Banca de DEFESA: ELISA GABRIELA MACHADO DE LUCENA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : ELISA GABRIELA MACHADO DE LUCENA
DATE: 31/07/2026
TIME: 14:00
LOCAL: Online
TITLE:

Reducing Computational Cost in Parallel Scalability Analysis through Online Accuracy-Aware Adaptive Sampling

KEY WORDS:

Parallel Applications, Performance prediction, Scalability analysis, Machine learning

PAGES: 55
BIG AREA: Ciências Exatas e da Terra
AREA: Ciência da Computação
SUBÁREA: Sistemas de Computação
SUMMARY:

The PaScal Suite is a set of tools designed to support developers in analyzing the scalability of parallel applications. However, depending on the number of evaluated configurations, the scalability analysis process may require a large number of application executions, resulting in lengthy experimentation times. To address this limitation, this work proposes a machine-learning-based approach integrated into the PaScal Suite to predict the performance of parallel applications across different machine sizes and problem sizes. The proposed method employs an Extra Trees regression model to adaptively estimate execution times using a reduced set of sampled configurations. By predicting the performance of unmeasured configurations, the approach decreases the number of required program executions while preserving the visualization capabilities provided by the PaScal Viewer. Experiments conducted with applications from the PARSEC benchmark suite demonstrate that the proposed model can reduce the total analysis time by up to 70%, while maintaining prediction accuracy through cross-validation using the sMAPE metric. The resulting integration enables developers, researchers, and students to perform scalability analyses more efficiently, reducing both computational cost and experimentation time.

COMMITTEE MEMBERS:
Presidente - 1673543 - SAMUEL XAVIER DE SOUZA
Interno - 2885532 - IVANOVITCH MEDEIROS DANTAS DA SILVA
Interno - 1153006 - LUIZ AFFONSO HENDERSON GUEDES DE OLIVEIRA
Externo à Instituição - LUCAS MELLO SCHNORR - UFRGS