[ppml] 2005-1:Business Need for PI Assignments

[Daniel Roesen]

On Fri, Apr 22, 2005 at 04:11:09PM +0100, Michael.Dillon at radianz.com wrote:
> If a company receives an AS number then they can announce
> any number of prefixes into the global routing table. Why not
> give them a /32 knowing that in 99% of the cases, this
> company will never need an additional allocation and will
> therefore only need to announce one route entry? That does
> conserve a dimension in tension.
> 
> God help us if we start to hand out /44's and then 5 years
> from now someone creates an IPv6 application that results in
> large numbers of /48 assignments, causing all the /44 allocations
> to run out and get a second one. Then, when we see the problem
> and adjust that /44 boundary up to /32, these organizations will
> once again overflow their 2nd /44 and get a 3rd /32 allocation.

We're talking of PI space, not LIR PA space. So a PI owner is never
supposed to assign subnets from his own PI space to customers (I
know there are grey areas).

The only thing that you achieve with handing out /32 PI is that people
don't become LIR anymore, but just buy/get PI and assign /48s to
customers ("screw documentation!").

By limiting PI to /48 (or larger for real documented need) you take
away this incentive to avoid becoming LIR (ARIN member) as customers
of such "PI ISPs" would be able to get only sub-standard sized
blocks.

PI is for multihomed sites, not ISPs assigning /48s down to customers,
and as such the same rules as for end users assignments (which PI
essentially is) should apply.

> I'd like to point out that my IPv4 DSL provider gives me a /32
> for my home DSL connection and they never even asked whether
> or not I had a router or a single PC. If this is justification
> enough to get one 4.3 billionth of the IPv4 address space, then
> why does a company need to go through contortions to justify 
> one 4.3 billionth of the IPv6 address space?

Because a v4 /32 is the absolute minimum sized assignment possible,
and a v6 /32 is about 4.3 billion times of that.

Over-conservation is bad. Throwing around with space like there's no
tomorrow is as well.

Look at techniques like embedded RP in IPv6 multicast for a clue why.
One unnecessarily endangers the option to use such techniques by
wasting address space.

Really. A /48 is 16 bit CIDRing(!) space. That's enough for almost
everyone in the forseeable future. If you can justify more (huge amounts
of subnets, large and diverse internal routing hierarchy crying for
more flexibility in subnetting) you can always get more, IF justified.


Regards,
Daniel

-- 
CLUE-RIPE -- Jabber: dr at cluenet.de -- dr at IRCnet -- PGP: 0xA85C8AA0