Detalhes bibliográficos
| Ano de defesa: |
2012 |
| Autor(a) principal: |
Antunes, Eduardo de Brum
 |
| Orientador(a): |
Rose, César Augusto Fonticielha de
|
| Banca de defesa: |
Não Informado pela instituição |
| Tipo de documento: |
Dissertação
|
| Tipo de acesso: |
Acesso aberto |
| Idioma: |
por |
| Instituição de defesa: |
Pontifícia Universidade Católica do Rio Grande do Sul
|
| Programa de Pós-Graduação: |
Programa de Pós-Graduação em Ciência da Computação
|
| Departamento: |
Faculdade de Informáca
|
| País: |
BR
|
| Palavras-chave em Português: |
|
| Área do conhecimento CNPq: |
|
| Link de acesso: |
http://tede2.pucrs.br/tede2/handle/tede/5159
|
Resumo: |
The increasing complexity of the applications demands more processing capacity, which boosts the development of a computational system composed of modules, such as processors, memories and specific hardware cores, called Multi-Processor System-on- Chip (MPSoC). If the modules of this system are connected through a Network-on-Chip (NoC) communication infrastructure and all processors are of the same type, they are known by homogeneous NoC based MPSoC. One of the main problems relating to MPSoCs design is the definition of which processors of the system will be responsible for each application task execution, objecting to meet the design requirements, such as the energy consumption minimization and the application execution time reduction. This work aims to carry out quickly and efficiently partitioning and mapping activities for the design of homogeneous MPSoCs. More specifically, the partitioning application's task into groups, and mapping of tasks or task groups into a target architecture type homogeneous NoC-based MPSoC. These activities are guided by requirements of energy consumption minimization and load balancing, and delimited by constraints of maximum energy consumption, maximum processing load and maxima areas of data and code of each processor. The work shows the complexity of partitioning and mapping activities separately and jointly. It also shows that the mapping is more efficient on energy consumption minimization, when compared to partitioning, yet the effect of partitioning cannot be neglected. Moreover, the joint effect of both activities saves in average 37% of energy. The mapping when performed at runtime may be inefficient, due to the short time and the large number of solutions to be explored. Having an approach that applies a static partition before the dynamic mapping, it is possible to achieve more efficient mappings |
