Re: PI & 1st allocation policy

[Gert Doering <]

Hi,

On Thu, Jul 10, 2003 at 06:37:28PM +0200, leo vegoda wrote:
> I'd like to re-start discussion of the PI and 1st allocation policy. 
> There is now less than two months before RIPE 46 in September. It would 
> be nice to be closer to conclusions by then.

Yep.  Thanks.

> The current policy is that anyone qualifying for an amount of PA space 
> qualifies for the same amount of space in PI. Routing may not be 
> considered as justification for additional addresses. 

Which is (and I agree with Kurtis here) bad as it encourages lying to
make the block bigger and improve routeability likeliness.

> When this policy 
> was set there were no initial criteria for qualifying for a PA 
> allocation. That meant that as RIPE NCC membership is open to all anyone 
> could obtain a /19 of PA by paying a membership fee - even if they only 
> justified a /25 and wanted a /24 of PI.
> 
> The minimum allocation size was changed to from a /19 to a /20 in August 
> 2000 following research indicting that fewer than them majority of /19 
> allocations were not even half used[1]. Then, in November 2001 the 
> current initial criteria for receiving an initial IPv4 allocation were 
> introduced. These followed a discussion at RIPE 39[2] and on the 
> lir-wg@localhost mailing list[3].

As Nurani pointed out a while ago, this was done mainly to slow down 
the IPv4 address space wastage (due to new LIR immediately getting a
/20, even if they would have been happy forever with a /24).

At that time, I was all for it, but I think it backfired:  people that
are "growing into being large enough" are now requesting multiple PI 
blocks - start with a /24, fill that, request another /24, fill that,
apply for PA, get PA, inject *3* routes into global table, instead of
only one.

These days, I'm far less worried about IPv4 space running out than about 
routing table growth - if every new LIR injects one additional prefix,
that's bad enough, but a policy that encourages multiple non-aggregateable
network blocks for an entity like a "LIR + customers" is *bad*.

(Of course startup LIRs could go for a suballocation first, and then
renumber into their own allocation - but from what I hear, many people
really really distrust their upstreams and do not WANT to do anything
with "foreign" address space).

> Almost two years on we see that we have about two years of PI space left 
> from the "traditional Class C" space. We also have a situation that can 
> make it difficult for new LIRs to obtain an initial IPv4 allocation. 
> This puts some LIRs in a difficult position. They need to balance the 
> benefits of a single, contiguous network against the costs of 
> renumbering from assorted PI assignments for their own and their 
> customers networks.

Exactly.   As customers usually don't see the need for renumbering, 
and it costs money, it's a very slow process (if it happens at all).

> At the same time we can see that only about 20% of LIRs receiving a /20 
> allocation have requested a second allocation. This suggests that a /20 
> is plenty for the majority of organisations.
[..]
> Based on these data, I would like to make a proposal for discussion. 
> Please feel free to knock holes in the proposal. I believe it is 
> important to explore the issue and so want to make sure that suggestions 
> are discussed.
> 
>                         % ----- Proposal ----- %
> 
> 1. Reduce the minimum allocation size from /20 to /21

Fine with me.

> 2. Remove the requirement to show an immediate need for 25% of the 
> allocated
>     address space (a /23 in this case)

Fine with me (see above).

> 3. No longer assign PI (Portable) address space to End Users
> 4. End Users requiring a portable address block could become an LIR and 
> receive
>     a /21 allocation.

I'm a bit worried about that.  

It's a useful approach as it means "end customers that want to be *so* 
independent from everybody else that they absolutely *need* to burden 
their prefix onto the whole world's routers actually have to pay a 
yearly fee to get those addresses".

On the other hand, it could be seen as kind of a "cartel" that denies
enterprises' access to the Internet or whatever (Daniel K. is the
chief political worrier for this ;-) ) which might stir up sleeping
dragons.


(But overall I like this approach.  I would have never proposed something
so radical myself, but I *have* come to the opinion that PI is *bad*).

Ah.  One thing is missing.  What about enterprises that want address
space that doesn't have to be routeable but *needs* to be unique (for
internal/VPN connections, for IXP meshes, and so on)?  Do we want to
enforce "become a LIR" on them?

[..]
> Finally, I'd like to offer some stats on ASN assignments and IPv4 
> allocations and assignments for the first half of 2003. These data might 
> be useful in the discussion:
> 
> ASN assignments: 599
> Allocation:      377
> PI Assignments:  408

Between the lines, I read "780 new routes in the global table, but only
600 are really necessary due to 'different routing policy = own AS'".  

(But one needs more data to be sure that this is really happening).

Gert Doering
        -- NetMaster
-- 
Total number of prefixes smaller than registry allocations:  55442  (55636)

SpaceNet AG                 Mail: netmaster@localhost
Joseph-Dollinger-Bogen 14   Tel : +49-89-32356-0
80807 Muenchen              Fax : +49-89-32356-299