Fluid-flow modelling of internet traffic in GSM/GPRS networks

  • Authors:
  • Tim Irnich;Peter Stuckmann

  • Affiliations:
  • Chair of Communication Networks, Aachen University of Technology (RWTH), Kopernikusstr, 16, D-52074 Aachen, Germany;Chair of Communication Networks, Aachen University of Technology (RWTH), Kopernikusstr, 16, D-52074 Aachen, Germany

  • Venue:
  • Computer Communications
  • Year:
  • 2003

Quantified Score

Hi-index 0.24

Visualization

Abstract

In order to find an analytical solution for performance evaluation of Internet access over the General Packet Radio Service (GPRS), the packet-switched extension of the GSM mobile radio network, we evaluate the well-known Fluid-flow modelling (FFM) approach. Mobile Internet users are modeled by an ON/OFF source with exponentially distributed sojourn times in the ON and OFF state. We choose the IP datagram delay at the radio interface as quantity of interest. As Fluid-flow analysis only delivers the mean equilibrium buffer content, we present some additional methodology to calculate the sojourn time in the Fluid-flow model, taking specific characteristics of the GPRS into account. ON/OFF source parameters are determined representing the mean offered traffic in typical GPRS load scenarios by stochastic simulation, using the GPRS emulation system GPRSim. The GPRSim comprises load generators representing typical GPRS usage and a prototypical implementation of the GPRS protocols. Comparison of simulation results and analysis is performed using the same traffic source model for both simulation and analysis, eliminating approximation of WWW traffic characteristics as a reason for deviations between simulation results and analysis. Additionally analytical results are compared with results of the GPRSim using a detailed WWW model for traffic generation. The results illustrate the influence of accurate source modelling and the limitations of the FFM to model the GPRS system with elastic traffic. It is shown that the FFM is not able to depict the elastic property of TCP-based Internet traffic, but is capable of modelling the multiplexing of inelastic traffic like generated by streaming or real-time applications over the GPRS radio link in sufficient detail for performance estimations.