Year

1999

Degree Name

Doctor of Philosophy

Abstract

In recent years, the variety, and use of Internet applications has grown enormously. As a result, the Internet is moving from providing solely unicast delivery and best effort service, to also providing multicast delivery and real time service support. Asynchronous Transfer Mode (ATM) is also being designed to support these two delivery and service classes. Furthermore, as the deployment of ATM in both the wide and local area has increased in recent years, there has been growing interest in the problem of providing Internet services over ATM networks. Selecting a delivery mechanism for carrying Internet traffic over ATM infrastructure is a core issue that must be resolved. Many schemes for providing Internet services over ATM have been proposed, however there is no clear consensus in the literature of which scheme should be followed. Moreover, a detailed review of the literature reveals that little analysis examining the performance of these schemes has been performed. This thesis addresses this problem by presenting a detailed analysis of schemes for providing unicast and multicast Internet delivery over wide area ATM networks. The first part of this thesis develops a queuing analysis framework to enable the detailed performance comparison of delivery schemes. This work extends existing queuing network methodology to model: point-to-multipoint ATM connections; the segmentation and reassembly of Internet packets into ATM cells; and the Internet Transmission Control Protocol Slow Start mechanism. The second portion of this thesis develops a reahstic application and network model for carrying World Wide Web traffic over the Internet. This model is used to compare the key alternative approaches for unicast delivery: the Hop-by-Hop approach; and the Buffered and Hybrid variants of the Cut-Through unicast delivery approach. We find that both the Buffered and Hybrid Cut-Through approaches provide significantly lower response times than the Hop-by-Hop approach, even when the end-to-end (or cut-through) VC must be created, as long as the signalling network is able to support the expected traffic volume. The analysis also shows that the Buffered approach out-performs the Hybrid approach in most network scenarios.

Share

COinS