The purpose of this paper is to investigate the problem of achieving optimal rates in a distributed environment using only the information available at the end hosts.
The main contribution of this paper is that a user-centric optimization algorithm, which does not require any explicit network feeback and is therefore able to be deployed on the Internet, achieves the system optimum at equilibrium.
(1) While fairness is a desirable property to have, fairness alone may
not be a good objective. The available bandwidtch should also be
allocated in such a way that the overall utility of the users is
(2) The authors' proposed algorithm can be implemented over the current Internet (without extensive modifications inside the network) by a simple modification of the existing TCP at the end hosts.
I would rate this paper as a 4 because this paper does present a deployable algorithm which is a slight modification to TCP at the end host. As the authors mentioned in the paper, their work is really a natural extension of previous work by other authors.
To justify their results, the authors addressed the problem in a mathematical setting by providing proofs of network performance under given assumptions. In terms of actually showing that this idea worked in practice, the authors only provided a "numerical example," which makes it seems as if their approach may not be as deployable as they claim. (Otherwise, why wouldn't they have implemented it instead of just giving us one simulation example).
One limitation of their approach is I'm not convinced of the effective deployability of their algorithm. (How does it work if not everyone in the network has their modification?)
One lesson researchs should take away from this work is that network equilibrium can/should be achieved through local user optimization of some pricing function as opposed to attempting implement a central authority within the network.