Zoe Antoniou

Scheduling Algorithms for Application-Level QoS Support of Continuous Media Streams

Thursday, December 16, 1999
3:00 PM
206 Egan

Abstract

The Quality of Service (QoS) of continuous media applications over ATM networks is affected in a rather direct manner by the timely delivery of Application-level Data Units (ADUs) which are typically mapped into several ATM cells. An investigation of the problem of resource management (scheduling) at the application-level time scale is proposed in this work. In particular, ADU based, deadline driven scheduling policies with ADU based QoS are proposed. The supported applications are considered to be diverse in terms of both the required QoS and the traffic characteristics.

In the first part an ADU-based Earliest Due First (EDF) scheduling policy is proposed. It is investigated under periodic ADU arrival processes which are typical of continuous media streams. The proposed policy can be implemented in terms of a simple First ADU In - First ADU Out (FAI-FAO) policy and it is shown to improve on the ADU dropping rate induced by FIFO policies of comparable complexity.

In the second part the more complex and effective Size-Deadline Sensitive(SDS) scheduling policy is proposed and investigated under general ADU arrival processes. It is shown to be optimal in the sense that it maximises the number of complete ADUs scheduled for transmission at any point in time, by taking into consideration the diverse ADU deadlines and sizes and combining the principles of two well known scheduling algorithms, the EDF and the Shortest Job First (SJF). The induced ADU and cell loss rates are compared against those induced by the standard EDF and SJF policies.

In the third part, the Deadline Credit (DC) policy is proposed and investigated under prerecorded continuous media traffic for both single hop and multi-hop networks. Its objectives are to utilise all of the available capacity by allocating bandwidth dynamically to all data streams, to distribute the available bandwidth fairly between all competing data streams and to maximise the number of ADUs that are served by always giving priority to the one with the shortest deadline. In order to achieve this it exploits the available bandwidth when the system is underutilised in order to send ADUs in advance. This results in building up a deadline credit for future ADUs and, thus, reducing the losses during periods of congestion. The induced ADU loss rates are compared against those of alternative approaches which do not allow any dynamic deadline credit build up. Also, results are presented which illustrate the fairness of the proposed policy.

Thesis Committee:
Prof. I. Stavrakakis (advisor, external member)
Prof. E. Manolakos (co-advisor)
Prof. M. Salehi