Heuristics for Mapping Real-Time Applications to NoC-Based Architectures Using Genetic Algorithms

• Iae Santos Bonilha UFSM
• Leandro Indrusiak University of York
• Osmar Marchi dos Santos University of York

Resumo

One crucial step in the development of an application for a NoC-based MPSoC, consists on the allocation of the tasks that compose the system. When we consider such architecture, besides the computation time of the tasks, we also have to take into account the communication time between tasks in order to allocate and guarantee that tasks do not miss their deadlines. Communication between tasks is a factor that impacts directly in the real time system temporal behavior. However, its combination with task execution is not fully taken in consideration in NoC-based architectures. In this paper we focus on the discussion and development of different heuristics that are used for task mapping with the use of genetic algorithms. We use different response-time equations in order to provide a feasibility analysis that works for both task execution and communication. Our evaluation of the algorithms show that certain heuristics can provide quite an improvement on the processor utilization.

Palavras-chave: Sociology, Statistics, Mathematical model, Computer architecture, Genetic algorithms, Real-time systems, Biological cells, Multi-core, Network-On-Chip