In a previous message, Alan Hannan wrote:
[ Dave wrote: ]
> I need to convince someone that singly-homed customer route flaps/
> withdrawals should *not* propagate beyond our AS. I've found some
> discussion of this in the July NANOG archives, and talk about cisco
> floating statics, etc... and that "one-way" traffic is insignificant.
Uhm, I'm not sure the concensus was that they shouldn't. I
believe most everyone would agree that flaps w/in CIDR blocks
should not propogate, and that people should only announce the
most general network possible.
But, if you've got a customer singly-homed to me, ideally, from an
architecturely scalable point of view, you would do well to static
them to your aggregation/POP router.
However, I'm not sure a quorum agreed that single-homed customers
should show up in backbone tables if their routes are/were down.
There are points to be made both ways, but the BB routing tables
are meant to be a snapshot of the net, and if a vector points to
provider P, and customer C is not reachable there, I don't really
think P should announce such....
Well, if C is aggregated into one of my blocks, and they go away for
a while, I'm gonna either blackhole it or send an Unreachable. The
same as I would if their non-aggregated route is not withdrawn. CIDR/
aggregation pushes this out to the edges anyway, so I'd think that this
is of far lesser consequence than thrashing the defaultless/core routers.
I suppose the folks who run said routers would know more than I could
at this point....