The principal goal of this research planning project is to prescribe a performance evaluation environment that supports the effective, efficient, and creative programming of parallel computer systems. Since the complexity of parallel computer systems makes program performance evaluation and prediction difficult, tools employing the experimental method to observe, analyze, and display system activity are essential. Accurate insights about program execution, particularly on future large systems, are possible only if innovative techniques are used in the areas of instrumentation, data management, visualization, and knowledge engineering. Within this context, this project will study the relationships between properties of parallel programs and properties of the performance representations produced during execution on scalable multicomputer systems. The validity and utility of particular views and statistics as indicators and predictors of program performance are to be investigated.
