The University of Southampton
University of Southampton Institutional Repository

System-level power management using online machine learning for prediction and adaptation

System-level power management using online machine learning for prediction and adaptation
System-level power management using online machine learning for prediction and adaptation
Nowadays embedded devices have the need to be portable, battery powered and high performance. This need for high performance makes power management a matter of critical priority. Power management algorithms exist, but most of the approaches focus on an energy-performance trade-off oblivious to the applications running on the system. Others are application-specific and their solution cannot be applied to other applications.

This work proposes Shepherd, a cross-layer runtime management system for reduction of energy consumption whilst offering soft real-time performance. It is cross-layer because it takes the performance requirements from the application, and learns to adjust the power management knobs to provide the expected performance at the minimum cost of energy. Shepherd is implemented as a Linux governor running at OS level, this layer offers a low-overhead interface to change the CPU voltage and frequency dynamically.

As opposed to the reactive behaviour of Linux Governors, Shepherd adapts to the application-specific performance requirements dynamically, and proactively selects the power state that fulfils these requirements while consuming the least power. Proactiveness is achieved by using AEWMA for adapting to the upcoming workload. These adaptations are facilitated using a model-free reinforcement learning algorithm, that once it learns the optimal decisions it starts exploiting them. This work enables Shepherd to work with different applications. A programming framework was designed to allow programmers to develop their applications to be power-aware, by enabling them to send their performance requirements and annotations to Shepherd and provide the cross-layer soft real-time performance desired.

Shepherd is implemented within the Linux Kernel 3.7.10, interfacing with the application and hardware to select an appropriate voltage-frequency control for the executing application. The performance of Shepherd is demonstrated on an ARM Cortex-A8 processor. Experiments conducted with multimedia applications demonstrate that Shepherd minimises energy consumption by up to 30% against existing Governors. Also, the framework has been used to adapt example applications to work with Shepherd, achieving 60% energy savings compared to the existing approaches.
Maeda-Nunez, Luis
75dd521f-c2c0-4006-90d0-9ae32a3534cf
Maeda-Nunez, Luis
75dd521f-c2c0-4006-90d0-9ae32a3534cf
Merrett, Geoffrey
89b3a696-41de-44c3-89aa-b0aa29f54020

Maeda-Nunez, Luis (2016) System-level power management using online machine learning for prediction and adaptation. University of Southampton, Faculty of Physical Sciences and Engineering, Doctoral Thesis, 155pp.

Record type: Thesis (Doctoral)

Abstract

Nowadays embedded devices have the need to be portable, battery powered and high performance. This need for high performance makes power management a matter of critical priority. Power management algorithms exist, but most of the approaches focus on an energy-performance trade-off oblivious to the applications running on the system. Others are application-specific and their solution cannot be applied to other applications.

This work proposes Shepherd, a cross-layer runtime management system for reduction of energy consumption whilst offering soft real-time performance. It is cross-layer because it takes the performance requirements from the application, and learns to adjust the power management knobs to provide the expected performance at the minimum cost of energy. Shepherd is implemented as a Linux governor running at OS level, this layer offers a low-overhead interface to change the CPU voltage and frequency dynamically.

As opposed to the reactive behaviour of Linux Governors, Shepherd adapts to the application-specific performance requirements dynamically, and proactively selects the power state that fulfils these requirements while consuming the least power. Proactiveness is achieved by using AEWMA for adapting to the upcoming workload. These adaptations are facilitated using a model-free reinforcement learning algorithm, that once it learns the optimal decisions it starts exploiting them. This work enables Shepherd to work with different applications. A programming framework was designed to allow programmers to develop their applications to be power-aware, by enabling them to send their performance requirements and annotations to Shepherd and provide the cross-layer soft real-time performance desired.

Shepherd is implemented within the Linux Kernel 3.7.10, interfacing with the application and hardware to select an appropriate voltage-frequency control for the executing application. The performance of Shepherd is demonstrated on an ARM Cortex-A8 processor. Experiments conducted with multimedia applications demonstrate that Shepherd minimises energy consumption by up to 30% against existing Governors. Also, the framework has been used to adapt example applications to work with Shepherd, achieving 60% energy savings compared to the existing approaches.

Text
Luis_Alfonso_Maeda_Nunez-Final_Thesis.pdf - Other
Available under License University of Southampton Thesis Licence.
Download (12MB)

More information

Published date: 22 July 2016
Organisations: University of Southampton, EEE

Identifiers

Local EPrints ID: 399995
URI: http://eprints.soton.ac.uk/id/eprint/399995
PURE UUID: 9f4ce80c-96d9-468e-98e0-06b363ec2ded
ORCID for Geoffrey Merrett: ORCID iD orcid.org/0000-0003-4980-3894

Catalogue record

Date deposited: 27 Oct 2016 16:00
Last modified: 17 Jul 2018 00:32

Export record

Contributors

Author: Luis Maeda-Nunez
Thesis advisor: Geoffrey Merrett ORCID iD

University divisions

Download statistics

Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.

View more statistics

Atom RSS 1.0 RSS 2.0

Contact ePrints Soton: eprints@soton.ac.uk

ePrints Soton supports OAI 2.0 with a base URL of http://eprints.soton.ac.uk/cgi/oai2

This repository has been built using EPrints software, developed at the University of Southampton, but available to everyone to use.

We use cookies to ensure that we give you the best experience on our website. If you continue without changing your settings, we will assume that you are happy to receive cookies on the University of Southampton website.

×