Energy Efficient Scheduling of Real Time Signal Processing Applications through Combined DVFS and DPM

Abstract : This paper proposes a framework to design energy efficient signal processing systems. The energy efficiency is provided by combining Dynamic Frequency and Voltage Scaling (DVFS) and Dynamic Power Management (DPM). The framework is based on Synchronous Dataflow (SDF) modeling of signal processing applications. A transformation to a single rate form is performed to expose the application parallelism. An automated scheduling is then performed, minimizing the constraint of energy efficiency and providing DVFS and DPM decisions. This framework uses an architecture model including the number of available cores, the per-actor processing load and the energy per-cycle, derived from time and power measurements of modelled applications. After introducing the proposed framework, the energy characterization of big.LITTLE SoC systems is described. A generic approach is presented to generate the energy model of a platform from power measurements as customized polynomials. Finally, the experimental results on a Samsung Exynos 5410 big.LITTLE processor show that the energy optimal execution is not obtained by Linux governors that can execute either as-fast-as-possible or as-slow-as-possible. Instead, the most energy efficient scheduling is obtained by adapting both DVFS and DPM to application needs.
Liste complète des métadonnées

Cited literature [16 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01299607
Contributor : Erwan Nogues <>
Submitted on : Wednesday, April 27, 2016 - 6:53:02 PM
Last modification on : Thursday, February 7, 2019 - 5:50:22 PM
Document(s) archivé(s) le : Thursday, July 28, 2016 - 10:48:58 AM

File

pdp_cam.pdf
Files produced by the author(s)

Identifiers

Citation

Erwan Nogues, Maxime Pelcat, Daniel Menard, Alexandre Mercat. Energy Efficient Scheduling of Real Time Signal Processing Applications through Combined DVFS and DPM . 24th Euromicro International Conference on Parallel, Distributed, and Network-Based Processing (PDP) , Feb 2016, Heraklion, Greece. ⟨10.1109/PDP.2016.15⟩. ⟨hal-01299607⟩

Share

Metrics

Record views

584

Files downloads

230