Banca de QUALIFICAÇÃO: HIAGO MAYK GOMES DE ARAÚJO ROCHA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
DISCENTE : HIAGO MAYK GOMES DE ARAÚJO ROCHA
DATA : 01/08/2018
HORA: 14:00
LOCAL: Auditório I
TÍTULO:

Mapping and Routing Problem: a math-heuristic optimization process


PALAVRAS-CHAVES:

Multiprocessor System-on-Chip, Network-on-Chip, Performance, Task Mapping, Routing Algorithm, Math-Heuristic.


PÁGINAS: 60
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
SUBÁREA: Sistemas de Computação
ESPECIALIDADE: Arquitetura de Sistemas de Computação
RESUMO:

The advances in integrated circuit manufacturing technology, enabled by the reduction in transistor's size, allowed the creation of complex, multiprocessed systems in a single chip, called Multiprocessor Systems on Chip (MPSoCs). Currently, one of the main communication models between the cores of these systems are the Networks on Chip (NoCs). MPSoCs based on NoCs seek to meet the demand and complexity of current applications, which require more processing power. Therefore, the study of optimization methods in MPSoCs becomes more relevant given its wide use in the design of modern embedded systems as well as supercomputers. Since they are part of  current embedded systems design, MPSoCs are subject to some restrictions, such as power dissipation and energy consumption. In addition to application's performance enhancenamen, which can be measured by the communication latency between the cores and the total execution time of the applications. The challenges mentioned above are related to the way in which the tasks of the applications are mapped into the cores of the MPSoC and how the task routing occurs. These problems are called the Mapping Problem (PM) and the Routing Problem (PR), respectively, and are extensively studied in the literature on the optimization of NoC-based MPSoCs.

Given the importance of the task mapping and routing and the strong relationship between these two problems, this work presents the formulation, of the entire mathematical programming, of the Problem called Mapping and Routing Problem (PMR). The PMR consists in minimizing the cost of allocating the tasks of an application in the cores of an MPSoC, considering the communication between the tasks. As well as the possible routes taken by the packets between each pair of tasks mapped into the cores. The proposed model considers homogeneous MPSoCs based on NoCs and uses a constraint that considers the bandwidth of NoC channels. The main goal of the model is to search for mapping solutions and the construction of optimized deadlock-free routing with low latency. In addition, it is proposed a math-heuristic optimization procedure for the PMR considering the context of dynamic task mapping. Math-heuristics is a problem-solving strategy that applies interoperability between metaheuristic or heuristic procedures and the mathematical model of the problem in order to find good solutions. To the best of our knowledge, no work has been found in the optimization literature for NoC-based MPSoCs that addresses the use of this strategy. As a result, it is expected to have an optimization proposal for dynamic task mapping that, when compared with other literature proposals, presents better latency results.


MEMBROS DA BANCA:
Presidente - 1882699 - MONICA MAGALHAES PEREIRA
Interno - 1694485 - MARCIO EDUARDO KREUTZ
Externo ao Programa - 2859606 - SILVIA MARIA DINIZ MONTEIRO MAIA
Externo à Instituição - ANTONIO CARLOS SCHNEIDER BECK FILHO - UFRGS
Notícia cadastrada em: 27/07/2018 07:47
SIGAA | Superintendência de Informática - (84) 3215-3148 | Copyright © 2006-2018 - UFRN - sigaa07-producao.info.ufrn.br.sigaa07-producao