From: Erika Rice (erice@cs.washington.edu)
Date: Sun Oct 17 2004 - 16:08:32 PDT
"Random Early Detection Gateways for Congestion Avoidance" by Sally
Floyd and Van Jacobsen:
Efficient congestion control and avoidance are important parts of a good
network. In their paper, "Random Early Detection Gateways for
Congestion Avoidance" Sally Floyd and Van Jacobson describe an algorithm
for congestion avoidance that does not depend on end hosts cooperating.
Using their method, routers in the network randomly drop packets based
on how full the queue is.
This avoids some of the problems with bursty traffic and fairness that
other algorithms have (although their algorithm is not guaranteed to be
fair, just likely to be fair because of the randomness of the packet
drops). It works by making the end hosts think there is congestion in
the network and, therefore, makes them back off before there is actual
congestion.
Although the algorithm presented by the authors performs well, it does
have a drawback. As a consequence of not depending on the cooperation
of the end hosts, it must be implemented in all of the routers. While
the algorithm does not violate the end-to-end argument (the routers need
not know of the virtual connections between hosts), it would be hard to
implement in practice because of this. Deploying this algorithm
throughout a large network would be very difficult.
However, the method could still be used. As the authors point out,
since each router only depends on its own ability to drop packets and
not on the behavior of end host or other routers, it could be deployed
incrementally. If this method could overcome the deployment issue would
likely be a good way of avoiding congestion without the risk of end
hosts behave unfairly by not backing off when told to.
This archive was generated by hypermail 2.1.6 : Sun Oct 17 2004 - 16:08:32 PDT