Neural network versus max-flow algorithms for multiprocessor real-time scheduling

Neural networks have been widely used in a large area of applications, like image processing, learning processes, identification and control, etc. but there is a lack for their use for approximate solving real-time scheduling problems. The authors have already shown the ability of a neural network based scheduling algorithm to deal with the scheduling of independent real-time tasks in a multiprocessor environment. The algorithm is approximate but has a remarkable convergence speed due to the highly parallel nature of the search. In recent literature, the authors have analyzed the performance of the algorithm when compared with the well-known rare monotonic and earliest deadline algorithms for the monoprocessor case. In this paper we present an analysis of the quality of the yielded solution for the multiprocessor case.