Register / Login

CFP
Paper Guide

Schedule

Keynotes
Summits
Presentations
Tutorials
BOFS

Venue
Travel & Hotel
Costs & Savings

FAQ

Media
Archives
Contact

Home

Workload Governor: Implementing aggressive platform power management

Power management has been dealt with in the past with two important infrastructures in the Linux kernel: cpufreq and cpuidle. Linux kernel policies like cpufreq on demand governor and cpuidle-menu governor make effective use of these processor power management features, giving power savings to the end user. However, the approaches so far have been restricted to processor power management during idle. We are seeing an increasing need, especially in the mobile internet device space, to align the entire platform into processor idle windows, where in all platform components and devices can enter active idle states, with the capability of managing their own power in idle state. Although cpuidle governors (especially menu governor) uses PM QoS requirements to understand platform-wide QoS requirements, this is very limited in its usage and impact.

In this paper, we propose the following: (1) Workload governor: A new CPUIDLE governor, which can understand the nature of workload running on the platform based on available platform/OS indicators, and use that knowledge to better manage CPU C-states (2)Extensions to CPUIDLE infrastructure that will allow for a platform-specific power management callback. This callback is intended to be a low latency call, which can implement multiple possible platform-wide low power active idle states, if the CPU was entering a deep C-state.

We will also describe some experimental studies that we have done with the above prototypes on a Linux based mobile device.

register | call for papers

Copyright © 2009 Linux Symposium Inc. All rights reserved.
Linux is a registered trademark of Linus Torvalds.