COMPUTER PERFORMANCE PREDICTION VIA ANALYTICAL MODELING - AN EXPERIMENT.

One of the primary goals of computer system modeling is performance prediction. A model of the current system is constructed and validated. The model is then altered to reflect a proposed change to the system. The performance predictions of the altered model are taken as indicators of actual future performance of the changed system. This study describes, documents, and analyzes a series of experiments to test the above scenario. The results point out several problems that commonly arise while constructing predictive models. Solutions to these problems demand a critical review of all aspects of predictive modeling.