This paper addresses the problem of protecting against ill-behaved sources that hog bandwidth. The paper proposes a fair queueing solution that addresses this problem at the gateway level. Essentially, the idea of a fair queue (borrowed from a previous paper) means that the router needs to keep track of each individual flow and alot to it only as much bandwidth as it requests.
The authors of this paper improved Nagle's fair queue algorithm in these ways:
I give the paper a rating of 3 for modest contribution. I think the methodology used is pretty convincing that fair queues run comparitively better than first-come-first-serve queues. However, there are some limitations in the simulations. For example, the authors used only FTP and Telnet flows to run tests. Since 1989, the network contains mostly HTTP flows. Also, in the recent years, real-time streaming and interactive applications have also become popular. These new applications need to be tested against the fair queue to see how well it performs.
As far as implementing the fair queue, I think the technology is already available for doing so, so that's no longer a drawback.