The art of performance tuning is, alas, still an art; there are few tools to help predict the effects of changes that are motivated by performance needs. In this work, we present dynamic dependency modeling techniques for predicting the effects of configuration changes. These techniques utilize a simple, exterior model of the system under test, and only require codifying the nature of dependencies between subsystems. With a few examples, we demonstrate how to utilize these models to predict the impact of system changes. Using simple tools and a reasoning process, it is possible to answer many questions about possible performance enhancements. The impact of this work is to advance performance tuning a small amount
Marc Chiarini, Alva L. Couch