Re: issues 18, 15, 6 -- return routability

Hi Stephane,

>I think the RR test should be done before the loss of connectivity.  
>
>Before would quickly accomplish the goal of proving that the peer actually
>owns the IP address.  If you wait until after the loss of connectivity,
>you'd have to source the RR using every interface you had configured, since
>you wouldn't be sure if your interface was actually working properly (or if
>it is the peer's interface that malfunctioned).  This would exponentially
>increase the number of messages per interface.
>  
>
I think you interpreted the "before/after" distinction slightly
differently that what had originally thought. I was asking
whether we do the test before the traffic stream is moved,
you are asking whether to do the test before the nodes
move. Both are valid questions, but I was assuming that
nodes just happen to move, and that we don't necessarily
always know that this happens before its already done.
Of course, a MOBIKE node that knows its address is going
to go bad soon can and should start using a new address.
But I don't think we can guarantee this is known in all cases.

(But if we know we are going to move, we don't need to
RR every address, just the one that we are going to use.
RR isn't need to test reachability, my understanding is
that some kind of probing would take place prior to that.
For instance, when a client moves, it should probably probe
connectivity to the SGW in order to make sure its new address
works and in order to punch a hole to the NAT mappings. Then
it can tell SGW that this is the new address, and the SGW
can perform an RR test to verify that the new address was indeed
what was claimed.)

--Jari