[Ecn-sane] per-flow scheduling

[Jonathan Morton]

> On 18 Jul, 2019, at 3:20 am, Dave Taht <dave.taht at gmail.com> wrote:
> 
>> What it should do instead is peek the queue and drop until it hits a
>> markable packet, at the very least.
> 
> I didn't say this well. It should drop otherwise markable packets until it
> exits the loop, and then mark the one it delivers from that flow, if it delivers
> one from that flow. That gets rid of all the extra mass ecn creates...

You know, I think I finally understand what you're talking about here.  You want to treat cases where the marking rate exceeds the flow's packet delivery rate as an overload condition justifying a shift to packet drops.

This actually seems like a sane idea.  There is I think one caveat: selecting the Codel 'interval' parameter will now have an increased penalty for getting it wrong, especially on the too-small side.

 - Jonathan Morton