Re: Fallacy by Kurt (was Re: IPv6 Policy Clarification - Initial allocation criteria "d)")

[Masataka Ohta <mohta@necom830.hpcl.titech.ac.jp>]

Kurt Erik Lindqvist;

>>Read the draft.

> Several times today alone, otherwise I wouldn't remember what it said.

OK.

> If we concentrate on the  draft-ohta-multihomed-isps-00.txt proposal 
> alone :

That's fine. The architectural draft is not essential for today's
topic.

> If I pick a Tier-2 or Tier-3 provider, I will in most 
> cases get addresses provided out of their PA allocation. 

Seemingly, you think that NLIs peering with a TLI must be
delegated TLI's prefix.

However, ISPs are still free to peer with any other ISPs
without sharing a prefix. Though it increases number of
prefixes within ISPs it is a local policy issue of ISPs
on how much extra prifixes they carry.

The requirement, instead, is that NLIs delegated a TLI's
prefix must peer with the TLI, not vice versa.

Thus, your original statement:

>>This doesn't fly. He can't set his own routing policy and he can't 
>> multihome.

is wrong and the remaining issue is renumbering.

>> If he changes the single upstream his customers needs to 
>> renumber.

That is true on any upstream prefix change, though adding a TLI
as a peer does not necessarily mean adding a prefix.

Then, you miss one thing.

There supposed to be a hard limit on the number of TLAs.

Considering that there will be new TLIs, old TLIs must
retire.

Thus, no TLI is assured to remain to be TLI forever.
I
> I, as a customer, would then much rather be a customer of 
> the TLI directly, that would assure me that I would have stable 
> addresses,

There is no such stability assured.

> That 
> allocation is permanent with the ISP as long as they exist and pays 
> their RIR (and in most cases long after they don't pay their RIR).

That is the assumption which is no longer valid with the introduction
of the hard limit on the number of global routing table size.

ISP hierarchy is inessential.

So is the theory.

Now, consider the reality.

If the hard limit is large enough (I think 8192 is so), TLI changes
will be infrequent.

TLA allocation policy should have mechanism to allow an ISP recover
from accidental loss of TLI status. For example, an ISP may be allowed
to use its TLA one year after its lease period has expired and to
continue to use it (with penalty fee or something like that) if the
ISP recover its TLI status witin a year.

NLIs seldom change their set of prefixes for obviouis commercial
reasons. Even if they do, there will be long notification and
transition periods. End users are free not to add new prefix,
though they sacrifice some robustness. When a prefix from a TLI
is deleted, NLIs with good economical condition will immediately
stop advertising old prefix to end users but keep peering with the
TLI and keep routing packets with old prefix for the time being.
Thus, end users can keep using the old prefix.

On the other hand, ISPs (including TLIs) without good economical
conditios will keep bankrupting and stopping operations without
much prior notices, which has nothing to do with the addressing
architecture.

							Masataka Ohta