Abstract

This paper is concerned with the performance tuning of the fuzzy logic controller (FLC) used for the process grain sized scheduling of parallel jobs. First, we have proposed a scheduling algorithm called agile algorithm. The performance of the agile algorithm depends on how the processes of the applications are coscheduled. The performance of the scheduling algorithm is evaluated using the features of the scheduling metrics like average waiting time, mean response time, mean reaction time, mean slowdown, turn around time and mean utilization. A rule based scheduling system is framed using the fuzzy logic controller with the help of the above mentioned metrics to schedule all the parallel workload data. The fuzzy controller helps to identify the scheduling strategy using the rule base developed and assign the corresponding scheduling class to the parallel jobs. The fuzzy logic controller uses Mamdani model for classifying the scheduling class and found that the error rate is high during the defuzzification and need to tune the fuzzy controller to reduce the error. The paper concentrates about the tuning of the fuzzy logic controller using a neural network where the rule base of a fuzzy system is interpreted as a neural network. The performance of the neural network in turn depends on the weight determination of its own network and the various optimization techniques like genetic and parallel genetic algorithm, particle swarm optimization, hybrid particle swarm optimization with the tabu search and the parallel implementation of the hybrid approach of the particle swarm optimization with the tabu search are employed to identify the weight determination of the neuro fuzzy network. The paper gives very good performance of the fuzzy controller using the neuro fuzzy system with weights identified using the above optimization techniques. The paper gives a complete analysis of the computational time occurred and the weight identification with the various optimization techniques which guides the neural network to speed up the training process.

Key words: Genetic algorithm, agile algorithm, mean reaction time, particle swarm optimization, tabu search.