Hi Toke,
The two critical references are
this paper and
this PhD thesis. The former describes "cherish-urgency" multiplexing. The "cherish" is what is different to today's scheduling. It is used to create a new class of algorithm whose goal is global optimisation, not local optimisation (and global pessimisation).
The latter describes a paradigm change from "build it and then reason about emergent performance" to "reason about engineered performance and then build it". It
works in practise, so whether it works in theory is left as an exercise to the reader.
The first step is to get the measurement right. I'm running a public
workshop in London on 8th Dec, and I am happy to accommodate anyone from this list at our internal cost.
Everyone working on AQM has done the best possible within the paradigm they are operating. There is a bigger box of possibilities available, but it needs you to engage with a paradigm change.
Martin