How do we make a Steamdeck scheduler work on large servers
summary
Meta evaluates SCX_LAVD on large servers and finds concrete problems and fixes: 50+ CPUs per CCX causes heavy DSQ lock contention and task thrashing so a middle-ground virtual-domain split improves L2 locality without losing work conservation; pinned tasks move to per-CPU DSQs and use a smaller time slice for cohabiting pinned tasks to avoid latency spikes; interrupt-heavy cores get interrupt-time accounting reimbursement and a capacity-aware load balancer to keep latency‑critical threads on responsive cores.
AI-written summary · about
chapters (6)
- 0:00Introduction to scheduler challenges
- 1:04CCX domain contention issues
- 2:31Virtual domain trade-offs
- 4:40Pinned task problems
- 8:12IRQ heavy workload handling
- 10:45Q&A session
related talks
46mOptimizing Scheduler for Linux Gaming
44mRestartable Sequences: Scheduler-Aware Scaling of Memory Use on Many-Core Systems
19mDo nothing fast: How to scale idle cpus?
35mOpening the Box: Diagnosing Operating-System Task-Scheduler...
21mPreview: Operating Systems
40mLatency hints for CFS task
22mWelcome message and DL Server
24mPREEMPT_RT Q&A with tglx