Re: Comments to draft-ietf-mobike-protocol-00.txt

Mohan Parthasarathy writes:
> If the operational pair is not working and the intiator is behind NAT,
> it could be multiple reasons : NAT reboot, some router crashed etc.
> You assumed that NAT rebooted above. If the NAT did not reboot,
> then the PATH did not really change. In this case, the initiator would
> still not see any response and hence try a new PATH later.

The first thing that initiator does when it does not see packets is to
start DPD on the operational address pair. It does this regardless if
it is behind NAT or not. The only change there is that if it is behind
NAT he should put CHANGE_PATH notify there too.

If the problem was NAT booting, then the CHANGE_PATH will update the
responder side (it is processed there as normal IP-address change). If
the lack of traffic was because of some other reason, then the
CHANGE_PATH does not do anything, as it would only change the
operational address pair to be same that as it is already now (i.e. no
operation for the responder side, this situation can happen quite
often in other cases too, so responder must be able to process that
correctly).

If this was because path having NAT broke down, then the DPD will
try other address pairs too when it cannot get packets back with the
current operational address pair. Then initiator will find the proper
working address pair, and finishe the DPD and do the another
CHANGE_PATH to that (as specified in normal case where the packet
having CHANGE_PATH is retransmitted to multiple address pairs).

So the only change really is the addition of CHANGE_PATH in the DPD
packet if behind NAT.

> Does this mean that the initiator behind NAT always have to try
> sending DPD with CHANGE_PATH and then try new PATH ?

Yes, but actually the DPD with CHANGE_PATH will be used to search for
the new path. I.e after it finishes we have new address pair that
works, because we do retransmit that packet to other addresses too.

> If we think this is a corner case, i am not sure whether it is worth
> it.

Nat booting is corner case, but adding the CHANGE_PATH to DPD when
behind NAT is so small change that I think we should do it. All the
other processing is still same than in normal case. 
-- 
[email protected]