought out situations. But given the
issues surrounding this topic, *this* document should not try and
address this topic. 

> But this last one will not have a great effect... i think it is rather
> easy for a medium/large organization to "forge" end sites...
> Another "workaround" can be made by "ficticious-merging": Little company
> A, and Little company B "together", stating Little company B is the end
> user...

The intention with the current wording is to make it clear what the
intention is and give the RIRs the flexibility to pushback on
organizations that try to obtain allocations that don't match the
intention. We can try to tighten up the wording, but I suspect that
ultimately we will not be able to produce iron-clad text that deals
with all possibilities and we need to assume the RIRs will excercise
some common sense.

If you have wording suggestions, please post them.

> Is in any way dangerous replacing these last two, or adding:
> "- Having an ASN" ?

Does this really help, or will this be easy enough to achieve in
practice that its not a real hurdle?

> > - Organizations requesting address space will provide connectivity for
> >    the organizations it has assigned /48s to by advertising such
> >    connectivity through the single aggregate allocated to that
> >    organization.

> Hmmm... this is the logical thing, seems to me like a compromise, but IMHO
> it is not a "prerequisite" to get an alloc.

It is not a prerequisite per se. It is to make it clear that an LIR
should only be assigning space to orgnanizations for which it provides
connectivity. It shouldn't be giving out address space to
organizations that just want address space, but intend to get their
connectivity through some other means (if at all).

Note: this does not mean that the *only* connectivity is through that
LIR/ISP. An end site can have service with multiple ISPs and arrange
for them to carry the end site's /48. I.e., this scheme does not
preclude multihoming as it is done today.

However, they are not portable addresses. It is important that if some
(or no) other ISPs are willing to carry the /48s (for whatever
reason), connectivity to the site is still possible via the aggregate
route that its LIR/ISP is also advertising. This means that if
filtering is used for longer prefixes (which will certainly be the
case at some point), we don't have a loss in connectivity.

> > - Organizations requesting address space have a plan for assigning
> >    address space (e.g., /48s) to other organizations, with the number
> >    of such assignments likely to result in at least 200 such
> >    assignments over the next two years.

> Having a plan is good, but going on a round number doesnt seem right...
> This 200 number is not right for example (and also unfair) to European and
> small countries NREN's!

My view is that /32s should *only* be given to LIRs for which there is
a reasonable probability that they will actually at some
(undetermined) time in the future actually might completely use the
entire allocation. This could be in 10 years or even 20. This is good
from an aggregation perspective. But since figuring out how to ensure
that all allocations result in full utilization is hard, the 200 over
2 years was (admittedly somewhat arbitrarily) picked as a reasonable
balance between being too strict and not strict enough.

Personally, I think that if an LIR in fact will not ever have many
more assignments than 200, it probably shouldn't be getting a /32 in
the first place. A /32 is quite a lot of space for only 200 sites. But
I think the current proposal achieves a reasonable balance between
competing issues.

> Some small countries NREN's have a lot less than 50 "end-points", so would
> it be acceptable to make a plan assigning a /46 (4x/48) to each end-point?

IMO, no.

> > - Organizations who are granted initial allocations, but after two
> >    years no longer satisfy the requirements above, are subject to
> >    having their allocations revoked.

> Hmmm... this is rather dangerous, but somehow it also reflects my own
> view, in a general way.
> Why? Because i think this is an issue that fails in the v4 World: getting
> free space back to the "free address" status pool.
> LIRs close, LIRs stop paying, and in reality things are not "reversed", or
> in some way at least audited. So my point is: Nice to have the rule, but
> it will serve for something? I have real doubts about it...

If the rule is not there, it's hard to see how we will ever be able to
cleanup "mistakes". I don't think this wording implies a big change in
how the RIRs operate. But it gives them them the authority to act, for
example, if it turns out that organizations are excercising bad intent
in obtaining address space. E.g., after two (or more) years never
having made a serious attempt to actually make 200 end site
assignments. Or, clearly being an end site rather than a true LIR.

This authority has been lacking in IPv4 with legacy allocations, for
example, so it needs to be written into the policy from day one.

> So this "rule" seems very nice on the "recycling" intention, but what is
> really needed are some ways/"rules" to really enforce it. :-)

I think details of enforcement is best left to the RIRs. And I would
expect that they will not act arbitrarily and capriciously before
doing so. I.e., I would expect they will consult with the community as
appropriate before carrying out policies that may have significant
impact on its members.

Thomas
-
- This list (global-v6) is handled by majordomo@lists.apnic.net