[Iccrg] Re: Easter egg in my flow Isolation (non) talk

[Mikael Abrahamsson]

On Sun, 28 Mar 2010, Matt Mathis wrote:

> My talk is on the meeting materials page or 
> http://staff.psc.edu/mathis/papers/FlowIsolation20100323.pdf

I strongly oppose the notion here that ISPs can't build out their core 
networks so it doesn't congest. There is no technical reason why it can't 
be done, and the cost of core network is still fairly small on a 
per-subscriber/megabit basis.

Making core routers cope with flows will make them substantially more 
expensive which will make them costlier to upgrade, which is a negative 
spiral. The core should be considered to be fast and stupid and 
intelligence should be in the access layer. I see no reason to change the 
classic IETF view on this.

What we need is per-flow intelligence on the PE-CE link (basically 
fair-queue), this is what is lacking today.

The core congestion seen today has often other reasons than technical, and 
thus doesn't have a technical solution that makes any real sense.

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