Paper Review:
A Comparison of Mechanisms for Improving TCP Performance over Wireless Links

Reviewer: Jie Zhou

Problem

One assumption of traditional transport protocols is that packet losses occur mostly because of congestion. However, lossy links such as wireless networks also suffer from significant losses due to bit errors ad handoffs. How to improve the performance of TCP in such networks.

Contribution

The paper studies the exsiting schemes for TCP on lossy networks. The author categorizes the schemes into three broad categories (end-to-end protocols, link-layer protocols, and split-connection protocols), and presents a comparative analysis of them.

Main Ideas

  • Knowledge of TCP can improve the performance of link-layer protocol.
  • Splitting the end-to-end connection is not a requirement for good performance.
  • Selective acknowledgement schemes are very useful in the presence of lossy links, especially when losses occur in bursts.
  • End-to-end schemes are promising because they do not need extensive support from intermediate nodes in the network.

    Critique

    I give the paper a rate 4 (significant contribution), because it provides thourough analysis of the current techniques for TCP in lossy links. Wireless networks are developing dramatically in recent years. It is very important accomodating wireless networks in Internet. The analysis in this paper provides a solid basis for further research.

    The paper is quite convincing: it not only gives extensive experiment data, but also deeply analyze the cause behind them.

    Two limitations in this paper:
  • In the experiments of end-to-end schemes, the author assume the receiver has sufficient knowledge about wireless loss. This is not true in reality.
  • The experiments are done in the absence of congestions.

    Open Questions:

  • How about the protocol enhancement in the presence of both network congestion and wireless losses?
  • What is the impact of large variations in connection round-trip times?
  • What is the impact of bandwidth and latency asymmetry on transport performance?