Evolutionary Computation Tuned by Fuzzy Logic for the Assignment of Hard Real-Time Tasks in MIMD NUMA Parallel Architectures

E. Ferro, J. Santos, R. Cayssials, and J. Orozco (Argentina)


Genetic Algorithms, Fuzzy Logic, Multitask Assignment, MIMD NUMA Architectures .


One of the main problems of MIMD NUMA parallel architectures is the mapping of the set of tasks to be executed over the set of processors. The problem is compounded if the tasks are real-time and not only the obvious time constraint, but also allocation, precedence, resources and communications constraints are considered. A method based on a genetic algorithm is presented to address the assignment problem. The coefficients of the cost function (inverse of the aptitude function) are tuned by fuzzy algebra. The tuned method outperforms the pure genetic. The evaluation is carried out over a randomly generated set of problems. The metric used is the Success Ratio.

Important Links:

Go Back