Performance analysis of error recovery and congestion control in high-speed networks

In the past few years, Broadband Integrated Services Digital Network (B-ISDN) has received increasing attention as a communication architecture capable of supporting multimedia applications. Among the techniques proposed to implement B-ISDN, Asynchronous Transfer Mode (ATM) is considered to be the most promising transfer technique because of its efficiency and flexibility.In ATM networks, the performance bottleneck of the network, which was once the channel transmission speed, is shifted to the processing speed at the network switching nodes and the propagation delay of the channel. This shift is because the high-speed channel increases the ratio of processing time to packet transmission time and also the ratio of propagation delay to packet transmission time. The increased processing overhead makes it difficult to implement hop-by-hop schemes, which may impose prohibitably high processing at each switching node. The increased propagation delay overhead makes traffic control in ATM a challenge since a large number of packets can be in transit between two ATM switching nodes. Because of these fundamental changes, control schemes developed for traditional networks may not perform efficiently, and thus, new network architectures (congestion control schemes, error control schemes, etc.) are required in ATM networks.In this dissertation, we first present an extensive survey of various traffic control schemes and network protocols for ATM networks. In this survey, possible traffic control schemes are examined, and problems of those schemes and their possible solutions are presented. Next, we investigate two key research issues in ATM networks (and other types of high-speed networks): the effects of protocol-processing overhead and the efficiency of traffic control schemes.We first investigate the effects of protocol-processing overhead on the performance of error recovery schemes. Specifically, we investigate the performance trade-offs between link-by-link and edge-to-edge error recovery schemes. Our results show that for a network with high-speed/low-error-rate channels, an edge-to-edge scheme gives a smaller delay than a link-by-link scheme. We then investigate the effectiveness of a priority packet discarding scheme, a congestion control mechanism suitable for high-speed networks. We derive loss probabilities for each stream and investigate the impact of burstiness of traffic streams on the performance of individual streams