Congestion management in high-speed networks

The focus of this dissertation is on congestion control algorithms in rate-based networks. We consider the problem of how to allocate available bandwidth fairly between competing flows and the effect of such allocation on window-based transport protocols, such as TCP. We study TCP behavior and analyze its performance in both one-way and two-way traffic scenarios, not only in pure rate-controlled environments, but also, in an environment where classical IP datagram networks are interconnected by rate-controlled segments.; We develop an explicit rate allocation algorithm for ATM switches, that achieves max-min fair allocation of the available bandwidth, and can satisfy minimum bandwidth requests.; The source of a connection responds to the explicit rate feedback received from the network by adjusting the rate of transmission using and increase/decrease policy. We analyze the source policy, derive approximate analytical closed-form expressions to describe the rate increase process, and use these approximations to analyze the impact of the source algorithm on the TCP slow-start process when operating over a rate-controlled network where the rate allocation is performed explicitly in the switches.; Next, we examine the behavior of a window-based congestion control algorithm over a rate-controlled channel (for example, TCP operating over ABR service). The bunching of acknowledgements (ack compression) in the end systems in such an environment under two-way traffic can lead to significant loss of throughput and unfairness. We develop an analytical model for the IP queue behavior at the end systems and use it to estimate the throughput of the TCP connections.; We explore solutions to the ack compression problem. We study three approaches to improve the throughput of the connections: (i) transmitting acknowledgments at a higher priority than data, (ii) using a connection-level backpressure mechanism to limit the maximum amount of data buffered in the outgoing IP queue of the source of the low-bandwidth connection, and (iii) using a scheduling mechanism to allocate bandwidth between data and acks. We show that only the third approach is able to provide isolation and to control the connection throughputs precisely.; Finally, we study the performance of TCP in an internetwork consisting of both rate-controlled and non-rate-controlled segments. We evaluate the effectiveness of intelligent packet discard policies such as Random Early Detection (RED) and Drop-from-Front in improving the TCP throughput and fairness in such environments. We also develop an explicit feedback scheme based on modifying the receiver's advertised window in TCP acknowledgments returning to the source. Our results show that this explicit window adaptation scheme results in superior throughput and fairness over the packet discard policies studied. (Abstract shortened by UMI.)...