Abstract
The availability of low-cost, high-bandwidth optical fiber will change the technology and use of wide-area networks by allowing lower-cost, higher-speed communication. Higher transmission rates combined with wide geographical distribution will require new protocols for dynamic resource allocation in wide-area computer networks.
This thesis analyzes a scheme to approach the problem of network congestion for high-speed packet switches that provide datagram services.
The analyzed algorithm, which has been propose by Yavatkar [Yava89], avoids congestion by restricting excessive traffic at the entry of the network and by operating the network adaptively on a rate basis at the "knee point" of the delay-throughput curve. At this point a possible increase in throughput is small but would cause a relatively large increase of transmission delay.
The algorithm is analyzed by using basic, simple topologies to get an understanding of its stationary and dynamic behavior. The thesis analyzes the characteristics, problems and compromises of the algorithm. The congestion avoidance and control mechanism basically trades resource efficiency for a small transmission delay inside the network and higher reliability. A low signal-to-noise ration of the traffic measurements limites the effectiveness of the rate control.
The analysis is based on computer simulation experiments using the simulation language CSIM.