RE: The state of IPv6 multihoming development

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

On Mon, 28 Oct 2002, Michel Py wrote:

> > First of all, in order to really be multihomed, both addresses
> > must be viable. So if the HTTP farm decides to direct its traffic
> > to the other address, Joe can't really stop them, other than not
> > advertise this address in the first place.

> How could this possibly work? If Joe opens an HTTP connection to the web
> server using a source address of PA2, there is no way the web server can
> send the return traffic to Joe using PA1 as of today.

That's why we're working on better stuff for tomorrow. If the web server
knows Joe's addresses, it can simply direct its packets at any of those
addresses. This is an essential capability; without it the server
wouldn't be able to get traffic flowing again if the address it was
sending the data to stops working.

> So, if it happens that from the web server's network, PA1 is preferred
> over pipe P1, and PA2 is preferred over pipe P2, indeed we have Joe
> choosing over which pipe the traffic will come out of the server farm.

If we forget the fine grained stuff we were talking about earlier, the
web serving network can the same thing we do today with BGP: prefer more
prefixes over the connection that has the most available bandwidth.
The difference with today's practices would be rather small, although in
theory it would be possible for two networks to try to optimize in
different directions so the traffic flow starts to oscillate.