Re: [GIST] NLI address choice for responding nodes

hi roland, 

i think the main thing is that the responder NLI should not
matter very much. The only requirement is that it should
lead to a functional MA between initiator and responder,
i.e. one that allows messages to be delivered. exactly how
the messages get delivered (e.g. on what interface) doesn't
have any significance, only that the messages arrive. all
the interaction with the routing system takes place on the
Query message (or inside the implementation). more below:

> -----Original Message-----
> From: Roland Bless [mailto:[email protected]] 
> Sent: 25 July 2008 12:12
> To: [email protected]; Hancock, Robert
> Subject: [GIST] NLI address choice for responding nodes
> 
> Hi Robert,
> 
> Max Laier came across the following issue while
> investigating mobility scenarios and actually implementing
> support for them in GIST:
> 
> - while it is described on p.50 (sec. 5.2.2) how the NLI
>   is to be determined for Query or Confirm messages, there is
>   no guidance how to do it for Response messages.
>   But the NLI in responses is important, because it
>   determines where an MA will be destined to in C-mode.
>   "A suitable choice is the interface that is carrying
>   the outbound flow." is not applicable for responding
>   nodes, though the rest of the paragraph seems to be
>   applicable to a responding node, too.

the 'rest of the paragraph' is applicable - in fact, IIRC
it was put in precisely to give guidance about NLI 
selection for the responder. (I think it was Andy's 
torture test packets in Paris, which included things
like sending a v6 NLI in a v4 response and so on...)

we could put the two sentences specific to Q/C messages
in a separate paragraph, and just say there were no
additional requirements for responses.

and further ...

> 
> - if a responding node has several addresses on the same
>   interface configured, it is not really obvious how to
>   choose the best NLI address. It is even in most cases
>   not possible to determine the associated interface
>   address via which the query came, because the interception
>   usually happens on the interface and not on particular
>   i/f addresses. It may be the case that you have non-public
>   and public-addresses configured on the same interface or
>   that you may have to choose between different CoAs. The
>   situation may be easier in case that the GIST node is
>   the flow destination. In this case you may use the MRI
>   information in order to find an address for the Response's
>   NLI and sending the response from it.

i think that all that matters is that the responder is
reachable. to work out how to be reachable, you only need
to know whether an address is reachable from the same network
region that the query came from. my algorithm would be:
- what physical interface did it arrive on
- choose any global address on that interface

> 
> - the only idea that I have is to either use MRI information
>   if the GIST node is the flow destination or to find out and
>   use the outgoing interface where packets are routed to the NLI
>   address of the Querier (i.e., to use the reverse route).

i think that's one reasonable algorithm. but not that if you 
have overlapping address spaces (e.g. you are a NAT) you still
have to be interface aware.

>   In the latter case, the MA will probably arrive at a
>   different interface than the incoming flow interface
>   at which the query arrived.

true, but (see above) I don't think that matters.

cheers,

r.

> 
> Any other suggestions?
> 
> Regards,
>  Roland
>