BALTS: Better At Least Than Singlehomed

[Iljitsch van Beijnum <iljitsch@muada.com>]

Since "the" solution seems more elusive than ever, maybe this is the time
to get a "better than nothing" effort off the ground.

The idea is very simple: when everything is working, send regular packets
and do everything just like in a single homed environment.

As soon as the source host learns it can no longer send information to the
destination host using regular means, it starts encapsulating the packets
into IP packets. It then transmits these packets using a different source
interface, gateway and/or destination address. The exact interface,
gateway and destination address choices are determined by local
configuration or heuristics.

At the destination address, the packets are de-encapsulated and processed
like regularly received packets.

In essence, this means creating on demand dynamic tunnels.

A good way to publish secondary addresses would be in a new resource
record in the ip6.int domain, or as an IP option in the first packet of a
stream.

The main problem would be detecting failures. If source address filtering
is configured correctly (or rather: in the way that suites us best) local
link failures will generate a "administratively blocked" ICMP message,
which can be used to trigger a change of source adddress. Host
unreachables can trigger a change of destination address, but there are
many failure modes that fail to provide these.

Advantages:

* simple
* changes limited to IP layer, no higher layer changes necessary
* a lot of potential for improvement down the road

Iljitsch van Beijnum