| The potential performance benefits of a parallel/distributed software application are not always easy to achieve. One primary reason for this difficulty is the lack of a comprehensive performance model which incorporates the software structure, expected workload and runtime environment of the application. To overcome this obstacle, a hierarchical methodology is described to provide a framework for analytic performance modeling. The methodology includes methods for linking different types of analytic models, processing uncertain information, evaluating sensitivity and controlling the complexity of performance equations. The methodology is intended to complement the analysis/design process, and can be applied throughout the software life-cycle. The results of the research provide tools and techniques which can help designers fulfill their performance goals. |
