Solve Multiprocessor Real-Time Scheduling Using Competitive Slack Neural Networks

Abstract

Neural network using competitive learning rule provides a highly effective method of attaining a sound solution and is capable of simplifying the network complexity. Intrinsically, the competitive scheme is used to solve fully utilized real-time scheduling problem. This study extends the competitive Hopfield neural network to solve both non-fully and fully utilized multiprocessor real-time scheduling problems with execution time and deadline constraints. Extra “slack neurons” added on to the neural network topology to represent pseudo job were applied to ease the non-fully utilized situation and facilitate solving the problem. Simulation results confirm that the competitive neural network imposed on the proposed energy function corresponding neural networks with slack neurons integrated ensures an appropriate approach of solving this class of real-time scheduling problems of single processor or multiprocessor system.