Analyzing Traffic and Multicast Switch Issues in an ATM Network.

Abstract

This dissertation attempts to solve two problems related to an ATM network. First, we consider packetized voice and video sources as the incoming traffic to an ATM multiplexer and propose modeling methods for both individual and aggregated traffic sources. These methods are, then, used to analyze performance parameters such as buffer occupancy, cell loss probability, and cell delay. Results, thus obtained, for different buffer sizes and number of voice and video sources are analyzed and compared with those generated from existing techniques. Second, we study the priority handling feature for time critical services in an ATM multicast switch. For this, we propose a non-blocking copy network and priority handling algorithms. We, then, analyze the copy network using an analytical method and simulation. The analysis utilizes both priority and non-priority cells for two different output reservation schemes. The performance parameters, based on cell delay, delay jitter, and cell loss probability, are studied for different buffer sizes and fan-outs under various input traffic loads. Our results show that the proposed copy network provides a better performance for the priority cells while the performance for the non-priority cells is slightly inferior in comparison with the scenario when the network does not consider priority handling. We also study the fault-tolerant behavior of the copy network, specially for the broadcast banyan network subsection, and present a routing scheme considering the non-blocking property under a specific pattern of connection assignments. A fault tolerant characteristic can be quantified using the full access probability. The computation of the full access probability for a general network is known to be NP-hard. We, therefore, provide a new bounding technique utilizing the concept of minimal cuts to compute full access probability of the copy network. Our study for the fault-tolerant multi-stage interconnection network having either an extra stage or chaining shows that the proposed technique provides tighter bounds as compared to those given by existing approaches. We also apply our bounding method to compute full access probability of the fault-tolerant copy network.