Banca de DEFESA: REILTA CHRISTINE DANTAS MAIA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : REILTA CHRISTINE DANTAS MAIA
DATE: 13/02/2026
TIME: 10:00
LOCAL: meet.google.com/gdq-sixm-wze
TITLE:

Accelerating Scalability Analysis in the PaScal Analyzer with the Paramount Iteration Approach

KEY WORDS:

Scalability Analysis, Performance Estimation, Paramount Iteration, PaScal Suite, High-Performance Computing

PAGES: 68
BIG AREA: Ciências Exatas e da Terra
AREA: Ciência da Computação
SUMMARY:

Scalability analysis of High-Performance Computing (HPC) applications imposes significant challenges, demanding excessive time and high computational resource consumption due to the need for full executions across multiple hardware configurations. To mitigate this problem, this work presents PaScal Paramount, a new feature integrated into the PaScal Analyzer tool designed to accelerate the data collection required for this analysis. The approach is based on the Paramount Iteration technique, which partially executes the application’s iterative loops to estimate its global behavior. Unlike preliminary approaches, the implemented methodology utilizes the Trimmed Mean of the initial iterations to project the total time, filtering out warm-up noise and system outliers to ensure statistical robustness. The experimental validation covered three distinct classes of applications: regular load (Matrix Multiplication), irregular load (Mandelbrot), and strictly computational (NAS EP Proxy). Results demonstrated that the tool can reduce analysis time by up to 89%, preserving the fidelity of visual efficiency and scalability patterns obtained from full executions. Furthermore, the study identified a trade-off point in the sampling rate that ensures estimation accuracy even in irregular load scenarios. Therefore, the proposed solution offers a flexible and agile alternative, making the scalability analysis cycle more practical, accessible, and fast.

COMMITTEE MEMBERS:
Interno - 1153006 - LUIZ AFFONSO HENDERSON GUEDES DE OLIVEIRA
Interno - 1543191 - LUIZ FELIPE DE QUEIROZ SILVEIRA
Interno - 1673543 - SAMUEL XAVIER DE SOUZA
Externo à Instituição - LUCAS MELLO SCHNORR - UFRGS