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

Reviewer: Oleg Elkhunovich


TCP is not tuned to provide good performance in lossy networks such as wireless. Authors of this papers explore different schemes of protocols that might solve this problem and allow for better performance over wireless links.


The contribution of this paper is a detailed analysis of different methods to overcome TCP's inability to provide good performance in a lossy network. Authors come up with some interesting ideas that point in new directions of research for this problem.

Main Ideas

  • TCP-aware link-layer protocol with selective acknowledgement performs the best
  • The split connection approach has problems because sender often stalls as due to timeouts on the wireless connection, resulting in poor end to end throughput. Well-tuned link layer scheme is better even than SMART-based split connection protocol. Therefore, 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


    Significance: 3
    The paper is important because it gives a very good comparison of different methods to overcome a stated problem. However, not all possibilities are considered and no real breakthrough conclusions are made. This paper merely opens a door for new promising research in combining SACKS and ELN.
    The methodology of this paper is quite good. It is very well organized and nicely splits up and discusses different schemes - candidate solutions. However, experiments seem to have some flaws and are rather idealized. There are still some questions about the correctness of the results in the real world. Also, no theoretical background is given to prove some statements that paper makes.
    Only experimental approach is considered, and experments themselves are closer to simulation than real world environment.

    Methological analysis of possible solutions to a problem can lead to an area of further interesting research.