On the validity of assumptions used to model local area networks

The accuracy of analytical and simulation models of local area networks (LANs) is dependent on underlying assumptions about traffic characteristics, and protocols. The authors present a comparison of some of these assumptions about LAN behavior against empirical data on the actual behavior of several LANs. Data concerning the packet size distribution, interarrival distribution, and traffic pattern are presented graphically, showing that the actual packet size distribution and traffic pattern do not follow the simplified distributions assumed in the models. However, the packet interarrival time distribution somewhat resembles the assumed exponential distribution. The accuracy of analytical models and simulations that rely on assumptions about the packet arrival process is assessed by comparing the results of simulations using the assumed distributions with simulations driven by the trace of an actual arrival process. Performance measures are compared and the difference is tested for statistical significance. Techniques for generating realistic loads that are higher than those observed, yet are based on actual observations, are presented and evaluated.<
>