Energy-Efficient Phase-Aware Scheduling for Heterogeneous Multicore Processors

While the multicore processors common today are typically homogeneous (i.e. composed of multiple cores of identical design), ever-more-stringent energy and performance constraints are making heterogeneous multicore processors increasingly attractive alternatives. Because applications vary significantly in the computing resources that they can effectively exploit, they observe drastically different energy consumption and performance depending on the characteristics of the processor core(s) they are running on. Heterogeneous computing cores can more efficiently meet the specialized needs of varied applications. However, while heterogeneous types of cores can be beneficial, the added complexity of heterogeneity can significantly complicate the scheduling of application threads to cores of different types. To achieve energy efficiency, a scheduler must attempt to minimize energy consumption by finding the optimal application to core mapping. Most existing approaches to mapping applications to heterogenous cores do so statically-that is they determine a priori the fitness of a particular application thread to a particular core type. This paper demonstrates that significant reduction in energy consumption can be achieved by dynamically adjusting this mapping as application behavior changes with new program phases. This paper further proposes an online scheduler that re- evaluates the program-to-core assignment when a phase-change of a program occurs, in order to optimize the scheduler for energy consumption. Our results show significant energy reduction over random scheduling of programs within a heterogeneous multicore processor.