Paper Review: Dynamic Behavior of Slowly-Responsive Congestion Control Algorithms

Reviewer: Kenneth Chin

The purpose of this paper is to convince people that slowly-responsive congestion control algorithms can coexist with the current TCP's AIMD congestion control algorithms without very much affecting the performance of the latter. Through simulations and analysis, the author argues that most of the TCP-compatible algorithms (with self-clocking) are safe for deployment in the sense that they are capable of avoiding causing the network to go into persistent overload persistent loss rates on sudden bandwidth reductions. The findings show that in return for smoother transmission rates, slowly-responsive algorithms lose throughput to TCP under dynamic network conditions.

In the paper, it is mentioned that slow-responsive congestion control algorithms are good for real-time applications. It seems that a smoother bandwidth benefits real-time applications. However, my view is that for real-time applications over the Internet where it does not guarantee in-order-arrival, the packet drop rates and how fast those dropped packets can be retransmitted and reach the destination within a reasonable period of time are the major issues that make real-time applications over the Internet successful. In the paper, most of the simulations were done in a predefined constant drop rate (say 10%). Therefore, the findings do not reflect truly whether the slow-responsive congestion control algorithm is better than TCP in the area of real-time applications. Moreover, in one of the analyses, it shows that in a more bursty condition, TFRC and TCP(1/8) did a poor job in maintaining throughput smoothness and high throughput. As a result, the findings give me an adverse impression that, under dynamic network conditions, slow-responsive congestion control algorithms are no longer TCP-compatible. Therefore there is no incentive to encourage the incorporation of TCP-compatible protocols into the Internet.

There are three main ideas in the paper:
  1. conservation of packets is a very important principle
  2. smoothness in an expense of lower throughput
  3. TCP-compatible protocols can coexist with TCP protocol
I would give this paper the 1st grade because it gives a negative contribution.