Non-intrusive Monitoring Framework for NoC-based Many-Cores

  • Angelo Elias Dalzotto Pontifícia Universidade Católica do Rio Grande do Sul
  • Caroline da Silva Borges Pontifícia Universidade Católica do Rio Grande do Sul
  • Marcelo Ruaro Pontifícia Universidade Católica do Rio Grande do Sul
  • Fernando Gehm Moraes Pontifícia Universidade Católica do Rio Grande do Sul

Resumo


Many-core Systems on Chip (MCSoCs) require resource management to achieve scalability at the computation and communication levels. The monitoring infrastructure feeds management tasks with raw data, enabling these tasks to detect behaviors corresponding to constraint violations or a trend that signalizes a future violation. Several works available in the literature use monitoring to apply their management techniques but do not specify how to implement the monitoring framework. We propose a monitoring framework for MCSoCs, with the following features: (i) generic: the infrastructure can carry data related to different monitored features; (ii) monitored data does not disturb NoC flows; and (iii) reduced overhead compared to other monitoring methods. The monitoring framework is loosely coupled to the MCSoC by using a dedicated NoC to carry monitoring and management messages, decoupling data traffic from management traffic. Results adopt the Observe-Decide-Act management method, comparing the proposed monitoring framework to a standard monitoring approach. Results show a reduction in the data NoC traffic (12%), faster management responsiveness to act on deadline violations (up to 77%), and reduced applications execution time (on average 8%).
Palavras-chave: monitoring, management, self-adaptability, multiple physical networks (MPN), NoC-based many cores
Publicado
21/11/2022
Como Citar

Selecione um Formato
DALZOTTO, Angelo Elias; BORGES, Caroline da Silva; RUARO, Marcelo; MORAES, Fernando Gehm. Non-intrusive Monitoring Framework for NoC-based Many-Cores. In: SIMPÓSIO BRASILEIRO DE ENGENHARIA DE SISTEMAS COMPUTACIONAIS (SBESC), 12. , 2022, Fortaleza/CE. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2022 . p. 93-99. ISSN 2237-5430.