h1>Paper Review:
TCP Vegas: End to End Congestion Avoidance on a Global Internet

Reviewer: Robert Dugas


This paper attemts to identify and address inefficiencies and overhead present in current TCP Reno implementations.


This paper proposes a number of modifications to TCP Reno and provides test data to back up the claim that these modifications represent improvements.

Main Ideas

  • A new retransmission mechanism which uses the sytem clock to timestamp to supercede "coarse-grained" periodic timeout checks and duplicate ack counting
  • Congestion avoidance using changes in throughput rate to linearly increase or decrease trasmission rate so as to avoid unnecessary clogging for informational purposes.
  • Modified slow-start to only double every other RTT so as to avoid "overshooting" the optimal rate.


    The modifications proposed seem to represent significant improvements to TCP which, in turn, will result in a more efficient global internet. Methodology:
    Although there is clearly a lot being presented in the paper, some things do appear swept under the carpet. For instance, a long explanation of a complex new slow-start algorithm is presented and its praises sung, then the section ends by saying it didn't work and they didn't implement it without further elaboration. In addition, overhauling the entire internet is always frightening without adequate test data. The authors subdue this fear by stating "there is no reason to expect Vegas to lead to a network collapse". As a final point, I *really* didn't like their graphs.

    The primary lesson learned from this paper is that even accepted standard might represent areas to improve upon. In addition, fancy graphs can be convincing.