Consider this my feedback on bandwidth constraint models.
The Russian Dolls model is well understood and can directly be used to
meet the scenarios in the requirements draft w/o using the network totally
inefficiently, so in the spirit of expedience, I fully support it.
The architecture and protocol supports new models, so should a production
network ever deploy any of this, they can surely help develop a new one to
plug in (such as maximum allocation).
Just in case you didn't have time to read
draft-lefaucheur-tewg-russian-dolls-00.txt, my input at this stage is
exactly in line with Jim's statements above. You can find more detailed
rationale for this view in draft-lefaucheur-tewg-russian-dolls-00.txt.