Dynamic counters and the efficient and effective online power management of embedded real-time systems

Abstract

Energy consumption has become an important issue for modern embedded systems. This is because, one does not only like to deploy high-performance systems and provide guaranteed services, but also request system deployments to last as long as possible. With online dynamic power management (DPM), one adaptively changes the system's power mode, i. e., schedules when to turn on and off on-the-fly, to achieve energy savings. This paper introduces dynamic counters and discusses their usage in the context of (online) DPM in a setting where systems have to fulfill hard real-time requirements. The proposed scheme enables one to conservatively bound the future workload and thereby to safely schedule on-and off-periods of devices. Simulation results indicate that the proposed technique is more efficient and more effective with respect to energy savings, compared to previous work.

Online workload monitoring with the feedback of actual execution time for real-time systems

“…With the observations that arrival curve can be conservatively approximated by a set of staircase functions, Lampka et al [10] proposed using dynamic counters to predict the future coming events. This method is extended by Huang et al [8] to use dual leaky-bucket counters to monitor the staircase function.…”

“…1. Instead of event counters in [8], [10], the dynamic counters in Algo. 1 are the allowable burst capacity.…”

“…The system is assumed to have only two states [10], which are the busy state for event processing, and the idle state for doing nothing. In Algo.…”

“…Event-based monitors verify that task activations do not exceed the designed upper bound by employing arrival curve [8], [10], [14], [15]. Workload-based monitors verify that the system workload does not exceed the designed upper bound by employing WAF [13].…”

“…For the single event stream, the monitor is based on the socalled dynamic counters [8], [10], which itself is derived from the arrival curve model. By adopting the arrival curve to model the incoming system workload as event streams, our monitor can cope with non-deterministic arrival patterns of the incoming workload.…”