Effective use of sleep states with context-aware selective resume

In response to increasing power costs and the social importance of improving energy efficiency, there has been significant work on taking advantage of a computer’s ability to enter low-power sleep states. However, current approaches do not allow fine-grained control of the transition process, so times between the power states are long. This limits the usefulness of these states, especially when computers are accessed sporadically. We introduce a context-aware selective resume framework that selectively resumes a system, with only the minimal set of devices needed for the waking task, to save power and shorten transition times. In this paper, we discuss the classes of applications that would benefit from selective resume, describe the design and implementation of CAESAR, our selective resume framework, and evaluate two applications, a temperature monitoring system and a fast memory server. Our evaluation demonstrates that context-aware selective resume can improve resume latency by 3.9X and energy savings over 5X compared to traditional approaches.