Journey of improving RT performance on Single CPU equipped PREEMPT_RT Linux
summary
Demonstrates squeezing RT and housekeeping onto one core by running PREEMPT_RT workloads as pinned VMs under ACRN type‑1 hypervisor, using CPU pinning, cache partitioning (CAT), and passthrough, and validating with cyclictest + stress‑ng + iPerf. Identifies hotspots: softirqs, igb driver, watchdog reads, link up/down and DHCP causing extra jitter; applies fixes like hard IRQs, static IP, service trimming and static IRQ affinity and reduces observed jitter from ~106 microseconds to ~22 microseconds.
AI-written summary · about
chapters (7)
- 0:00ECI and realtime Linux
- 7:47Workload consolidation use cases
- 13:26Virtual machines vs containers
- 17:19Cache partitioning for realtime
- 23:46Tuning single CPU core
- 28:25Benchmarking RT gains
- 31:37Call for collaboration
related talks
36mJourney of improving RT performance on Single CPU equipped PREEMPT_RT Linux
50mInside the RT Patch
39mEvaluation of PREEMPT_RT in Virtualized Environments
53mPREEMPT_RT - How Not to Break It.
52mEffectively Measure and Reduce Kernel Latencies for Real-time Constraints
35mPreempt-RT Raspberry Pi Linux
40mRT Linux in Safety-Critical Systems: The Potential and the Challenges
26mReal-time Linux: what is next?