Re: draft-shirasaki-dualstack-service-02.txt CPE assigned Host Address Lifetime issue

[JINMEI Tatuya / 神明達哉 <jinmei@isl.rdc.toshiba.co.jp>]

(I've not fully been following this thread, but)

>>>>> On Mon, 16 Feb 2004 10:51:31 +0900 (JST), 
>>>>> SHIRASAKI Yasuhiro <yasuhiro@nttv6.jp> said:

>> So now the hosts (PCs) will have an old  non-link-local IPv6 address, with
>> its lifetime still valid for the next week or month, and receives a new
>> non-link-local IPv6 address. When sending out packets, it will have to make
>> a default address selection based on the two source addresses. It will have
>> good chance of selecting the old address, which the BRAS doesn't have a clue
>> when forwarding packet back to this (destination) address.

> I've just checked two implementations and noticed a different behavior
> against prefix changes. When the CPE changed advertising prefixes
> in a flash, one implementation did the things as you mentioned,
> but the another one has revoked an old prefix immidiately and started
> communications with a new prefix only. If all hosts behave as
> the latter implementation, we don't need any special treatment
> with prefix changes...

Please let me check, are you talking about a host that behaves like
this?

- the host receives a prefix P via RA from a router, and configures an
  address P:A.
- then the router stops advertising the prefix P, and starts
  advertising another prefix Q.
- the host receives the new RAs with prefix Q, and configures a new
  address Q:A.  At this stage, the preferred and valid lifetimes of
  the old address does not expire.  However, the host never uses the
  old address as the source address even if RFC3484 suggests the
  old address be preferred.

If so, I'd say the host is not fully compliant (at least not compliant
to RFC3484), and network operators should not assume such a behavior.

					JINMEI, Tatuya
					Communication Platform Lab.
					Corporate R&D Center, Toshiba Corp.
					jinmei@isl.rdc.toshiba.co.jp