Performance of collision-free protocols for local area networks with star topology

Abstract

The main objective of this dessertation is to investigate the performance characteristics of a broadcast star network with collision avoidance switch and a multichannel local lightwave network with grouping property. So far, quite a few attempts have been made to model the broadcast star network with collison avoidance switch. Most of those models assumed constant message length and random selection policy at the central hub. If the central hub can choose the oldest message among the messages which are transmitted at the same slot, tnen the service discipline becomes first-come first-served (FCFS). Although the FCFS didcipline has many desirable featrues such as fairness and lowest variance of message delay, this discipline is not easy to implement since the central hub does not know the arrival times of messages. If the central hub chooses one message randomly among the messages which are transmitted at the same slot, then the service discripline becomes random-order of service(ROS). While the ros discipline is easy to implement, the message delay is not bounded and the variance of message delay becomes large in this descipline. For this reason, we first proposed a service disciplin which is relatively easy to implement and which has the variance characteristics of message delay almost the same as the FCFS case (ideal case). the proposed service discipline is combination of FCFS and ROS disciplines and it uses reservation and priority fields in the message header. Using the proposed seervice discipline, we presented models for a broadcast star network with collision avoidance switch separately in two casesi a broadcast star network with small propagation delay and a broadcast star network with large propagation delay. We assumed that the network consists of an infinite population of stations which collectively generate new messages according to a Poisson distribution and that the size of each message is distributed according to a general distribution. We also ...