Leakage-Aware Reallocation for Periodic Real-Time Tasks on Multicore Processors

Hongtao Huang, Feng Xia, Jijie Wang, Siyu Lei, Guowei Wu

Abstract

It is an increasingly important issue to reduce the energy consumption of computing systems. In this paper, we consider partition based energy-aware scheduling of periodic real-time tasks on multicore processors. The scheduling exploits dynamic voltage scaling (DVS) and core sleep scheduling to reduce both dynamic and leakage energy consumption. If the overhead of core state switching is non-negligible, however, the performance of this scheduling strategy in terms of energy efficiency might degrade. To achieve further energy saving, we extend the static task scheduling with run-time task reallocation. The basic idea is to aggregate idle time among cores so that as many cores as possible could be put into sleep in a way that the overall energy consumption is reduced. Simulation results show that the proposed approach results in up to 20% energy saving over traditional leakage-aware DVS.

Introduction

In modern real-time embedded systems, such as automotive controllers or industrial robotics, meeting strict timing deadlines is non-negotiable. Historically, energy efficiency in these systems was managed through Dynamic Voltage Scaling (DVS), which reduces power by slowing down the processor. However, as semiconductor technology has advanced to smaller scales, leakage power—the energy lost even when a transistor is inactive—has become a dominant factor.

Slowing a processor too much can actually increase total energy consumption because the core remains active and "leaking" for a longer duration. This defines a Critical Speed: a threshold below which further DVS scaling is counterproductive. While putting idle cores into a sleep state can mitigate leakage, the process of waking a core incurs a significant energy cost known as switching overhead. If idle periods are fragmented and short, the energy spent waking the core may exceed the energy saved during sleep.

The Leakage-Aware Reallocation Strategy

The researchers identify a "Balanced Partition Paradox." Traditional scheduling algorithms typically aim to balance the workload evenly across all available cores. While this is optimal for pure DVS, it is detrimental for leakage-aware systems because it fragments idle time into many small intervals that are too short to justify entering a sleep state.

To address this, the paper proposes Leakage-Aware (LA) Reallocation, a runtime strategy that prioritizes "clumping" tasks together. The algorithm works as follows:

  1. Initial Partitioning: Tasks are initially assigned using a "Worst-Fit" heuristic.
  2. Greedy Reallocation: At runtime, when a new job arrives, the system evaluates the core's expected idle time. If the idle period is shorter than the "sleep threshold" (the minimum time needed to make sleeping energy-efficient), the algorithm attempts to move the task to a different core.
  3. Candidate Core Selection: Tasks are migrated to a specific set of candidate cores. A migration is only permitted if it does not violate real-time deadlines and does not force the processor speed above the Critical Speed.

By intentionally unbalancing the load, the system creates longer, continuous idle intervals on some cores, allowing them to enter deep sleep states more effectively.

Experimental Evaluation

The proposed approach was evaluated using a custom-built simulator modeling a 70nm CMOS processor. The simulation considered random sets of periodic tasks with varying execution times to account for "early completion" slack.

Energy consumption vs. switching overhead (Esw)

The results demonstrate that LA-Reallocation consistently outperforms traditional leakage-aware DVS. As shown in the performance graphs, the proposed method achieves up to a 20% reduction in normalized energy consumption. The advantage is particularly pronounced when switching overhead is low.

Energy consumption vs. execution time ratio (cc/WCET)

The algorithm is most effective at moderate system utilization (around 0.4). Furthermore, the LA-Reallocation method scales better with core counts; while "Pure DVS" energy consumption spikes as more cores are added due to increased leakage from idle cores, the reallocation strategy effectively "shuts down" unnecessary cores.

Conclusion

This research highlights that for modern multicore processors, energy efficiency requires a departure from traditional load-balancing. By aggregating idle time through runtime task reallocation, systems can significantly reduce leakage power without compromising real-time constraints. Future work may explore more complex task dependencies and per-core independent voltage scaling to further refine these energy-saving techniques.

Summary updated

Abstract

It is an increasingly important issue to reduce the energy consumption of computing systems. In this paper, we consider partition based energy-aware scheduling of periodic real-time tasks on multicore processors. The scheduling exploits dynamic voltage scaling (DVS) and core sleep scheduling to reduce both dynamic and leakage energy consumption. If the overhead of core state switching is non-negligible, however, the performance of this scheduling strategy in terms of energy efficiency might degrade. To achieve further energy saving, we extend the static task scheduling with run-time task reallocation. The basic idea is to aggregate idle time among cores so that as many cores as possible could be put into sleep in a way that the overall energy consumption is reduced. Simulation results show that the proposed approach results in up to 20% energy saving over traditional leakage-aware DVS.
BibTeX
@inproceedings{huang2010leakage,
  author = {Hongtao Huang and Feng Xia and Jijie Wang and Siyu Lei and Guowei Wu},
  title = {{Leakage-Aware Reallocation for Periodic Real-Time Tasks on Multicore Processors}},
  booktitle = {FCST},
  year = {2010},
  url = {https://openreview.net/forum?id=H8X4NBOBUX},
}
Figure