SCHED_DEADLINE: It's Alive!
Talkat Embedded Linux Conference + OpenIoT Summit North America 2017 · on Tuesday, 2017-02-21· at 15:00· Grand Ballroom I/II·
summary
SCHED_DEADLINE gains concrete patches: a bandwidth reclaiming (greedy reclamation of unused bandwidth) prototype lets tasks temporarily exceed their static runtime up to a configurable u_max, tracked by per-CPU active-utilization and a zero‑lag timer to avoid abuse on wakeup. Runtime adjusts by a simple ratio for runtime scaling with frequency, supporting big.LITTLE and multiprocessor migration by moving contributions between CPUs. Early synthetic tests show a 30–40% reduction in missed deadlines versus fixed-budget behavior; group/hierarchical deadline support is next.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:14Agenda
- 5:10Why is this developing
- 8:22New features
- 10:56greedy reclamation
- 13:03tracking
- 19:42reclaiming
- 22:26current UX
- 23:36solution
- 24:56multiprocess
- 26:48working progress
- 29:39best case runtime
- 39:09results
- 40:02group scheduling
- 41:35hierarchical approach
- 44:56multiprocessor approach
- 46:00future
related talks
23mSCHED_DEADLINE desiderata and slightly crazy ideas
50mUsing SCHED_DEADLINE
22mWelcome message and DL Server
46mOptimizing Scheduler for Linux Gaming
32mImprove CPU Utilization by Designing a New Schedule Algorithm
44mRestartable Sequences: Scheduler-Aware Scaling of Memory Use on Many-Core Systems
40mLatency hints for CFS task
42mRV and the deadline scheduler: how to verify your subsystem at runti...