Scheduling and Load Balancing in RHEL8 and Beyond
summary
Shows that RHEL8.3 adopts a Linux 5.5 load‑balancer rework that fixes a cgroup imbalance by counting running tasks instead of only scaled fair load, restoring a NAS Parallel Benchmarks NAS LU 44 threads run on a 48‑CPU system from about one‑tenth speed back to normal. A remaining sawtooth of migrations with 24 threads comes from a conflicting NUMA auto‑balancer; RHEL8.4/5.9 align that logic and stop the ping‑pong. Traces use the number_running tracepoint and plot_number_running to visualize results.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:12Agenda
- 0:57About Me
- 1:57The Scheduler
- 6:40The Issue
- 9:11The Results
- 12:30C Group Rework
- 14:55C Group Rework with 24 threads
- 17:18Pneuma Balancing
- 20:05Future features
related talks
55mPerformance tuning of Red Hat Enterprise Linux
52mPerformance tuninng for Red Hat Enterprise Linux.
41mLoad Balancing Roundup
35mOpening the Box: Diagnosing Operating-System Task-Scheduler...
20mLinux Kernel Scheduling and split-LLC architectures: Overview, Ch...
36mTuning Memory and the Network Stack in Linux
24mA Case Study in Chaos Testing: Uncovering Kernel Scaling Issues
32mImprove CPU Utilization by Designing a New Schedule Algorithm