Modeling and performance analysis of carrier sense multiple access protocols on star-like topologies
Ivan Vukovic · 1994
This thesis is about carrier sensing protocols on communication network topologies that we term star-like. Unidirectional in nature, star-like topologies include Broad-band Ethernet, StarLan, Fibernet II and wireless single hop networks with a base station. We relax a widely used, but inaccurate, equidistant-stations assumption. We specifically investigate the impact on the throughput performance of the station and traffic load distribution in space. For this, we propose a new space-time model which traces events in star-like topologies. The arrival process is modeled as a time-homogeneous: space-inhomogeneous Poisson process. Under these assumptions we calculate the throughput of nonpersistent, p-persistent and 1-persistent CSMA and CSMA/CD. Our analytical results indicate that the previous analyses, based on the equidistant stations assumption, significantly underestimate the normalized throughput. Simulation results of IEEE 802.3 Broadband Ethernet LAN match well with our analytical results. The simulations also reveal that the Poisson retransmission assumption is a good approximation. Under the Poisson assumption we calculate the probability of success for a random arrival as a function of the distance from the repeater. This probability provides an exact measure of unfairness in star-like topologies. Finally, we exploit our space-time diagram to investigate the impact on performance due to power capture at the repeater. The analysis of CSMA with power capture is an important problem in packet radio and wireless networks which has not received much attention in the past. We analyze two capture models which yield an upper and a lower bound on the throughput. The models assume that capture is solely a result of different propagation losses for different packets. Our numerical results show that even with a pessimistic capture model the throughput of CSMA improves at the expense of degraded fairness of the system.