An Unbiased Look at the Energy Aware Scheduler (EAS)
summary
Energy Aware Scheduler (EAS) extends CFS to place tasks on big or little cores using PELT load estimates and DVFS info, then later replaces PELT with WALT (windowed demand) and Q HMP’s test-demand math. Measurements show ~20% lower power on YouTube playback but EAS suffers bursts — more than twice the frame drops during Chrome scrolling versus Q HMP. WALT plus HMP demand fixes both power and QoS enough that WALT becomes the mainline choice, though task‑packing inefficiencies remain.
AI-written summary · about
chapters (15)
- 0:00Start
- 3:51CFS operation principles
- 6:27big. LITTLE in a nutshell
- 8:01Scheduler for big LITTLE?
- 9:20EAS: basic principles
- 13:23Qualcomm HMP scheduler
- 17:01Task demand scaling
- 20:30Test: Youtube playback power
- 21:12Test: frame drops per sec.
- 27:20WALT: demand contribution calculation
- 31:11Result interpretation
- 32:51EAS: current status
- 34:40EAS and task packing
- 36:32Conclusions
- 37:16Credits
related talks
22mEAS Unit Testing
24mEnergy model accuracy
40mLatency hints for CFS task
16mDynamic Energy Model to handle leakage power
42mSteps Towards a Gaming-Optimized Scheduler
46mOptimizing Scheduler for Linux Gaming
17muclamp in CFS: Fairness, latency, and energy efficiency
38mTask Scheduling for Multicore Embedded Devices