Study of Approaches for Reducing the Evaluation Time of Parallel Program Performance

• Elisa Gabriela Machado de Lucena UFRN
• Samuel Xavier-de-Souza UFRN

DOI: https://doi.org/10.5753/erad-ne.2025.11739

Abstract

Scalability analysis in parallel applications can be costly due to the need for multiple executions. This work proposes a comparison between an analytical modeling and a neural network (MLP) approach to estimate execution time. The goal is to evaluate the trade-off between computational cost, accuracy, and generalization. The expected outcome is to enhance the Pascal Analyzer by integrating predictive techniques that speed up performance analysis.

References

da Silva, A. B., Cunha, D. A., Silva, V. R., de A. Furtunato, A. F., and Xavier-de Souza, S. (2017). Pascal viewer: A tool for the visualization of parallel scalability trends. In International Workshop on Extreme-Scale Programming Tools, pages 250–264. Springer.

Furtunato, A. F. A., Georgiou, K., Eder, K., and Xavier-De-Souza, S. (2020). When parallel speedups hit the memory wall. IEEE Access, 8:79225–79238.

Rodrigues, M. d. S. L. (2025). Estimação adaptativa baseada em acurácia para tempo de execução de aplicações paralelas. B.S. thesis, Universidade Federal do Rio Grande do Norte.

Sze, V., Chen, Y.-H., Yang, T.-J., and Emer, J. S. (2017). Efficient processing of deep neural networks: A tutorial and survey. Proceedings of the IEEE, 105(12):2295–2329.