Identifier/locator recap and hop-by-hop vs. source routing

[Pekka Nikander <pekka.nikander@nomadiclab.com>]

Erik Nordmark wrote:
> But my concern is that host multi-addressing will more or less have 
> the hosts track with source/destination combinations (which
> approximate routing paths at some rough level) work vs. doesn't. To
> get fast failover this essentially turns into doing end-to-end
> "hello" traffic instead of relying on the local hello traffic
> performed by the routing protocols. So I think there are severe
> limitations in making this scale. ...
Erik's comment above let me back to an observation that
I made some time ago, but which I failed to write up
before this.

It looks like that if we go for the locator/identifier
separation and multi-addressing, we go for an architecture
where we combine today's hop-by-hop routing with some limited
source routing.  The routers continue to do hop-by-hop
routing on addresses, just as today.  However, either the
hosts or some middleboxes start doing what is effectively
some kind of limited source routing:  based on the identifier
they select the best destination/source address pair from
a number of possibilities.  In that way they affect the
route that the packet takes, i.e., perform source routing.

To perform high quality source routing there should be
some idea of the topology.  That idea does not need to be
detailed; often simple RTT estimates might be enough.
But something is needed.  I think Christian's slides
contained a fairly good summary of this.

I don't know if this observation is of any value.
I just wanted to post it in the case that it might
help people in their thinking.

--Pekka Nikander