[6bone] 2001:478:: as /48

[Michel Py]

>> John Fraizer wrote:
>> OK.  This brings about the same implications as in IPv4 space.
>> If you want traceroutes through exchange [x] to work without
>> having at least ONE hop that has "* * *" for a return, you will
>> accept the prefix and route it to *TRANSIT CUSTOMERS*.

> Robert Kiessling wrote:
> No, this is wrong. I don't know where this myth comes from.
> All you need to accept packets from that address space, but
> there's no need at all to be able to route *to* it, to avoid
> holes in traceroutes. Thus you don't need to accept any BGP
> routes from that block.

One more time I will agree with Robert here. The traceroute is based on
the TTL expiring. Therefore, all that the network has to know is the
route to the target, specifically meaning that each hop needs to have
NLRI for the announcing router.

So, the source hosts sends a datagram to the target with increasing TTL
values, and each hop that dumps the datagram in the bit bucket sends a
"time expired" message to the source host, and that's the way traceroute
works.

If, by "accident", the TTL expires in a router that the originating host
has no specific route to, it does not matter, as what is required is
that the router that expires the TTL is able to send the "time expired"
to the originating host, not the opposite.

> Pekka Savola wrote:
>.. unless you use stuff like Unicast RPF which the most don't
> (in this context).

That would be easily solved by accepting the aggregate for that space.

Michel.