Predicting the impact of scheduling modifications on system performance: case study

Abstract

A measurement-based model is used to conduct a detailed evaluation of scheduling policies of the Alliant FX/80. The model is also used to evaluate the real workload performance impact of various processor configurations. The model is constructed from measurements obtained during normal machine operation. It is capable of predicting the completion time of a given application executing under real workloads. The evaluation of scheduling policies presented demonstrates the flexibility and power of the modeling methodology. It is shown that the model is not limited to single-point evaluations of system changes. The model has the ability to investigate worst case behavior, as well as estimate the probability that an application will finish by a given deadline. Results from empirical studies which validate the model are also presented.<>