v6 numbering confusion

[Joe Abley]

Me again :)

On Tue, Feb 23, 1999 at 11:13:45PM +1300, Joe Abley wrote:
> 
> One issue that confuses me slightly is that of the aggregation of addresses.
> If one pTLA delegates a 48-bit prefix to us, should we then announce that
> prefix to any successive pTLAs who allow us to build a tunnel to them?
> 
> This sounds a bit like the hole-punching that abounds with current IPv4
> numbering and aggregation. Isn't this a bad thing?

OK. A couple of people got back to me on this, but the answers seem strange...
and not entirely in-line with what I am finding when setting up some
v6 over v4 tunnels around the place.

The advice I got was that if pTLA (a) delegates prefix P(a) to me, and
pTLA (b) delegates prefix P(b) to me, then I should not advertise P(a)
to pTLA (b) [and vice versa], since it breaks the aggregation model.

That makes sense. Kind of. However, there are a few complications that
I can see:

1. If I am assigned n prefixes from n pTLAs to whom I connect, then I
end up with n different unicast addresses for every v6-capable host in
my network. If one of the pTLA tunnels is down, then the corresponding
host addresses (taken from within the corresponding prefix) are unreachable.
If I am using round-robin DNS to spread load across different host
addresses, then in this failure mode my hosts are unreachable 1/n of the
time, and any TCP sessions using the affected unicast address will die
following the failure.  Hmm.

2. If I delegate addresses to our customers (we have customers who want
to join via our v6 network), it seems as though I need to delegate n
prefixes to each customer, and they need to number each of _their_ hosts
n times as well.

Didn't the renumbering problem just get n times worse?

On a completely different topic, I have noticed that some 6bone
experimenters are happy to number the point-to-point tunnel networks
between their routers and mine using quite short-prefix subnets (e.g.
64 bits), whilst others are super-frugal and use prefixes as long as
127 bits.

Is there a convention developing about numbering point-to-point circuits?


Joe