Improve CPU Utilization by Designing a New Schedule Algorithm
summary
Claims a new offline scheduling class added to the Linux kernel that enforces a 20% cap per CPU for offline work, exposes an interface to mark pods as offline, and implements lightweight load balancing plus binary accounting. Tests show offline CPU utilization rising from ~15% to ~65% without hurting online tail latency; a problematic mixed setup doubled failure counts in one scenario and added ~20 ms per hop, but the new scheduler smooths offline bursts and preserves online reliability. Patches are released in TLinux.
AI-written summary · about
chapters (9)
- 0:00Intro
- 3:05Latency Sensitivity
- 5:54Classical Issue
- 7:28Solution
- 10:00Load Balancing
- 12:35Scheduling
- 15:15Generic Interface
- 20:40Off Balancing
- 23:00Conclusion
related talks
32mImprove CPU Utilization by Designing a New Schedule Algorithm
19msystem pressure on CPUs capacity and feedback to scheduler
25mLinux needs a Scheduler QOS API -- and it isn't nice(2)
50mSCHED_DEADLINE: It's Alive!
19mDo nothing fast: How to scale idle cpus?
46mOptimizing Scheduler for Linux Gaming
38mFixing Real-Time Scheduler Throttling in the Linux Kernel
32mCPU Power Saving Methods for Real-time Workloads