From: Susumu Harada (harada@cs.washington.edu)
Date: Sun Oct 17 2004 - 20:39:33 PDT
"A Binary Feedback Scheme for Congestion Avoidance in Computer Networks
with a Connectionless Network Layer"
K.K. Ramakrishnan and Raj Jain
The paper describes the details of a congestion avoidance scheme called
the Explicit Binary Feedback Scheme, which I believe is the same as the
DECbit scheme although the authors do not mention that name. The paper is
very well structured, introducing the motivation for dealing with
collisions, noting the two approaches, namely collision control and
collision avoidance, and presenting their algorithm for implementing
collision avoidance.
The paper does a very nice job of presenting numerous alternative
strategies that the authors considered before settling on their version of
the solution, such as in the choice of the decision frequency (section
5.1) and in the choice of the decision function (section 5.4).
The emphasis of the Explicit Binary Feedback Scheme is to allow hosts to
adjust its sending behavior before collision is actually encountered, and
to provide fair distribution of the network resource among the hosts, all
while using minimal network resource in the communication of this
congestion information from the router to the hosts. It is quite elegant
how a single bit set by the router provides the hosts with enough
information to attempt to maximize network throughput collectively. It
also portrays a network eutopia of sorts, where all the hosts are nicely
cooperating and agreeing to adjust their window sizes based on this
scheme.
This raises concerns about the robustness of this scheme in the face of
malicious or misbehaving hosts that fail to abide by this scheme. What
the paper does not address is, what happens when collisions actually do
start occurring? It seems that inevitably some form of congestion control
mechanism (such as the ack-timeout and source quench mechanisms mentioned
in the introduction) needs to be integrated in addition to the congestion
avoidance scheme.
I would also have liked to see their experimental results with more
communication channels than just two. It also seems they made an
oversimplified assumption by saying that "the workload we considered for
the purposes of our design was that each source is considered to have
packets ready to transmit at all times." This would hardly appear to
provide any variation in the network traffic contributed by each host, a
highly unrealistic scenario.
Overall the paper was well organized and easy to follow, with clear
description of the proposed scheme and alternatives. It is interesting to
note that despite the fact that the paper was written in 1988, the scheme
is still not widely adopted today. It would be interesting to see a study
of whether the reason is due to political/economic reasons, or technical
deficiency of the scheme.
This archive was generated by hypermail 2.1.6 : Sun Oct 17 2004 - 20:39:34 PDT