Heterogeneous maximal-throughput bursty traffic model with application to packet switches

The paper presents an analysis of a discrete-time Markov-modulated process for generating heterogeneous bursty traffic in input-queued switches. The goals of the model described are two fold. First, it constitutes an expansion of the classic ON/OFF process to address scenarios in which multiple queues receive bursts from a shared input stream. Second, in concert with the first goal, it enables 100% link utilization (hence maximal traffic generation throughput) by avoiding the mandatory OFF slots introduced by classic ON/OFF models. This work is especially valuable in evaluating the performance of packet switching fabrics under bursty conditions