Particionador paralelo de grafos utilizando algoritmos heurísticos para aplicação em simuladores paralelos de reservatórios de petróleo

• Autor(a) principal: Silva, Leonardo Rogério Binda da
• Data de Publicação: 2014
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Espírito Santo (riUfes)
• Texto Completo: http://repositorio.ufes.br/handle/10/5337
• topic: Petroleum engineering; Graphs partitioning; Heuristics; Parallel computing; Reservoir simulation; Clusters of Computer; Particionamento de grafos; Computação paralela; Simulação de reservatórios; Clusters de computadores; Engenharia/Tecnologia/Gestão; Engenharia de petróleo; Representações dos grafos; Heurística; Programação paralela (Computação); 620.9

Oil is currently the most widely fuel used in the world. To obtain it to the greatest possible economic viability is a relentless pursuit of the producing companies. In this scenario, the numerical reservoir simulation using parallel computers with distributed memory (clusters) is emerging as an important tool. These application handle mashes of discrete points that represent the field of oil reservoir. An important step of the simulation using clusters is the partitioning of this mesh points so that each cluster process node can perform its calculations on a portion of this mesh. The domain meshes can be represented by graphs. Partitioning meshes then becomes a problem of graph partitioning. If the graph vertices number that represents the mesh is very high, serial partitioners can have performance problems. Graph partitioners using clusters appear as interesting alternatives in this situation, minimizing the time spent in partitioning. This research deals with the implementation of a parallel graph partitioner to be used in clusters based on partitioning heurists proposed and implemented serially by Bonatto (2010). The parallel partitioner has been developed using the Java programming language and MPJ Express messages passing library. Efficient abstract data types have been proposed and implemented in order to optimize the performance. The parallel graph partitioner performed the cutting of different graphs, obtaining, most of the time, smaller cuts than the ones found by serial partitioner of Bonatto (2010) and by programs such as METIS and CHACO. Improvements to the Bonatto (2010) serial partitioner have been proposed. Analysis of speedup and parallel efficiency have been performed to find out the gains of times abtained with the parallelization of the heuristics.