Since the problem of scheduling multiprocessor tasks with deadlines is known as a hard combinatorial optimization one, an approximation or heuristic algorithm has been studied. A Boltzmann machine is such an approximation method based on neural networ...
Since the problem of scheduling multiprocessor tasks with deadlines is known as a hard combinatorial optimization one, an approximation or heuristic algorithm has been studied. A Boltzmann machine is such an approximation method based on neural network that can be executed in highly parallel. In this paper, we propose a mapping scheme of the scheduling problem onto a Boltzmann machine and a Boltzmann machine-based scheduling method for multiprocessor tasks with deadlines. It is proven that maximization of the consensus function of the Boltzmann machine corresponds to finding a schedule with the maximum number of tasks that can be completed without violating their deadlines. It is argued by simulations that the Boltzmann machine configured with this mapping scheme outperforms the traditional heuristic scheduling algorithms.