LinuxConf Europe 2007

Photos and reports

Timetable

Programme

Registration

LPI Exams

Conference Dinner (Sunday)

Duxford Excursion (Monday)

Exhibitors and Sponsors

Accommodation

Venue

Travel

About Cambridge

Kernel Summit 2007

Other GUUG events

Other UKUUG events

Venkatesh Pallipadi - Intel Corp.

Processor Power Management features and Process scheduler: Do we need to tie them together?

(In collaboration with Suresh B Siddha at Intel Corp.)

Today's microprocessors provide a wide variety of power-saving features. Processor P-state is the capability of running the processor at different voltage/frequency states. Processor C-state is the capability to go into various low power idle States (with varying wake up latency). Operating system policies like cpufreq-ondemand governor and cpuidle-menu governor use these processor power management features. But the Linux process scheduler is not aware of these processor power management features. There is a power v. performance scheduler policy, but it is a systemwide-global option set by the administrator.

In this paper, we will look into different mechanisms of incorporating processor P-state and C-state knowledge into the Linux process scheduler. We focus on their merits and demerits and various challenges involved. We will also look at ways to do this in a user-transparent way in which a processor-power management-aware process scheduler can dynamically select the scheduler power-savings or peak-performance policy. Another aspect that we will look at is how detailed scheduling information (idle micro-accounting) can benefit the existing cpufreq-ondemand governor.

Submitted paper

Paper (PDF) .