Execution-variance-aware task allocation for energy minimization on the big.LITTLE architecture

Abstract

With the increasing computational demands of applications, the research of energy-efficient heterogeneous multiprocessor scheduling is becoming more and more important. The problem can be addressed as allocating tasks to the right cores and setting the appropriate operating frequencies. In this paper, we present a task allocation algorithm for a single-ISA big.LITTLE architecture which consists of two different types of cores: high-performance core and energy-efficient core. Based on the execution variance of two types of cores, a heuristic that approximate the optimal load distribution is proposed. We experimentally simulate our method and compare it with the state-of-the-art solutions. Experimental results show that average 20.4% less energy consumption (and up to 60.2% for some cases) can be saved from the typical Little-Core-First with satisfying the deadline constraints of all tasks.