[address-policy-wg] Re: getting second IPv6 PA as a LIR

On Tue, 3 May 2011, Daniel Roesen wrote:

> Why do businesses pay for their competitor's business?

I'm going to sum up my reasons for my views in this email and then most 
likely stop writing as this is going nowhere.

I started using the Internet in early 90ties and as soon as I heard about 
PI, I thought it was a great idea and why shouldn't I have one? It's great 
for me, I never have to renumber. Then I started working with routing and 
udnerstood how this global system worked, and all of a sudden I understood 
the global implications of PI. I still think it's a great idea for each 
individual, but bad for the humanity as a whole. It's just like driving a 
hummer to go shoe-shopping is a bad idea in that it's wasteful to use this 
resource.

A DFZ route has to be carried by everybody, meaning as the DFZ increases, 
everybody has to upgrade their platforms, whether they might need it for 
bandwidth reasons or not. It also means people have to buy more expensive 
platforms. The largest 1U L3 switch FIB size I know of is the Extreme 
Summit 480X, it can do almost 512k IPv4 routes. With IPv4 and IPv6 table 
still growing, I'd hesitate to recommend anyone to buy it right now 
because it won't be able to handle DFZ in 3-5 years. This means someone 
who wants to take the DFZ needs a bigger and more expensive platform that 
most likely uses more energy over its lifetime, not to talk about what it 
costs to manufacture it.

In my 10-15 years in core routing, I've seen the table rise from ~100 to 
~350k routes, but the platforms haven't gotten very much faster compared 
to the routing table size. They still take tens of seconds to converge 
when it comes to programming all these routes into FIB, and the algorithms 
we use to calculate the routes are from the 1980ies or earlier. As far as 
I know, there are no magical new much better ways of doing this on the 
horizon, most of the improvement suggestions involves doing overlay 
networks or having end systems be more flexible on their behaviour 
regarding how tightly services are bound to addresses.

So to sum it all up. I don't trust moores law to keep up with routing 
table growth if we let "everybody" get PI, but it seems I am in 
very much minority here with that view.

-- 
Mikael Abrahamsson    email: swmike at swm.pp.se