From: Masaharu Kobashi (mkbsh@cs.washington.edu)
Date: Mon Oct 18 2004 - 00:54:42 PDT
1. Main result of the paper
The paper describes new algorithms incorporated in BSD TCP after the
first congestion collapse of the Internet in 1986. These algorithms
include round-trip-time variance estimation, slow-start, and additive
increase / multiplicative decrease. The author also presents rationale
behind the design of the algorithms and the impressive measurements of
their effects.
2. Strengths in this paper
The proposed new algorithm has actually improved the Internet's
congestion problem significantly. It is a great contribution to the
Internet community. It is surprising that such effective methods are
based on rather simple algorithmic changes.
One interesting aspect of their method for avoiding congestion is
that they do not rely on the cooperation of routers on the paths.
It is impressive that they can achieve that much improvement with the
the changes of end points' algorithms only.
Another contribution of the paper is it made the causes of the congestion
clear by the statement which says there are only three ways for packet
conservation to fail (the three cases are listed after that). This
observation is central to their improvement as well as following
further enhancements.
3. Limitations and suggested improvements
The author does not give justifications for using some constants
for variables in formulas. It is not persuasive and looks ad hoc.
The paper assumes that the path from the source to the destination
is the same as the path for ACKs from the destination to the source.
But it is not necessarily so. In the cases where these paths are not
the same, the reasoning for leading to equilibrium may not hold when
the causes of packet loss on the two different paths are different.
It does not consider possibilities of multiple ACKs generated by
malfunctioning hosts, malicious hosts, or nodes on the paths.
4. Relevance today and future
The assumption that packet loss due to damage to packets
is extremely are is no longer true in the rapidly increasing
wireless networks. In the current trend of the Internet usage patterns,
the proposed algorithms may not work as expected, although I heard
further enhancements have been done to TCP after this initial
improvements.
This archive was generated by hypermail 2.1.6 : Mon Oct 18 2004 - 00:54:43 PDT