[arin-ppml] Abandonment of 103/104

michael.dillon at bt.com michael.dillon at bt.com
Thu Dec 24 07:14:31 EST 2009

> Addressing is routing is addressing. How many cities does a 
> typical ISP span? With addresses aggregated by city, each 
> ISP+city pair would have to be announced as a separate 
> prefix,

No. ISPs don't HAVE TO do anything. And most importantly,
it is not ARIN's job to tell ISPs how to implement routing
for city aggregates. It is up to ISPs and their customers
to work out what needs to be done which may very well include
special regional peering that allows for cold potato routing
and it may well involve different cost-sharing agreements.
But it is not our job to work out those details. 

In fact, it is better to leave this stuff unsaid in order
to encourage ISPs to innovate and try out different 
solutions in different regions.

> just as the multinational ISPs have to announce their 
> regional RIR-allocated prefixes separately today. And as 
> discussed in the RRG thread, the prefix announcements don't 
> successfully aggregate across ISPs in the same city.

They do successfully aggregate if the one single city aggregate
is the only route announced to the DFZ, and any longer prefixes
are only announced within the city and the nearby region.
ISP's in Europe may need to carry more than just the CA prefix
for Paris, to avoid trombone routes but North American ISPs
can get away with one routing table entry for Paris.

> There are few enough multinationals that the split by RIR is 
> not a serious problem though even there you'd get slightly 
> better aggregation by having each org declare a home RIR and 
> get its global addresses from only that one. Start divvying 
> up by much smaller geographies like cities and you could see 
> a routing table explosion.

But that routing table explosion would be less than putting
every company's PI block in the DFZ and it would also
be regionalised. And it would be optional. If a global ISP
does not want to deal with the complexity, they can just send
all their Paris traffic to the LINX in London, even if the
traffic comes from Strasbourg or Berlin or Warsaw.

If that is the way that it has to be for users of CA blocks,
then when they CHOOSE to use a CA block, they will have
evaluated the tradeoffs and decided that higher latency
is acceptable.

> Every conceivable form of geographic addressing has been 
> exhaustively explored in the research community since the 
> early ideas like CA. Vast numbers of man-hours were spent 
> chasing it in the hopes it would prove to be a Holy Grail. 
> It's a blind alley. There's nothing there to be found.

That's because there is no Holy Grail. In addition, a lot
of that effort was directed to IPv4. The CA addressing
solution is only for IPv6, where the vastness of the address
space allows us to run a decade or two of experimentation
with very low risk of hitting any kind of IPv6 exhaustion.
And CA addressing is not meant to replace any current 
practice of routing, but to supplement it. 

--Michael Dillon

More information about the ARIN-PPML mailing list