Usage history-directed power management for smartphones

Abstract

Smartphones are now equipped with powerful application processors to meet the requirement of performance demanding apps. This often leads to over-provisioned hardware, which drains the battery quickly. To save energy for battery-powered smartphones, it is necessary to let the processor run at a power-saving state without sacrificing user experience. Android has implemented a set of CPU governors by leveraging dynamic voltage and frequency scaling (DVFS) according to the computational requirements of apps. However, they commonly adopt very conservative policies due to limited information, therefore leaving considerable energy reduction opportunities unexplored; or they are not responsive to user interactions, leading to poor user experiences. In this work, we start from an important observation on the repetitive patterns of smartphone usage for each individual user, and propose UH-DVFS, a usage history-directed DVFS framework leveraging on this observation. UH-DVFS can identify repetitive user transactions, and store their execution history information within a table. When such a user transaction is launched again, the table is consulted for an appropriate CPU frequency adaptation to reduce the energy consumption of the user transaction without sacrificing user experience. We have implemented the proposed framework on Android smartphones, and have tested it with real-world interaction-intensive apps. The results show that our framework can save energy from 10% to 36% without affecting the quality of user experiences.