RE: Alternatives to source address rewriting (was RE: Preserving established communications (was RE: about draft-nordmark-multi6-noid-00)

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

> So you would envision manually installing the rules in each
> internal router
> that says "packets with source address prefix X gravitate towards exit
> router Rx"?

That was my initial idea, yes

Note that this is not a dynamic information, since it only changes when the
site changes ISP or when the ISP renumbers, and both events shouldn't be
very frequent (at least not so frequent to demand a dynamic protocol)

> Or do you envision internal routers infering this from information already
> present in the IGP?

No

> Or do you envision tunnels between the exit routers to get the packets
> to the correct exit based on the source address prefix?
>

Actually this is an alternative to source address based routing, but i find
it not very optimal, since it implies MTU reduction and sub optimal routing

> > I think that the host should be able to decide if the packet
> will be routed
> > according the destiantion address (in which case rewriting is probably
> > needed to deal with ingress filtering) or it will be routed based on the
> > source address (in which case no rewriting occurs)
>
> In the latter case, if the packet arrives at an exit router and
> the ISP matching the source address is known to be down, should the router
> just drop the packet?

Well if a router don't have a route to route a packet, it should discard it
but that case would depend on the behaviour of the host, let's see this
below...

[...]

> But the host will never know better than the routing system because it
> is operating on aggregated information.

Don't understand this...
The routing system is the one who works with aggregated information implying
information loss.
The host deals with no aggregated host information

> What the host might know better than the routing system is that the
> last N packets sent using a given source and destination didn't result
> in packets being returned from the peer, thus something might be
> broken.
>

Well, there are different approaches, here...
For instance when the host receives such an information it could just try
with all the source destiantion address cobination available for that
communication, and see which one returns faster. Such approach would not
only provide detection of available path but also some idea of the faster
path. Clearly it may be expensive
The other option would be just to change both source and destiantion address
and see what happens and if it doesn't work try another combiantion. The
problem here is response time.

> Thus for the host to tell the routing system "please try sending
> this packet
> over a different path than the default one" might be useful.

This may work... ( i mean i don't see any reasons why not :-)

And what would be the benefits of such approach in wrt source address based
routing?
This feature is clearly not available in routers today, it requires a bit to
inform it to the routers and also that it would ony be usefull to deal with
M6 capable communications. We still have to provide some way to provide
ingress filtering compatibility for non M6 capable communications. The
benefit of source address based routing is that it also addresses both
situations (or kind of)

Regrds, marcelo


 But
> that is quite
> different than having the routing system blindly honor the source address
> for routing lookups.
>
>   Erik
>
>