This thesis presents an analysis and comparison of different numerical strategies for solving sparse linear systems arising from finite element structural dynamics problems. In particular, the work focuses on the evolution of a legacy LU-based solver using Skyline storage and compares it with modern approaches based on multifrontal (MUMPS) and supernodal (Intel MKL PARDISO) methods. The study evaluates computational performance, cache behavior, memory consumption, and dynamic allocation patterns using profiling tools. The results show that the adoption of optimized libraries targeting modern x86_64 architectures leads to a significant reduction in execution time, up to approximately 70% compared to the original implementation
uesta tesi analizza e confronta diverse strategie di risoluzione numerica per sistemi lineari sparsi derivanti da problemi di dinamica strutturale agli elementi finiti. In particolare, viene studiata l’evoluzione di un solver legacy basato su decomposizione LU con storage Skyline, confrontandolo con solutori moderni basati su tecniche multifrontal (MUMPS) e supernodali (Intel MKL PARDISO). L’analisi si concentra su prestazioni in termini di tempo di calcolo, comportamento della cache, consumo di memoria e numero di allocazioni dinamiche, ottenuti tramite strumenti di profiling. I risultati evidenziano come l’adozione di librerie ottimizzate per architetture moderne x86_64 consenta una riduzione significativa dei tempi di calcolo, fino a circa il 70% rispetto all’implementazione originale.
Performance optmization of a legacy FEM solver
COLOMBO, ALESSANDRO
2025/2026
Abstract
This thesis presents an analysis and comparison of different numerical strategies for solving sparse linear systems arising from finite element structural dynamics problems. In particular, the work focuses on the evolution of a legacy LU-based solver using Skyline storage and compares it with modern approaches based on multifrontal (MUMPS) and supernodal (Intel MKL PARDISO) methods. The study evaluates computational performance, cache behavior, memory consumption, and dynamic allocation patterns using profiling tools. The results show that the adoption of optimized libraries targeting modern x86_64 architectures leads to a significant reduction in execution time, up to approximately 70% compared to the original implementation| File | Dimensione | Formato | |
|---|---|---|---|
|
Performance_Optimization_of_a_Legacy_FEM_Solver.pdf
accessibile in internet per tutti
Dimensione
4.41 MB
Formato
Adobe PDF
|
4.41 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/261398