From: Ioannis Giotis (giotis@cs.washington.edu)
Date: Sun Oct 17 2004 - 21:37:10 PDT
Congestion control is one of the main issues in large packet networks.
There are two obvious ways to implement some kind of control, at the end
hosts or in the intermediate gateways. The authors present the RED
gateways that perform congestion control.
The paper is full of analytical results, simulations and comparisons
with other strategies. RED gateways implement queue monitoring by
averaging , in order to avoid mistreating burst traffic. When certain
criteria are met, the gateways mark a bit in the packet header or drop
the packet with a probability that depends on the queue.
An interesting part of the paper is the analysis performed to determine
the optimal probability of marking a packet and the choice of averaging
parameters used in queue monitoring. Clearly, different values
completely change the behavior and performance of the gateways.
The RED gateway implementation is well documented with a lot of
supporting simulation results. However, it is still unclear on how the
gateways would perform in larger networks, and more importantly if the
choice of parameters made by the authors are still good in other
scenarios.
Another advantage is the ability to insert these gateways in existing
networks with no support from the end hosts and improve the performance
gradually as the hosts begin to recognize the congestion control bit.
Overall, I found the paper well written, and I enjoyed the presence of
both analytical results from queuing theory and simulations that support
the results. Of course, these results seem obsolete now as networks have
scaled significantly and there exists more experience on the behavior of
ordinary congestion on gateways.
This archive was generated by hypermail 2.1.6 : Sun Oct 17 2004 - 21:37:10 PDT