Network Calculus for Mean Delay Analysis Through a Network

In this paper, a framework is developed to estimate the mean delay performance of (σ, ρ, π) regulated flows in networks with acyclic routing. We first show that the mean delay performance can be bounded by on-off type processes with exponentially distributed off periods. We then obtain per-flow bounds on the mean delay. We show that when there is no peak rate constraint, the Pollaczek-Khinchine formula for M/G/1 queues provides a tight bound thus establishing the Better-than-Poisson property for such flows. We then consider flows inside a network and show that they can be characterized by a stochastic burstiness parameter and show how the aggregate performance can be bounded from the asymptotic Better-than-Poisson property of regulated flows.