From: Tom Christiansen (tomchr@ee.washington.edu)
Date: Sun Oct 17 2004 - 23:21:46 PDT
This paper presents a novel method for congestion avoidance and control.
This method dynamically adapts network packet sizes to better utilize
available network bandwidth. The proposed method for congestion control is
very simple. It requires an additional state variable in the sender as well
as a few (3) additional lines of code.
The implemented slow-start algorithm does an incredible job at matching the
available bandwidth and provides a significant improvement over the
previous implementation. However, as the name indicates, it starts
slowly... Hence, this protocol will very likely degrade network performance
for small bursts of data, where it will excel in streaming applications. I
wonder if there is a scenario where multiple hosts on the same network
could "fight" over the available bandwidth and end up oscillating between
increasing and decreasing the packet size, thus, only utilizing half of the
available network bandwidth.
Positive: It is a fairly easy paper to read and the results are impressive.
The paper does provide a few pointers to theory which may be pursued at a
later time. I find the inserted pseudo code helpful for understanding the
principle of operation for the congestion avoidance algorithm.
Negative: While providing a teaser for the theory, the theoretical
background for the algorithms is treated rather superficially. The author
does mention that a theoretical background article is in the works, but
apparently it was more important to get the results out the door before the
theoretical background was described...
This archive was generated by hypermail 2.1.6 : Sun Oct 17 2004 - 23:21:53 PDT