The case study discussed in this chapter is used to illustrate how measurement data can be used to obtain performance parameters for a multiclass QN model. The model is used to evaluate various important scenarios, such as balancing disks, varying the concurrency level, selecting security levels, and purchasing new Web servers.
Experimentation is often used in lieu of analytic modeling. It is important to properly design experiments so that relevant factors and their levels may be used. There should be a balance between the number of options considered and the budget allocated to the experiments. Variations in the results of experiments due to errors and/or non-controlled variables are normal and expected. They are taken into account when analyzing the results of experiments.
Confidence intervals are shown to be quite useful. They are useful in determining the integrity of selected workload classes (e.g., Section 6.3). They are also useful in determining whether or not different configurations are significantly different (e.g., Section 6.7).