Artificial Neural Network based Task Scheduling for Heterogeneous Systems
Manjari Gupta, Lava Bhargava, S. Indu · 2020
Heterogeneous systems containing diverse computational resources have opened a new avenue in the field of multicore processors. Increasing versatility in applications and their performance requirements has shifted the focus to heterogeneous architecture. To leverage the additional computational capacity provided by these systems, efficient scheduling algorithms are required. The recent development in the field of machine learning provides the necessary tools in the form of prediction models and can be employed to achieve desired results. One such machine learning technique is the Artificial Neural Network (ANN) which has been used in the proposed work to achieve a high throughput heterogeneous system. An ANN-based scheduler that maximizes the performance of the cores by predicting application behavior in the next scheduling interval using core and workload statistics has been presented. Two distinct prediction models are trained for each core type thereby improving the accuracy of the model. The proposed method yields average performance improvement in the range of 6.5%-9.7% from conventional Fairness-aware scheduler and 2.5%-1.5% from a novel Dynamic ANN scheduler.