Self-similar traffic modeling and Clos-like packet switch architectures

Abstract

Two methods of generating self-similar Internet traffic are reported in this document. The first method uses Heavy-tailed Pareto distribution and generates fixed length packets, while the second method focuses on Fractional Gaussian Noise as means of generating variable-length packets. Various testing have been performed on the traffic to corroborate its self-similarity and measure relevant statistics. Using the self-similar traffic environment, the performance of a sectored three-stage packet switch architecture is evaluated which benefits from both photonic and electronic technologies and provides variable bandwidth between sector pairs by changing the configuration of the photonic centre stage according to arriving traffic. This architecture reduces complexity, at the price of a modest spatial speed-up, which is easy to provide with photonic technology. As an alternative to reconfiguration, a novel load-balancing Clos-like architecture and method without miss-sequencing is proposed. The architecture involves no central schedulers and is implemented by distributing cross-point queues over all elements of the switch and deploying pollers to select the queues to be served. This method is simple and practical and shows delay performance that approaches that of an ideal output queued switch.