Scheduling and management of real-time communication in point-to-point wide area networks

Abstract

Applications with timing requirements, such as multimedia and live multi-user interaction, are becoming more prevalent in wide area networks. The desire to provide more predictable performance for such applications in packet switched wide area networks is evident in the channel management provided by Asynchronous Transfer Mode (ATM) networks and in the extensions to the Internet protocols proposed by the Internet Engineering Task Force (IETF) working groups on integrated and differentiated service. The ability to provide guarantees on the performance of traffic flows, such as packet delay and loss characteristics, relies on an accurate model of the traffic arrival and service at each node in the network.

This thesis surveys the work in bounding packet delay based on various proposed queuing disciplines and proposes a method for more accurately defining the traffic arrival and worst case backlog experienced by packets. The methods are applied to the first in first out (FIFO) queuing discipline to define equations for determining the worst case backlog and queuing delay in multihop networks. Simulation results show a significant improvement in the accuracy of the delay bounds over existing bounds published in the literature. An improvement of two orders of magnitude can be realised for a ten hop path and the improvement increases exponentially with the length of the path for variable rate network traffic. The equations derived in the thesis also take into consideration the effect of jitter on delay, thereby removing the requirement for rate controllers or traffic shaping within the network.

In addition to providing more accurate delay bounds, the problem of providing fault tolerance to channels with guaranteed quality of service (QoS) is also explored. This thesis introduces a method for interleaving resource requirements of backup channels to reduce the overall resource reservations that are required to provide guaranteed fault recovery with the same QoS as the original failed channel. An algorithm for selecting recovery paths that can meet a channel's QoS requirements during recovery is also introduced.