Paper Title
MLSCHED: Machine Learning Based Job Scheduler
Abstract
This paper introduces MLSched, a novel scheduling scheme utilizing machine learning and deep learning
techniques, including LSTM, ANN, and Linear Regression. Targeting heterogeneous multicore systems, MLSched enhances
throughput by intelligently predicting thread parameters and IPC values for optimal thread scheduling. Compared to
existing schemes, MLSched demonstrates a 1.2X speedup and a 20% improvement in system throughput across Parsec and
Splash benchmarks, showcasing the effectiveness of machine learning in computer architecture.
Keywords - Heterogeneous multiprocessor, Machine learning, Thread Scheduling, Long Short Term Memory