Re: shim6 @ NANOG (forwarded note from John Payne)

[Erik Nordmark <erik.nordmark@sun.com>]

marcelo bagnulo braun wrote:
> > Wouldn't the hosts in X and Y have two IPv6 addresses?
> > Thus BGP itself at site X wouldn't know that B:Y and D:Y is the same 
> > host, hence BGP wouldn't compare them to find the best.
>
> yes, but i think could work anyway
> I mean, suppose that we are performing source address rewriting by the 
> multihomed site exit routers
> Now suppose that intra site routing is performed based solely in the 
> destination address (just as it is in the current paradigm)
> Assume that the multihomed site obtains BGP feed through both of its 
> providers.
> Now, based on the BGP information obtained, the intra site routing will 
> preffer one or other exit to reach a given destiantion.
> So, when a host within the multihomed site sends a packet to that 
> destination the exit route will be determined by this BGP information 
> hence the policy information pushed by the transit ISPs can affect the 
> intra site route selection
> Since the site exit routers can rewrite the source address, they will 
> select the appropriate source prefix

My response to Iljitsch was about his assertion about the current 
workings of shim6 - not taking into account any hypothetical extensions 
like router rewriting about locators.

But even with such a hypothetical extension, there would be nothing in 
BGP that would know that the B:Y and D:Y address prefixes are for the 
same site, hence nothing in BGP that would compare the metrics (AS path 
length or whatever) for the routes to those two separate prefixes.

   Erik