SCHED_DEADLINE desiderata and slightly crazy ideas
summary
Describes missing and in-development work for SCHED_DEADLINE: an RFC and patches for pressure inheritance to let mutex owners execute with donors’ runtime/period so runtime enforcement stays active; planned two-level CGroup support to put CFS groups inside deadline reservations and offer end-to-end pipeline guarantees; experiments with putting the deadline scheduler at the root to reclaim idle time and with least-laxity ordering to avoid priority inversion; and scalability fixes favoring partition or semi‑partition placement (split jobs) to reach ~80–90% usable CPU on multicore.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 0:46Future stuff
- 1:25References
- 3:28Why do we care
- 4:12Rate monotonic
- 4:37Cat deadline
- 5:14Overview
- 6:20Priority Inheritance
- 6:51Pressure Inheritance
- 9:56Scheduling Context
- 13:04Group Support
- 13:45Twolevel scheduler
- 15:02Realtime scheduler
- 18:07Scalability
- 19:23Partition Scheduler
- 20:29Same Partition Scheduler
- 21:18High Scalability
- 21:41Extra Organizing
- 22:24Outro
related talks
50mSCHED_DEADLINE: It's Alive!
50mUsing SCHED_DEADLINE
22mWelcome message and DL Server
25mLinux needs a Scheduler QOS API -- and it isn't nice(2)
46mOptimizing Scheduler for Linux Gaming
42mSteps Towards a Gaming-Optimized Scheduler
40mLatency hints for CFS task
19msystem pressure on CPUs capacity and feedback to scheduler