[narten@us.ibm.com: PI addressing in IPv6 advances in ARIN]

[Nick Hilliard]

> If you want to solve scalability problems, work on revising the algorithms
> and routing protocols (i.e. BGP) and forwarding apparatus to be able to
> scale up with billions+ routes in the table.  The scalability problem that
> many people talk hot air about in IPv6 is at the routing table, not at
> people using the internet.  Let them use it; fix the problems where they
> exist, and stop restricting who gets a routing slot.

More importantly, routing table growth in the ipv4 world has trailed 
Moore's observation, so regardless of optimisations in routing table and 
forwarding table lookup algorithms, the problem can be well contained 
for the forseeable future by hardware improvements alone.  This is 
notwithstanding the fact that the ipv6 prefix table will naturally be 
less fragmented than the ipv4 one, given the fact that LIRs will assign 
PA space in preference to PI, and that we won't start out by creating 
problems like the 192/8 swamp from the outset.

No, there is no doubt that PI ipv6 space is a good move.  We've had 9 
years of a lot of people thinking very hard about new ways to deal with 
ipv6 multihoming, and no-one has come up with anything which is 
demonstrably better.  Meanwhile, the lack of multihoming has been cited 
as one of the (many) reasons that people are not deploying the protocol.

In many ways, ipv6 PI space is a stereotype example of the "Worse is 
Better" maxim.   No-one is going to pretend that it's optimal, because 
it isn't.  But it _will_ work and it's a simple solution to a 
potentially complex problem.

As regards the TLA design and all that, this was never going to work 
from the start.  The internet is not strictly geographical or 
heirarchical; it's effectively chaotic at an AS level.  Everyone makes 
their own decisions about how they want their AS to be routed, and in 
this scenario, it is frankly unrealistic to expect completely 
heirarchical addressing to be of any long term significance. 
Fragmentation of the global routing table (whether ipv4 or ipv6) is a 
fact of life and any attempt to pretend otherwise is plain naive.

Apart from anything else, PI ipv6 space will completely undermine shim6, 
which will now die the painful death it deserves.  And I don't see how 
anyone could argue that this is a bad thing.

Nick