Hybrid Simulation of a FIFO Queuing System with Trace-Driven Background Traffic

Abstract

This paper introduces a novel hybrid packet-event / fluidflow network simulation scheme. A packet-event simulation technique uses the arrivals and the departures of packets to model the queuing system. The applications that need finegrained performance details, are simulated with an adapted event based approach. The impact of the background traffic on these foreground packets is simulated by virtual packets, which are created and put in the queue each time a foreground packet arrives. The calculation of the number and the size of these virtual packets is based on a fluidflow approximation of the buffer occupation probability density function of the background stream. A Many Sources Large Deviations traffic descriptor is used to characterize the fluid-flow. A numerical evaluation of this hybrid simulation scheme is performed with a video streaming application as foreground traffic and measured network traces as background traffic.

Stochastic hybrid simulation

“…In this simulation scheme, first proposed in Lauwens et al (2007), the splitting of the traffic into foreground and background traffic is combined with the general idea of using a large deviations technique to get insight in the rare events of interest. The background fluid-flows are no longer described by a set of differential equations but a stochastic model is used.…”

Stochastic hybrid simulation with applications to queuing networks

“…The novel stochastic hybrid methodology is a model in which the foreground packets are simulated in an eventdriven way. In this simulation scheme, first proposed in [1], the splitting of the traffic into foreground and background traffic is combined with the general idea of using a large deviations technique to get insight in the rare events of interest. The background fluid-flows are no longer described by a set of differential equations but a stochastic model is used.…”