Introduction to Realtime Linux
summary
Overview of concrete results and setup: compares PREEMPT_RT (single-kernel) and Xenomai (dual-kernel) on a Cortex‑A9 rocket board, using hackbench worst‑case load and an OSADL latency box to fire 10 kHz interrupts. User‑space reaction latencies sit under 100 µs worst case (around 80–95 µs), isolating a CPU drops kernel reaction to ~60 µs, and using the ARM FIQ path reaches ~10 µs average with ~30 µs worst case. PREEMPT_RT achieves comparable real‑world determinism without special libraries.
AI-written summary · about
chapters (30)
- 0:00Introduction
- 1:11Overview
- 2:22What is Realtime
- 3:12What is not Realtime
- 4:02Timing guarantees
- 6:32Error condition
- 7:08Soft Realtime
- 8:15Realtime Linux Users
- 9:15Realtime Linux Requirements
- 10:02Priority Inversion
- 11:23Priority Inheritance
- 12:04Dual Kernel
- 12:35Micro Kernel
- 14:40Realtime Linux
- 15:33RTAI
- 19:30Cinema
- 22:00Cinema Structure
- 22:51Dual Kernels
- 24:44Realtime preemption
- 28:10mainline integration
- 29:43how to make Linux realtime
- 32:18Realtime preemption overview
- 35:02Hellbench
- 35:52Latency box
- 37:53Use cases
- 38:47Results
- 44:57FIQ
- 45:58Final comparison
- 46:13Conclusion
- 47:44Questions
related talks
41mRealtime Linux Beyond PREEMPT_RT: Exploring Xenomai's Dual-Kernel Approach
38mBuilding Hard Real-Time Linux Systems with Xenomai
26mReal-time Linux: what is next?
44mReal-Time Linux on Embedded Multicore Processors
47mReal-Time Linux: What Is Next?
52mEffectively Measure and Reduce Kernel Latencies for Real-time Constraints
39mEvaluation of PREEMPT_RT in Virtualized Environments
40mRT Linux in Safety-Critical Systems: The Potential and the Challenges