Study and optimization for high performance processing with GALPHAT

• Ano de defesa: 2020
• Autor(a) principal: Igor Kolesnikov
• Orientador(a): Celso Luiz Mendes, Reinaldo Roberto Rosa
• Banca de defesa: Gilberto Ribeiro de Queiroz, Irapuan Rodrigues de Oliveira Filho
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Instituto Nacional de Pesquisas Espaciais (INPE)
• Programa de Pós-Graduação: Programa de Pós-Graduação do INPE em Computação Aplicada
• País: BR
• Link de acesso: http://urlib.net/sid.inpe.br/mtc-m21c/2020/04.20.16.10

The parametric computational modeling of galaxies is a process with a high computational cost. The statistical component of modeling, which may involve model refinements in relation to the source brightness distribution achieves more satisfactory results when the approach is Bayesian. In this research, we are using GALaxy PHotometric ATtributes (GALPHAT) as our main tool for data processing. The GALPHAT modeling of a galaxy observed by the Sloan Digital Sky Survey (SDSS) can last about 6 hours. In the current scenario of cosmology, this type of modeling, to be scientifically effective, must be performed on a set containing about thousands of objects. The sample analyzed within the scope of the FAPESP thematic project that LABAC participates contains more than 24,309 objects, an amount that demands the use of high-performance computing (HPC) to enable effective modeling of the entire sample. In this postgraduate project, we have as the main objective to study and optimize HPC solutions that allow GALPHAT processing on a SDSS sample in the fastest possible way. For this, we have two HPC systems that can work in a coordinated way to optimize the modeling strategies. The first system belongs to LABAC and is based on Intel Xeon Phi 7250 platform. The second system belongs to the partition of the multi-core platform of the Santos Dumont supercomputer. The research, therefore, includes the initial process done to set up and run GALPHAT on both platforms, thus using different types of processors and compilers. Considering the different processing steps, in different modeling strategies we applied refactoring and complete modules rewriting. Our studies have found the optimal combination of software, hardware and optimizations to minimize processing time. This is the first step in implementing and integrating the graphical user interface to make GALPHAT easier to use. This dissertation, therefore, presents all of the activities that were performed to allow, as a final result, to process, in a timely manner, via HPC, the entire selected sample including the description of benchmark among the computational systems used. It includes the development of the auxiliary visualization system as well.