Things to do to make the network better
Paul A Vixie
paul at vix.com
Mon Jan 5 17:31:40 UTC 1998
> That's great if you're a downstream provider with no transit customers.
> However, when you become a transit provider, it becomes much more difficult
> to determine inside vs. outside, since you're more in the middle between
> two "outsides" that pass traffic through you.
i don't agree. i know that the complexity grows out of hand quickly, but you
have to know, at any edge of your network, the set of routes which can come
from neighbors and the set of transit routes which you have to send to those
neighbors. the exception seems to be routes you get from your own transit
providers, which, numbering as they do in the 10's of thousands, are too many
to list. however, even if you had 10,000 transit customers, you would have
to list them in various places to make sure that they were all advertised to
your various bgp xp neighbors.
it is fairly common to refuse to hear your transit customers' prefixes from
external peers. if your customer is multihomed and if you gave permission
for a cutout, you are probably still not going to use external paths to reach
your own customer even when your igp mesh is screwed up. this lowers the
value of multihoming, and exceptions have been made, and this highlights the
value of portable address space for multihomed customers. but the thing is,
while it's easy to tell IOS "do not listen to these routes from outside" where
"these routes" is the same set of routes you _advertise_ to the outside, it is
NOT easy to say "do not accept packets which are from these nets unless they
come from an igp neighbor." this is something which, so far as i know, we
need the router vendors to fix.
bsd/os 3.* has a feature whereby stable routing is assumed and packets whose
source address makes no sense for the ingress interface are just rejected. i
know that this extra routing table lookup would kill performance in IOS, and
i know that there are a lot of "core" routers for which stable routing cannot
be assumed, but even in big networks where are "leafy" routers where you can
be assured that if a packet comes from a source address which is different from
the forwarding vector if it were a destination address, that something bad is
happening. it seems to me that we had this discussion when the SYN flooding
problem was first published on the cover of 2600 magazine, and it seems to me
that there is no IOS knob, even now, called "ip enforce-stable-routing".
More information about the NANOG