RE: Source address selection insufficient?

[Erik Nordmark <Erik.Nordmark@sun.com>]

> So, i guess that at this point, it seems to me that the more reasonable
> thing to do would be to honor the host source address choice, and adapt
> routing to it

That seems to be the robust approach.

> The draft also considers another option that is called site exit discovery.
> 
> In this case, the routers inform the hosts about the appropriate site exit
> through which the host has to send the packet so that it can reach the
> selected destination with the selected source address.
> The host then tunnels the packet through the proper exit router.
> 
> This would be more like source routing rather than source address based
> routing, and this solution honors the source address selection performed by
> the host (it shouldn't have the problems of the other case)

I think this is just a particular implementation of source-based routing -
implemented using an overlay between the hosts and the border routers.

If we are going down the source-based routing architecture path, I think
it is desirable to do this the right way by starting with tunnels between
border routers and then increasing the domain of routers that 
do source-based routing following this picture in the draft:

                 Multiple site exits
                 |     |     |     |
            -----+-----+-----+-----+-----
           (                             )
           ( Source based routing domain )
           (                             )
            ----+----+----+----+----+----
           (                             )
           (   Generic routing domain    )
           (                             )
            -----------------------------

The host-to-border overlay is undesirable because you have to change all
the hosts, and once you've put this part of routing in the hosts it will
be hard to evolve the routing and the adminstrator might loose policy
control as a result.

But that is mostly my gut feel.

  Erik