[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Bad routes update

Magnus Ahltorp <[email protected]> writes:
> > Like it or not, we will have to accept multiple announcements of
> > "punched-through" address blocks way below the /28 level.
> No, we don't have to accept that. Accepting it will just make people
> think that the multi-homing problem is solved.

It isn't clear that the "multi-homing problem" has any other solution.

Think of it from a systematic viewpoint. Is there actually a way for a 
multi-homed host both to get service from "two parts of the tree" and
yet not announce that into the default free zone? Thinking about it
for several minutes indicates "no".

There are, of course, possible solutions, but they involve radical
changes to the internet architecture. Stuff like having EIDs that do
not contain routing information has been contemplated, but no one has
yet produced a real functioning protocol based on such mechanisms. v6
as currently defined is not a new architecture with routing separated
somehow from endpoint addressing. 

Given the architecture we have right now, how else can one really
imagine doing multi-homing but with added announcements? You can't get
around the fact that people will need the routing data to make a
routing decision in the current architecture.

Unless we come up with a new internet architecture in addition to a
new addressing plan, it is unlikely we'll have a technical fix for

I think the long term solution here is *financial*, not *technical*. A
BGP route announcement imposes a cost on all of the routers in the
default free zone. By finding a way to pass the cost of BGP
announcements back to the announcer, we can assure that only those
networks that "need" an announcement (with need defined by market
forces) will make such announcements. If someone really needs
multi-homing, they then help pay for the cost this imposes on everyone