This represents a laborious effort with lack of originality

Review: TCP Vegas

Reviewer: Di Niu

This represents a laborious effort with lack of originality

This paper does not sound interesting to me, as it does not provide
any further insights into congestion control. Rather, it only
meticulously tuned engineering details to improve TCP performance.
Generally speaking, this is more of engineering than research. It is
primarily a modification made to [7] "Congestion avoidance and
control" by Jacobson, and thus lacks originality in a researcher's
perspective. I guess this paper was accepted by JSAC solely because
it did very solid experimental studies.

Several problems of Reno TCP are pointed out in the paper. Remedies
are proposed accordingly. Specifically, a new retransmission
mechanism, a new congestion avoidance mechanism and a modified slow-
start mechanism are adopted in TCP Vegas. However, these new
techniques are still based on the basic insights proposed in [7], and
therefore are much of engineering details. For example, in Vegas, the
new retransmission mechanism tries to further reduce its dependency
on the coarse-grain clock ticks by estimating RTT more accurately.
Retransmission details have also been modified. However, TCP Reno has
the same aim at this point. Thus, vegas is much of an engineering
improvement rather offering further research insights. Similarly, the
congestion avoidance mechanism is a trivial improvement on Tri-S.

Moreover, although the new schemes mitigate the the problems of TCP
Reno, they do not cure them. For example, it is said that Reno has
the problem of not knowing the safe window size at the start stage.
Vegas proposes to mitigate this problem by allowing exponential
growth only every other RTT and some other detailed methods. However,
the new methods pose two problems, and Vegas* is proposed which
further has raised new problems. The paper ends up being trapped into
the problem-solving cycle incurred by itself.

Doubtlessly, the paper is solid and strong in its experimental
studies. However, given enough time and strong wills, different
people can generate totally different engineering details with
respect to the same topic. In other words, the paper lacks of
originality.
Received on Wed Oct 04 2006 - 17:21:16 EDT