--------------------

Pasi Eronen (2004-08-24):

Based on talks at San Diego, I've created a new issue
on the issue tracker:

   "If both parties have several addresses, do we assume 
  that all pairs have connectivity between them?"

There was discussion about this in mid-April (one example 
mentioned was two SGWs connected via two different private 
networks), and most people thought we should not assume this.

Assuming this would also seem to largely rule out IPv4/v6
mobility.

If we decide _not_ to make this assumption, there are cases 
where we have to change both parties's addresses simultaneously.

It also seems the protocol will require some kind of address
lists, so something like SMOBIKE won't work (unless we assume 
one party knows several addresses for the other party via some 
other means than address lists sent within the protocol).

--------------------

Jari Arkko (2004-08-24):

I think the two SGW example is a pretty convincing one,
so I'd rather not assume full connectivity.

--------------------

Francis Dupont (2004-08-30):

obviously no so not only we need address lists but we need some kind
of loose address pairing.  BTW the question is not new and was
proposed when I tried to find a good usage case for the loose address
pairing my proposal provides...

PS: obviously = just consider a mix of IPv4 and IPv6 addresses for

--------------------

Pasi Eronen (2004-09-03):

Yes, I certainly agree that we should not assume this. I brought this 
up because at some point, Francis was suggesting that each end 
should be responsible for only changing its own address 
(e.g. 2004-07-29), but probably I misunderstood him here.

This question also has an important relationship to issue 16: in 
general, it may be difficult to tell the difference between lack of 
full connectivity and a failure somewhere in the middle of the 
network (and usually there is even no need to tell the difference).

So the answers to these issues are related: if we handle the lack 
of full connectivity properly, we most likely have everything needed 
to recover from problems that can only be detected by end-to-end 
messages (e.g., failure in the "middle of the network"); and vice
versa.


--------------------

Francis Dupont (2004-09-04):

In your previous mail you wrote:

   Yes, I certainly agree that we should not assume this. I brought this 
   up because at some point, Francis was suggesting that each end 
   should be responsible for only changing its own address 
   (e.g. 2004-07-29), but probably I misunderstood him here.
   
=> perhaps a SHOULD is a bit too strong here, IMHO this is only the
common/likely case. There is a known exception: when the addresses
are paired, it is better to change both addresses.
I have two arguments in favor of my opinion:
 - a node has a better control on the condition of its own addresses
 - when an address becomes unusable, the only critical thing is to
   change the local address of *inbound* SAs.

--------------------

Jari Arkko (2004-10-11):

While the discussion in this thread wondered into some
related subjects as well, it seems that everyone supported
not assuming full connectivity between all (operational)
addresses. Can we close this issue?

--------------------

Issue closed on 2004-10-12: "No, we should not assume full
connectivity."

--------------------