Uma biblioteca para auxiliar o desenvolvimento de aplicações que processam grandes matrizes na memória externa

• Ano de defesa: 2014
• Autor(a) principal: Silveira, Jaqueline Alvarenga
• Orientador(a): Andrade, Marcus Vinícius Alvim
• Banca de defesa: Ribeiro, Carlos Antônio Alvares Soares , Ferreira, Ricardo dos Santos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Viçosa
• Programa de Pós-Graduação: Mestrado em Ciência da Nutrição
• Departamento: Valor nutricional de alimentos e de dietas; Nutrição nas enfermidades agudas e crônicas não transmis
• País: BR
• Palavras-chave em Português: Processamento de dados; Algoritmos; Memória cache; Sistemas de informação
• Palavras-chave em Inglês: Data processing; Algorithms; Cache memory; Information systems
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://locus.ufv.br/handle/123456789/2798

This Work presents a new library, named TiledMatrix, to support the development of applications that process large matrices stored in external memory. The library is based on some strategies similar to cache memory management and its basic purpose is to allow that an application, designed for internal memory processing, can be adapted for external memory. It provides an interface for external memory access that is similar to the traditional method to access a matrix Moreover, it divides the matrix into tWo-dimensional blocks, Where the blocks are organized in the disk to take advantage of access pattern of applications. In order to reduce the transfer time of a block to and from the disk, TiledMatrix uses an algorithm for compression/decompression of data that Was implemented in parallel to further improve the efñciency of the library. The TiledMatrix Was implemented and tested in some applications that require intensive matrix processing such asz computing the transposed matrix and the computation of Viewshed, ñow accumulation and median. These applications Were implemented in two Versionsz one using TiledMatrix and another one using the Segment library that is included in GRASS, an open source GIS. In average, they Were 7 times faster With TiledMatrix and, in some cases, more than 18 times faster. The TiledMatrix performance Was also compared With the ArcGLSÍ For these tests, it Was considered only two applications that are included in ArcGIS: the Uiewshed and the ñow accumulation. As tests have shown, in some cases, the applications Were more than 200 times faster using the TiledMatrix.