[Iccrg] Is ECN too complicated?

[John Leslie]

Mikael Abrahamsson <swmike at swm.pp.se> wrote:
> 
> The major core routing platforms today do not as far as I know support ECN 
> at all, and these are the ones with big buffers (and support gazillions of 
> queues etc). I asked for ECN support and received pushback for reason 
> being "I was the only one who asked for this" (basically).

   We do need them to leave ECN marking alone. (Has someone documented
which core router do and don't?)

> So as I stated in the meeting, the ISP community is going more in the 
> direction of "let's make a 100GE router/L3-switch with minimal buffers 
> that never congests and upgrade to" than "let's find this platform that 
> can squeeze the max out of this 10GE we already have".

   And that is correct... but when the actual usage of those routers
produces even one packet drop per thousand, we're leaving the comfort
zone. :^(

> DWDM-capacity is cheap, doing buffering/intelligent things in routers is 
> expensive.

   Agreed. But there are some Network Service Providers too cheap to do
either. Can we document which these are?

> I can go more into detail about this if someone feels it's appropriate.

   I don't feel the need for more detail of why intelligence in core
routers is expensive. What we do need is an upper bound on packet drop
(without ECN marking coming first) in actual core routers.

--
John Leslie <john at jlc.net>