Study of a low power optimization method based on multithreaded load imbalance for embedded multicore

Wang Qing,Ji Zhenzhou,Zhu Suxia
DOI: https://doi.org/10.3772/j.issn.1002-0470.2013.09.004
2013-01-01
Abstract:In order to improve the energy efficiency of parallel application programs on chip multiprocessor systems, a high-performance low-power execution model is proposed for load imbalance in multi-threaded programs. Then, according to the model, an algorithm for control of parallel threads' execution frequency based on the run-time feedback scheme is presented to realize dynamical scaling of the execution frequency based on the load imbalance information of parallel threads to decrease energy consumption without affecting the performance of the parallel programs. The model and the algorithm were implemented on a multicore simulator. The experimental results show that the proposed model can save on average 13% of energy consumption, significantly reduce the energy consumption and improve the performance and power ratio.
What problem does this paper attempt to address?