RE: Transport level multihoming

[Peter Tattam <peter@jazz-1.trumpet.com.au>]

On Wed, 8 Aug 2001, Tony Hain wrote:

> Peter Tattam wrote:
> 
> > Until I get it published via the I-D editor, here's a URL to it.
> >
> > http://jazz-1.trumpet.com.au/ipv6-draft/preserve-tcp.txt
> 
> Maybe I missed something, I have no problem with the address exchange, but
> how is a host supposed to know which of its addresses the correspondent will
> be able to reply to? If the prefix required for the SYN/ACK drops after
> connection establishment, won't the host need to be snooping the routing
> system to know which of its addresses to use as the source of the next
> packet? If both hosts are snooping, wouldn't they be in a position to try and
> bias the end-to-end path by switching the destination address to one on the
> list provided in the SYN/ACK? 

This is the difficult bit alluded to in the draft.

In the BSD implementation, the address switch is based on the TCP retry
mechanism (I think after so many retries).  It cycles sequentially thorugh the
list supplied.

This is the tricky bit in the whole idea.  Making the first quess (for the SYN
exchange) is the hardest.  Switching to change addresses is not quite as
difficult as the list returned by the peer can be ordered to make the
subsequent guesses more accurate.

The thesis that accompanies the BSD implementation has an analysis of these
aspects.  It is worth reading.

My next step is to suggest ways that nodes can cache important information
regarding the state of forward routing so that it can make the appropriate
decisions.  It need not be tied to the routing system, just learnt as it goes.

> 
> Tony
> 
> 

I'm sorry it looks like hand waving.  This is new for us all.

Peter

--
Peter R. Tattam                            peter@trumpet.com
Managing Director,    Trumpet Software International Pty Ltd
Hobart, Australia,  Ph. +61-3-6245-0220,  Fax +61-3-62450210