Review: Binary Feedback Scheme

Paper: A Binary Feedback Scheme for Congestion Avoidance in Computer
Networks

Name: Robert Danek.
Course: CS2209, Fall '06.

The paper begins by explaining the difference between congestion
avoidance and congestion control. In the latter, the goal is to "fix"
the problem of congestion at the point at which it begins; however,
congestion avoidance, as its name suggests, attempts to avoid the
problem of the network entering a congested state by operating the
network at the knee of the response time curve.

This paper proposes a congestion avoidance scheme, as opposed to a
congestion control one. One of the main drawbacks of control schemes,
which this paper wanted to avoid, are that they usually involve sending
"choke" packets. This means injecting more packets into an already
overloaded network.

The scheme for avoiding congestion that the paper proposes works as
follows: When a router detects congestion, it will set a binary bit in
part of the network packet header of outgoing packets to 1. At the
receiver end, the value of this bit gets copied into the transport ACK
that is sent back to the source. When the source detects that this bit
is set, it throttles the number of packets it's injecting into the
network. The source will be using some kind of windowing scheme for
sending packets, and the throttling is done by decreasing the window size.

This scheme is extremely efficient in that it does not introduce any new
packets for signalling congestion, aside from an extra bit that needs to
be passed around.

The paper then explores a number of detailed factors affecting the
implementation of the scheme, including Router Policy and User Policy.
Router policy deals with how routers detect congestion; the paper uses a
mechanism that involves monitoring the average queue length, and if it's
greater than one, then congestion has been detected. User policy (at the
source) involves such things as determining how many acknowledgements
should be taken into account before adjusting window size, and by how
much the window size should be changed.

The paper goes on to explore the scheme from both an analytical and
simulation point of view on a number of factors. The factors explored
include feedback singal generation, and policies for decision making.

Overall this was a good paper. It defined a model for examining the
behaviour of the binary feedback scheme, and it explored it from both an
analytical perspective and also ran a number of simulations. It would
have been interesting to see some empirical data on how this scheme
behaves when implemented in a real network.
Received on Tue Sep 26 2006 - 01:04:57 EDT