Comparison of simulation and measurement using state of the art Web traffic models

This paper identifies and presents relevant differences between network measurement and simulation for testing network mechanisms. The intention is to uncover a broad spectrum of differences by using a realistic and dynamic traffic load. For this task, a recent web traffic model is implemented in the ns-2 simulation and as a traffic generator application for FreeBSD. Before performing the actual comparison TCP implementations in measurement and simulation have been analyzed in detail for differences, which were removed for the comparison, as far as possible. Our investigations show that measurement and simulation exhibit negligible performance differences concerning 1st order statistics like mean arrival rate, and noticeable differences in high load situations concerning metrics like the average response time of web objects. We identify the reasons behind these differences and propose improvements to make simulation behave closer to reality.