A Framework to Improve Performance and Energy Efficiency of Embedded Intelligence Service Systems

Due to the high real-time requirement and limited resources, the computing performance and power consumption are critical to embedded intelligent service systems. The thread group scheduling strategy, as a common method to optimize multi-core processor’s performance, performs poorly in power optimization. However, with data volume continues proliferating, the power consumption is still surging. Correspondingly, many power optimization methods, if not scheduled well, will swap between cache states frequently, thus degrading device performance. In this paper, we propose a framework that combines the thread scheduling strategies with dynamic power mode control strategies together to make better trade-off between system performance and power consumption. Experimental results show that compared to using only scheduling policies, the systems combining Dynamic Power Management (DPM) and Bank Usage Table (BUT) policies have 14.8% and 9.5% performance growth as well as 9.9% and 13.2% power consumption reduction, respectively. The energy delay product (EDP) is decreased by 20.1% and 22.5%, respectively.