End to end performance in real time IP networks with self-similar behaviour

IP network technology is being increasingly used to support real-time telecommunications services. Network operators need to be able to dimension service partitions and virtual buffer sizes for efficient resource utilisation whilst meeting service level agreements, but without resorting to over-dimensioning. To provide the necessary quality of service guarantees for real time traffic, it is vital to understand the impact that self-similar behaviour can have on end to end delays. This paper introduces approximate analytical methods for assessing end to end delay variation bounds which yield accurate results over the full range of partition loadings, in contrast to other accepted approaches which are only valid for heavy traffic. Comparison of results against network simulations, in which local criticalities produce heavy-tailed delay distributions as a sub-component of the end to end performance, show good agreement. This demonstrates a key result, that the effect of self-similar behaviour on real-time service performance can be accurately assessed with easy-to-use analytical techniques incorporating Markovian assumptions.