Evaluation of 'user-oriented' and 'black-box' traffic models for link provisioning

Abstract

To offer users a sufficient performance level, network links should be properly provisioned. The required bandwidth capacity may be determined through the use of a model of the real network traffic. In this paper, we study the use of two classes of traffic models: (i) 'user-oriented models', which capture the behavior of individual flows, and (ii) 'black-box models', which statistically describe the superposition of many users (and do not distinguish between individual flows). User-oriented models have the advantage that they allow for sensitivity analysis: the impact of a change in the user parameters (access rate, flow-size distribution) can be assessed. In general, however, our measurements indicated that black-box models are easier to estimate, and yield accurate provisioning guidelines.