| Abstract: | A hybrid analytic and experimental approach to the analysis of large systems is described. The approach is iterative under the assumption that a correct analysis will not be made first time. It is primarily aimed at the analysis of the performance effect of proposed changes prior to coding them. The requirement is for a fairly approximate analysis that is easy to make. The approach is illustrated by an experiment on a relational database system but we believe it to be applicable to many kinds of large systems. The different reasons for analysing performance of a system are discussed and then the ways of approaching the analysis, with an outline of our approach. The details of our approach are given, and illustrated by our experiment. |
