Abstract
In this paper, we analyze the issues that cause a device to consume more energy than it really needs and propose Just-in-time Device Scheduler (JDS), an online device scheduling algorithm, to address the issues. JDS consists of two major components: Iterative slack allocator and Dynamic handler. Iterative slack allocator statically assign slack to the tasks to reduce the energy consumed by device due to misprediction of job execution. Dynamic handler fine-tunes slack for the jobs in runqueue at run-time to reduce the energy waste incurred by unnecessary slack. JDS is compared with EEDS, the state-of-the-art online device scheduling algorithm, in evaluation. The experimental results show that JDS saves up to 33% of device energy consumption and 72% of device idle energy compared to EEDS.