Um estudo sobre modelos de computação paralela

AUTOR(ES)
DATA DE PUBLICAÇÃO

1995

RESUMO

Models of Computation are one of most important tools in algorithm design With these models, the work of an algorithm designer becomes easier, because these models leave out many characteristics of real machines. In parallel computing there is a great need for a general model, because we have many different parallel machines. The advent of a parallel computing model could make the area grow more than it is already growing. In this dissertation we study some parallel computing models. First we take a look at a representative set of parallel machines, in order to learn the differences between each architecture. Our set of machines contains some of the most important commercial machines such as: CM-2, Sequent Symmetry, MasPar MP-l and CM-5. After this, we study the models themselves. The models chosen were: PRAM, BSP [Va190] and LogP [CKP+93]. Many researchers argue that the PRAM is the best model for algorithm design although it is not realistic. The proposal of the BSP mo dei is bold, since it also seeks to influence parallel architecture design The proposal of LogP model although similar to the BSP, does not require parallel machines to have synchronization mechanisms. This makes LogP the most realistic but also the most difficult model to use. We evaluate these models based on the problems of Fourier Transform and Gaussian Elimination. After this study we made an evaluation of the three models

ASSUNTO(S)

programação paralela (computação)

ACESSO AO ARTIGO

Documentos Relacionados