Machine Learning support Energy Aware Efficient Multiprocessors Mapping of Real Time tasks

Using processor which supported a Dynamic Voltage Scaling (DVS), can lower power consumption by scaling down the processor' frequency while all task sets still don't miss their deadline. Recent usefulness in machine learning techniques inclusive artificial neural networks has led to the evolution of robust and feasible, prediction models for a variety of fields. This paper immediate a unique model which achieve maximum CPU throughput scheduling while using machine learning to predict the performance of all the available CPU at the scheduling quantum granularity. We show how lightweight ANNs can equip highly true performance predictions for a different set of applications thereby aiding to improve scheduling effectiveness. The proposed structure model makes a decision for assigning the current task to a specific processor with determined speed. This decision is based on a number of criteria including the current task period and deadline, and the available processors' current utilization. The proposed model is composed of a number of artificial neurons layers that are trained to promote relationships between the input parameters and the produced aim output target. An ANN was learned to achieve an accurate prediction of a task to processor mapping and that processor suitable operating frequency. Accomplished results were compared to conventional schedulers which demonstrate, significant performance benefits. The proposed model achieved good feasibility performance and minimum power consumption. Comparing our ANN model throughput to other conventional schedulers shows comparable throughput with round robin schedulers.