Effectiveness of congestion avoidance: a measurement study

The authors describe the implementation of the binary feedback congestion avoidance (CA) policies and report measurements of the CA scheme in an OSI testbed at MITRE. The goal is to experiment with network configurations and traffic loads varied and complex enough to allow realistic performance evaluation of CA. These measurement experiments present evidence that the congestion avoidance policy operates effectively with a variety of traffic loads and configurations. The conditions used for the measurement experiments with CA include the coexistence of traffic participating in the CA scheme with intermittent traffic that does not participate in CA; and use of CA with realistic traffic such as bulk data file transfer type traffic and remote login traffic. These results show that there is a dramatic reduction in the mean and standard deviation of the response time as well as the number of retransmissions for all classes of traffic, even when some of the sources of traffic do not participate in the CA policy. The behavior of the CA policies with bidirectional traffic was also studied.<<ETX>>