In this thesis are presented some of the results obtained as contribution to two different projects. The fi rst one is the REDSEA project, an EU founded project whose goal is to develop exascale network interconnect, described in more detail in section 1.1. The contribution to this project is, in turn, two-fold: First are presented the results related to the collection and analysis of network traces for the applications contributed by eXact lab. This activity is aimed at suggesting critical requirement for the network characteristics, in a co-design approach. The traces were collected for the LAMMPS molecular dynamics package and the results are presented and discussed in chapter 2. The other contribution to the REDSEA project is the porting of applications to the project speci c architecture. In particular, a parallel version of the Self Organizing Maps (SOM) algorithm has been developed. The goal was to develop a massively-parallel implementation of this algo- rithm, based on MPI and optimized for the RED-SEA architecture. The implemented code, available on github, has also full support for OpenSH-MEM parallelization, with the main goal being the ability to run on GSAS, a Partioned Global Address Space (PGAS) environment developed by FORTH. The results related to these activities, along with a comparison between the two implementations, are presented and discussed in chapter 3. The activities related to the second project presented in this thesis are dis- cussed in chapter 4. The main task was to extend the SOFA framework (via a plugin) to allow harnessing the ray-tracing cores of modern GPUs to perform collision detec- tion. The SOFA framework is a powerful framework used for real-time FEM based simulations, with applications, among many others, in surgery and engineer- ing. The developed collision detection pipeline employes the NVIDIA OptiX ray- tracing Engine that enabled us to exploit dedicated, and very powerful, hard- ware to perform collision detection, a very relevant task for many scienti c and industrial applications.
Exploring innovative approaches in industrial and academic HPC applications(2022 Dec 20).
Exploring innovative approaches in industrial and academic HPC applications
-
2022-12-20
Abstract
In this thesis are presented some of the results obtained as contribution to two different projects. The fi rst one is the REDSEA project, an EU founded project whose goal is to develop exascale network interconnect, described in more detail in section 1.1. The contribution to this project is, in turn, two-fold: First are presented the results related to the collection and analysis of network traces for the applications contributed by eXact lab. This activity is aimed at suggesting critical requirement for the network characteristics, in a co-design approach. The traces were collected for the LAMMPS molecular dynamics package and the results are presented and discussed in chapter 2. The other contribution to the REDSEA project is the porting of applications to the project speci c architecture. In particular, a parallel version of the Self Organizing Maps (SOM) algorithm has been developed. The goal was to develop a massively-parallel implementation of this algo- rithm, based on MPI and optimized for the RED-SEA architecture. The implemented code, available on github, has also full support for OpenSH-MEM parallelization, with the main goal being the ability to run on GSAS, a Partioned Global Address Space (PGAS) environment developed by FORTH. The results related to these activities, along with a comparison between the two implementations, are presented and discussed in chapter 3. The activities related to the second project presented in this thesis are dis- cussed in chapter 4. The main task was to extend the SOFA framework (via a plugin) to allow harnessing the ray-tracing cores of modern GPUs to perform collision detec- tion. The SOFA framework is a powerful framework used for real-time FEM based simulations, with applications, among many others, in surgery and engineer- ing. The developed collision detection pipeline employes the NVIDIA OptiX ray- tracing Engine that enabled us to exploit dedicated, and very powerful, hard- ware to perform collision detection, a very relevant task for many scienti c and industrial applications.File | Dimensione | Formato | |
---|---|---|---|
Matteo Barnaba_Thesis.pdf
accesso aperto
Descrizione: MHPC thesis
Tipologia:
Tesi
Licenza:
Non specificato
Dimensione
1.62 MB
Formato
Adobe PDF
|
1.62 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.