Real-time Scheduling Fault Simulation
summary
Evaluates concrete fault-injection techniques: strace and bpftrace (eBPF) inject syscall errors easily, the kernel fault injection framework exists but won’t work out-of-the-box on many distros, and kprobes let you attach deeper but hit locking and verifier limits. Finds scheduler core is ~57k lines (v6.10) with non-exported internals like pick_next_task, making precise jitter, early/late wakeups or deadline overruns hard without kernel patches. Concludes practical path is adding scheduler hooks or targeted patches; no finished farm-scale results yet.
AI-written summary · about
chapters (13)
- 0:00Real-time scheduling simulation
- 1:50Linux real-time scheduling history
- 3:13Linux real-time scheduling classes
- 7:07Disrupting real-time scheduling
- 8:54User-space fault injection
- 11:22Kernel fault injection framework
- 13:35eBPF for scheduling disruption
- 18:15Disrupting at scheduling points
- 21:13Linux scheduler deep dive
- 25:15Kernel patching challenges
- 29:31Scheduler disruption challenges
- 30:56Future work and next steps
- 33:05Q&A
related talks
39mReal-Time Scheduling Fault Simulation
26mIt's Coming From Inside the House: Kernelspace Fault Injection with KRF
25mApplying Program Verification to Linux Kernel Code: Challenges, Practices, and...
16mscx_chaos: Pushing the sched_ext API to schedule badly
19mCAFault: Enhance Fault Injection Technique in Practical Distributed Systems via...
28mExtreme OS Kernel Testing
42mExercising the Linux scheduler with Yogini
47mHunting Heisenbugs