Techniques for the study of QoS in IP networks

Setting the many parameters available to the operator in a QoS-enabled IP network in order to achieve peak performance is a highly non-trivial task. Individual settings are not independent; and their consequences depend upon the traffic and application mix, the behaviour of higher-level protocols, the remainder of the network, and the customer's response. To complicate matters further, settings which optimise mean achievable bandwidth for large files may be very different from those best for Web-browsing applications; and within a single class the preferred values to minimise mean or median delays may be different from those which minimise different percentiles. To study the behaviour of all these, a range of techniques needs to be employed. This paper gives examples of analytic, simulation and experimental approaches, and shows how they are all needed for an in-depth understanding. [PUBLICATION ABSTRACT]