Re: [v6ops] draft-ietf-v6ops-ula-usage-recommendations - work or abandon?

[Mikael Abrahamsson <[e-mail]>]

On Fri, 13 Nov 2015, George Michaelson wrote:

> Strong +1. We need to stop using FUD on future routing cost to limit how we
> design systems for enterprises and homes. By all means lets constrain
> current behaviour to scale with memory and CPU, but to set limits 5-10-15
> years out based on fear of a situation which is unlikely to emerge, and
> will emerge into different technology and models.. Well its sad.
>
> Its Dionysius Lardner "60mph will kill humans, stop speeding up steam
> trains" thinking.
>
> Whats even weirder, is that its some people who helped displace prior art,
> telephone number centric thinking who are doing it. How come the former
> rebels have become the empire?

I have been involved in Internet core routing for 15 years. I've seen the 
routing table grow from around 100k entries, to 650k or so. Yet, as far as 
I can tell, state of the art platforms still can't install routing updates 
faster than around 10k prefixes per second, which is not significantly 
faster than 15 years ago.

What has changed is a lot of advanced mechanisms to work around this 
problem, plus faster CPUs so changes propagate faster so we have less 
flapping of routes.

No significant improvement in the routing system has happened during this 
time, and I don't see any in the works. Doing route lookups is costly, 
larger tables mean more power consumption etc.

I'm sure we can design routers that can handle 10x the prefixes they can 
today, but at what cost?

Also, why should a core router in Europe have to care about an office 
connection being affected by something on the other side of the globe? IP 
has become successful because the core doesn't have to keep state for 
L4 connections, but if we radically increase the state needed to be kept 
for L3 destinations, we will most likely run into problems. Yes, they 
might be solved, but currently we do not know how.

So I would very much like to stay away from an architecture that suggests 
"PI for everybody".

-- 
Mikael Abrahamsson    email: [e-mail]