Task behaviour inputs to a heterogeneous multiprocessor scheduler

Abstract

A novel and intuitive way of scheduling entities on a heterogeneous multiprocessing system is presented. The key idea is to understand the behavioral characteristics (foreground/background, IO-bound/CPU-bound) of a scheduling entity to predict the need for its processing bandwidth. This is then used by the scheduler to influence the selection of the big cluster (high-performance) or the LITTLE cluster (power-saving) for optimal performance and power gains. The experimental results confirm that this method leverages the characteristics of a heterogeneous architecture to achieve the desired benefits.