RE: survivability, rewriting

["marcelo bagnulo" <mbagnulo@ing.uc3m.es>]

> > I guess that if a host has the M6 mechanism working, it would
> be reasonable
> > to assume that the site supports M6 and so that exit routers
> would rewrite
> > the source address.
>
> How so?
> If we decide to standardize something which goes in the stack
> then it will hopefully appear in future versions of various stack
> implementations asynchonously and hosts with those new versions
> will appear in
> an uncoordinated way in various parts of the network.
> Thus you can't assume that somehow the site network admin is
> in charge of what OS/stack version is installed on all the hosts
> in their site.

No, but the mechanism doesn't have to be on by default, i guess.
I mean, the same problem would happen if you place host in a single homed
site and that you have configured multiple IPs for whatever reason.
In this case you don't want that the host starts using M6 mechanism, just
becuase it has multiple IPs configured.
So M6 is off by default (to ensure compatibility with single homed sites)
and it is turned on by the admin, who should do it if the site is multihomed
and routers are capable of rewriting packets.

>
> > But anyway you could also use a bit in the router advertisement to
> > communicate it to the hosts, right?
>
> Yes, but you also need a way to get this information into all the routers
> in the site. If the router renumbering RFC was widely implemented and used
> it would provide a mechanism on which to add such additional information,
> but that protocol doesn't seem to be implemented much.
>
> My point is that if there are mechanisms which do not require such
> coordination between the site border routers and the hosts in the site
> they are preferable (all other things being equal) since this coordination
> is additional complexity i.e. additional places where things can go wrong.
>
Agree...
So you would need a mechanism to discover that the host is in a multihomed
site, right?

> > I don't see why the second part is conflicting with Iljistch goal...
>
> It doesn't, which is why I called it "an observation" in the
> clarification.
>

Ok

Regards, marcelo

>   Erik
>
>