RE: The state of IPv6 multihoming development

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

[Can everyone please clean up those CC lists once in a while? I get two
copies of each message which really doesn't bring me twice the joy.]

On Sun, 27 Oct 2002, Michel Py wrote:

> That being said, Craig has some legitimate traffic shaping concerns.
> Even though the host might not make direct routing decisions, by
> choosing either which of its own addresses it's going to use or which of
> the destination's addresses it's going to use, the host itself takes a
> traffic shaping decision, which is not something that Craig wants and I
> support him on that.

There are two cases:

1. The multihoming functionality it implemented in a separate box. In
   this case, you have all the traffic engineering knobs you'll ever
   need in this box. (Which will provide much better traffic engineering
   functionality than current BGP mechanisms, although at the expense
   of more work for thix box.)

2. The multihoming functionality is implemented in individual hosts.
   Without additional mechanisms, you have to revert to rather crude
   methods to do traffic engineering: simply drop enough packets to make
   the host switch to another IP address. Alternatively, we can come up
   with a mechanism for routers to inform hosts which exit they should
   use. As I explained in an earlier message, if the routing decisions
   are also based on the source address in the packets, it should be
   possible to control both the outgoing and incoming path for each
   flow. That's pretty much the holy grail of traffic engineering.

> In short: a large organization needs to be able to engineer traffic,
> which I don't realistically see happening when individual hosts can take
> decisions that can affect traffic engineering.

The host controls the traffic, but that doesn't mean the host gets to
decide. The same way the driver gets to control the car, but he had
better not drive the wrong way in a one way street.