examined and solutions are suggested through a comparison of measures. ClfARACTERlZMlG COMPUTER PERFORIUANCE WITH A SINGLE hlUM3ER

Reducing computer performance to a single number has become one of the more controversial (and confus-ing) subjects in performance evaluation. At best, the need for a single performance number is seen as a nec-essary evil; many argue that performance is multidi-mensional and can only be accurately represented with a series of performance numbers. Despite any such ar-guments, the fact remains that single numbers will be used for performance comparisons. This being the case, it is important that the best singIe-number measure be used. Properties of good measures are proposed in this article, and commonly used measures are studied and compared to determine which have the desired proper-ties. An area where performance is of considerable inter