Accurate Unix benchmarking: art, science, or black magic?

The author offers a simple mathematical model of the interrelation of hardware and software to illustrate the necessity of knowledge of, and control over, the system under evaluation. He presents a list of hardware and software variables in Unix systems that should be considered when running benchmarks and a set of guidelines to benchmarking technique. While Unix is used as the case study, most (if not all) of the unwritten rules generally apply to other operating system environments. The author applies these guidelines to various benchmarks to illustrate the need for proper technique.<>