Download presentation
Presentation is loading. Please wait.
Published byMercy Atkinson Modified over 9 years ago
1
Progress Report 2015/08/03
2
Related Works Materials from ◦ Energy Aware Scheduling workshop @ Kernel Summit 2014 ◦ Energy-aware Scheduling and CPU Power Management micro-conference @ LPC 2014 ◦ Power-aware Scheduling workshop @ Kernel Summit 2013 ◦ Power-efficient Scheduling micro-conference @ LPC 2013
3
Energy Aware Scheduling(EAS) A set of extensions to the Linux kernel that introduce energy model based decision making for task scheduling and power-performance control. Aims to make power-performance control more centralized with the scheduler being the primary driver for power-performance decisions. ◦ Current situation: the scheduler, cpufreq and cpuidle. ◦ Goal: simplify power-performance management with a scheduler-driven policy and a small set of well-defined tunables. Aims to provide tools that assist with energy model creation, qualification, the quantification of energy usage per workload and power-performance tuning.
4
Some Topics Tools and Benchmarking for energy improvements ◦ Idlestat: a trace analyzer based on a per- platform energy model(P-state and C-state) to help estimate the energy consumption of a measured workload. ◦ rt-app: creates workload patterns based on workload description files (current JSON). Two workloads described currently: audio playback and a generic web-browsing workload.
5
Some Topics(Cont.) Scheduler task and CPU load/utilization tracking ◦ Require good CPU-utilization tracking. ◦ Make load tracking aware of CPU frequency scaling. Energy Model in kernel ◦ consists of P- and C-state tables attached to different levels in the sched_domain hierarchy. ◦ a simple way to determine the implication of a scheduling decision.
6
Some Topics(Cont.) cpufreq and scheduling cpuidle and scheduling ◦ integrate frequency scaling and cpuidle with the scheduler. Power management and virtualization ◦ Guests shouldn't attempt to do power management.
7
Next Scheduling mechanism/algorithm ◦ Assume the energy model is given. Try the tools and benchmarks ◦ Idlestat and rt-app More recent related works.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.