In the thesis, a critical comparison of two different codes based on the Spectral Element Method is carried on, providing results in terms of performance, compatibility and scalability. In particular, these two programs are adapted to simulate seismic wave propagation on a large scale for HPC, using Message Passing Interface and working on parallel computing architectures. A detailed analysis of the numerical tools used, of the structure of each program and of the methods used to compare results is provided. Strategies for passing from one code to the other are defined, highlighting common points and differences, focusing on the description of input and output data structure. Partial times of performance are analyzed, in order to spot faster and slower processes for future optimizations. Scalability curves are obtained for the same case, increasing the number of processors used on the cluster; after a confirmation of the reliability and consistency of calculations, scalability curves are compared and results are discussed. These analyses are performed on a benchmark case, called Croissant Valley and on a real case, the 2012 Emilia earthquake. In order to provide new solid bases for further development of the two codes, starting from analyzing evidences on soil behavior under cyclic loads, the Muravskii model, based on a modification of the well known Masing's rules, is described and implemented on Matlab. Numerical strategies applied and the flowchart are presented, providing a description of the main issues faced and their solutions.
La tesi tratta la comparazione critica di due codici differenti basati sul Metodo degli Elementi Spettrali (SEM), fornendo risultati in termini di prestazioni, di compatibilità e scalabilità. In particolare, questi due codici sono adattati per simulare la propagazione di onde sismiche su larga scala per HPC, utilizzando la tecnologia MPI per computer ad architettura parallela. Si presenta un'analisi in dettaglio degli strumenti numerici utilizzati, della struttura dei programmi e dei metodi per compararne i risultati. Si definiscono le strategie per passare dall'utilizzo di un codice all'altro, evidenziando punti comuni e differenze, focalizzandosi sulla descrizione della struttura dei dati di input e output. Si analizzano i tempi parziali delle prestazioni, per trovare i processi ottimizzabili. Si ottengono curve di scalabilità al crescere del numero di processori e, una volta confermata l'affidabilità e la consistenza dei calcoli, si discutono i risultati. Le analisi sono svolte sui casi Croissant Valley e sul terremoto dell'Emilia (2012). Per fornire nuove basi per lo sviluppo futuro dei codici, a partire dalle evidenze sperimentali del comportamento del terreno sotto carichi ciclici, si descrive il modello di Muravskii, basato sulle regole di Masing e la sua implementazione su Matlab.
Critical comparison between HPC codes for non-linear seismic wave propagation : performance, compatibility and scalability
MAMBRINI, ALESSANDRO
2016/2017
Abstract
In the thesis, a critical comparison of two different codes based on the Spectral Element Method is carried on, providing results in terms of performance, compatibility and scalability. In particular, these two programs are adapted to simulate seismic wave propagation on a large scale for HPC, using Message Passing Interface and working on parallel computing architectures. A detailed analysis of the numerical tools used, of the structure of each program and of the methods used to compare results is provided. Strategies for passing from one code to the other are defined, highlighting common points and differences, focusing on the description of input and output data structure. Partial times of performance are analyzed, in order to spot faster and slower processes for future optimizations. Scalability curves are obtained for the same case, increasing the number of processors used on the cluster; after a confirmation of the reliability and consistency of calculations, scalability curves are compared and results are discussed. These analyses are performed on a benchmark case, called Croissant Valley and on a real case, the 2012 Emilia earthquake. In order to provide new solid bases for further development of the two codes, starting from analyzing evidences on soil behavior under cyclic loads, the Muravskii model, based on a modification of the well known Masing's rules, is described and implemented on Matlab. Numerical strategies applied and the flowchart are presented, providing a description of the main issues faced and their solutions.File | Dimensione | Formato | |
---|---|---|---|
masterA.pdf
solo utenti autorizzati dal 15/04/2019
Descrizione: nuovo
Dimensione
24.11 MB
Formato
Adobe PDF
|
24.11 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/10589/140277