Modelling multi-dimensional QoS: some fundamental constraints

In this paper, we model multi-dimensional QoS in a unified framework, and study some fundamental constraints from the network and the traffic on realizing multiple QoS goals. Multidimensional QoS requirements are quantitatively represented using a QoS region. Based on the theory of effective bandwidths, the framework connects the throughput, the delay, and the loss rate in a uniform formula. Important traffic and network factors, namely, the burst size and the link speed are involved. With this framework, it is found that the burst size sets hard limit on the QoS region that can be achieved, and that the matching between the link speed and the node processing power can greatly improve the limit. It is also made clear that while pure load imbalance among links does not affect the QoS region, the heterogeneities of burst size or link speed may severely degrade the QoS performance. Applying the theory to real-time services in differentiated services architecture, we show it provides a useful tool for QoS prediction and network dimensioning.