Analytical performance evaluation of Internet access over GPRS and its comparison with simulation results

We discuss the results of recently published analytical concepts for the performance evaluation of packet-switched mobile radio networks (see Vornefeld, U., Proc. European Wireless 2002, 2002; Proc. Int. Commun. Conf. '02, 2002). We compare them with the results of a simulation system (GPRSim), which emulates the GSM general packet radio service (GPRS) in great detail (see Stuckmann, P., Technical Report, http://www.comnets.rwith-aachen.de//spl sim/pst; Stuckmann and Paul, O., Proc. 10th Symp. Wireless Personal Multimedia Commun., p.597-602, 2001). We try to identify possible reasons for deviations between analytical and simulation results in order to support further development in this area. In Vornefeld's analytical modelling framework, source modelling is done by characterization of WWW traffic by means of marked Markovian arrival processes (MMAP); accordingly, GPRS performance measures are derived by analysis of an MMAP/G/1 queue. By evaluation of WWW traffic generated during stochastic simulation, we show that important characteristics of recent Internet traffic are depicted by Vornefeld's analytical source model. We find that the analytical approach is well suited for understanding the general impact of various system parameters. Nevertheless, although the analytical model uses a very detailed characterization of WWW traffic, for quantitative performance analysis, simulation results are still required.