Performance tuning of Red Hat Enterprise Linux
summary
Covers concrete RHEL tuning steps: use tuned profiles to pick latency vs throughput presets, inspect topology with lsCPU/numactl and prefer local NUMA allocation (measured 1.6s vs 2.0s runtime when bound), use perf/perf record to find hot locks, and choose transparent vs explicit huge pages — a demo shows transparent huge pages cutting a run from 12s to 7s (~1.77×). Describes vm tunables (dirty_background_ratio 10% / dirty_ratio 20%, min_free_kbytes scaling), cgroups CPU shares vs quota, and tools like numad/newmidi for automated placement.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:56Agenda: Performance Analysis & Tuning
- 3:22Red Hat Enterprise Linux Performance Evolution
- 4:40Tuned Overview
- 8:52Tools to display CPU and Memory (NUMA)
- 15:45Interaction between VM Tunables and NUMA
- 18:38Memory reclaim Watermarks
- 33:18Transparent Hugepages
- 38:48Cgroup default mount points
- 47:10Per file system flush daemon
- 48:34dirty ratio and dirty background ratio
- 50:24Some RHEL kernel features
related talks
52mPerformance tuninng for Red Hat Enterprise Linux.
48mOptimizing Linux Servers
51mLinux Systems Performance - Architecture
25mScheduling and Load Balancing in RHEL8 and Beyond
52mEffectively Measure and Reduce Kernel Latencies for Real-time Constraints
51mTuning RT Kernel to Improve Schedule Performance with Intel Platform
15mOptimizing Application Performance on Kubernetes
25mTune It Up! Enabling Low Latency in Kubernetes Clusters