Overview

In the previous chapters, we covered modeling from several perspectives, ranging from simulation to queuing models to Petri nets and to operational analysis. For those perspectives, only limited amounts of data are actually measured on an actual system. Often, the simplifying assumptions that are made so that model results are calculable enable us to obtain only an approximate analysis of the system's behavior. Also, the load conditions that are presented to an analytical or simulation model often are not tested in a real-world situation. These factors have two ramifications. The first is that more detailed analysis is difficult because of the lack of adequate real-world data. The second is that, even with a detailed model, validation of the model and its results must be weak at best. The latter statement is especially true for general-purpose simulation models such as those discussed throughout this book. Before a simulation can be used to predict the performance of any system, the results of its execution must be compared against a known baseline, and the simulation must be adjusted accordingly. One method of achieving this is through the instrumentation and collection of performance data on an actual system. The results of these measurements are compared with the predicted results from a simulation model of the same system. When the results agree to within some predetermined tolerance, the model is considered validated.