A measurement-based loss scheduling scheme

In high-speed communication networks such as the ATM-based B-ISDN, a frequently addressed issue in traffic management is how to effectively manage network resources to meet different quality of service (QoS) requirements of multiple traffic streams. We consider a measurement-based loss scheduling scheme, termed the QoS-scheme, for high-speed packet networks. We show that the QoS-scheme is optimal in terms of bandwidth utilization among all stationary, space conserving loss scheduling schemes, and that it is also optimal among all stationary loss schemes when all traffic streams are equally demanding. We examine the properties of the QoS-scheme and how these properties may be used to compute the minimum bandwidth required for a given traffic scenario. We also discuss the implications of these properties and compare the QoS-scheme with other loss scheduling schemes such as the last-in-first-out, partial buffer sharing, and push-out scheme through simulation studies.