Metric and workload effects on computer systems evaluation

While a system's performance is a function of its design and implementation, the workload researchers subject the system to and the metrics they use also can affect performance evaluation results. There are two main approaches to performance evaluation: Analysis involves simplifications in the interest of mathematical tractability, and simulation offers a more realistic alternative that directly uses real workload recordings. The problems the author describes appear more frequently during simulation because of its ability to directly reflect complex situations, even those the evaluator does not know about or understand.