Performance of unslotted CSMA-CD protocols for high-speed fiber optic networks

Abstract

The main objective of this dissertation is to investigate the stabilities, fairnesses, and delay-throughput performances of the unslotted CSMA-CD protocol and its variants for the fiber optic bus and ring networks with a finite number of stations, each of which has an infinite storage buffer. The unidirectional transmission property of an optical fiber leads to remove the negative effect of the propagation delay on the performance of the conventional CSMA-CD systems and to give the preference of accessing the channel to upstream stations, and therefore each network can be treated as an M/G/1 priority queueing system. In this work we investigate the performances in five different aspects. First, we analyze the unslotted 1-persistent CSMA-CD (CSMA-CD/U/1) protocol for both the networks. The preferential-access property leads to treat each network as an M/G/1 priority queueing system with the preemptive-repeat-identical retransmission strategy. We analyze this queueing system by using the busy-period analysis method. From the numerical and simulation results, we show that CSMA-CD/U/1 is better than the unslotted non-persistent CSMA/CD protocol, but it makes the ring network unstable due to the channel saturation effect. Second, we consider a method to make CSMA-CD/U/1 more efficient by reducing the number of incomplete packets generated by preemptions. For this purpose, as a protocol with an improved retransmission strategy, we present an unslotted 1-persistent CSMA-CD protocol with a combined preemptive-repeat-identical/preemptive-resume retransmission strategy (CSMA-CDC/U/1) for both the networks. Under this protocol, if the information field being transmitted is preempted, its remaining portion is retransmitted with the same header. However, when the header field is preempted, it is retransmitted from its beginning. Treating each network as an M/G/1 priority queueing system with the combined retransmission strategy, we analyze this protocol by using the busy-...