NPAC Technical Report SCCS-479

An Interpretive Framework for Application Performance Prediction

M Parashar, S Hariri, T Haupt, G Fox

Submitted April 21 1993


Abstract

The last few decades have seen an impressive developments in every aspect of parallel computing technology; viz. processing and storage technology, interconnect technology and software technology. Although these systems incorporate large amount of computing power, they are not general enough to efficiently support today's computation-intensive problems (e.g., the Grand Challenges), that warrant multiple computational models and levels of parallelism. We believe that the future of parallel computing lies in the integration of the plethora of "specialized" architectures into a single Heterogeneous High Performance Computing (HHPC) environment that allows them to cooperate in solving complex problems. Software development in any Parallel/Distributed environment is a non-trivial process and requires a thorough understanding of the application and the architecture. Evaluation tools form a critical part of any software development environment. These tools enable the developer to visualize the effect of various design choices on the performance of the application, to study scalability of the application with system and problem size and to investigate the effects of changes in system run-time status and its configuration on the application execution. The objective of t his paper is to propose an interpretive model for a source driven performance prediction framework which can meet challenges presented by an HHPC environment. The model provides a comprehensive characterization methodology to abstract and parametrize the behavior of the application and the computing environment. Interpretative techniques are then used to predict the performance of the abstracted application on the abstracted computing environment. A prototype performance prediction framework has been developed for the iPSC/860 using the proposed interpretative model. Numerical results obtained on this system are presented. These results confirm the potential of interpretative performance prediction techniques and their applicability to an HHPC environment.


PostScript version of the paper