What I'd like to know, actually, is what feedback I can give to the edges with respect to what traffic they can give to make things better or worse.
I'd also like to know what kind of a throughput hit I will take by doing flow switching instead of optimum or distributed switching (what does it cost me to know what a DS3 is actually doing?), and whether it's potentially scalable.
And ultimately, a model in which we can characterize real limits to networks (humanly noticeable lousy end-to-end throughput being the real evaluator) given the same architecture but different traffic patterns (e.g., something like discovering that a tweak to TCP in N-thousand low-bandwith TCP flows per time unit will degrade overall end-to-end goodput through an internet with O-thousand TCP flows of distribution L that has this topology by about Z%)