Paper review: Charge-Sensitive TCP and Rate Control in the Internet

Reviewer: Mike Liu

  1. State the problem the paper is trying to solve.
  2. The main problem the paper is trying to solve is how to develop a system to achieve the system optimal rate in a distributed environment, allowing total user utility to maximized, using only the information available at the end hosts.
  3. State the main contribution of the paper: solving a new problem, proposing a new algorithm, or presenting a new evaluation (analysis). If a new problem, why was the problem important? Is the problem still important today? Will the problem be important tomorrow?  If a new algorithm or new evaluation (analysis), what are the improvements over previous algorithms or evaluations? How do they come up with the new algorithm or evaluation? 
  4. The main contribution of this paper is that it offers a clear reanalysis of the TCP congestion problem and offers a very novel solution to solving it using a Charge-Sensitive TCP and Rate Control. By incorporating a pricing scheme, the user solves their own optimization problem on a larger time scale which allows the window-based transmission rate control mechanism to solve the network problem on a smaller time scale. The improvements this approach has over previous schemes, namely the currently implemented TCP, is that it attempts to correct for the high packet loss rate and delay bias, and does this with modifications only to the end hosts. The authors came up with this algorithm by incorporating work done by Kelly et al. on propotional fairness using theirprimal algorithm. It is also a natural extension of the fair window-based end-to-end congestion control mechanism developed by Mo and Walrand.
  5. Summarize the (at most) 3 key main ideas (each in 1 sentence.) 
  6. The three 3 key main ideas are: (1) The problem of achieving system optimal rates can decomposed into two problem, the network and user problems, in which the users solve the user optimization problem on a larger time scale, while the underlying window-based transmission-rate control mechanism solves the network problem on a smaller time scale. (2) This algorithm does not require any explicit feedback from the network and can be deployed over the Internet with modifications only to the end hosts, and this is because the price user i pays per unit time, hi, equals its target queue size, pi. (3) The case where the choice of each user's action, when modeled as a noncooperative game, has nonnegligible effect on the price per unit flow at the resources and in the case of a single bottleneck, there exists a unique Nash equilibrium of the game.
  7. Critique the main contribution
  8. What lessons should researchers and builders take away from this work. What (if any) questions does this work leave open?
  9. The lessons that researchers should take away from this work are that a rather difficult problem can often be solved by modeling it more simply, innovatively, and clearly. The solution of the problem often lies in the problem itself. The authors of this paper demonstrated this by modeling the system problem by decomposing it into two small problems, the user and network problems, which were much more readily solvable using a known algorithm, Kelly's primal algorithm. The question this work leaves open is how well can this solution be implemented and what are the transmission rates achieved through the authors' scheme in comparision to rates achieved by the current TCP.