Reliability Availability and Serviceability (RAS) on Linux
summary
Describes concrete RAS work in Linux: Kernel adds trace-based delivery, wait-for-event, per-client pipelines and system timestamps in Linux Kernel 3.10; a userland daemon rasdaemon hooks those trace events, consolidates MCA, EDAC and PCIe error streams, logs into a database, and maps reported faults to DIMM labels so hardware replacements are actionable. Tool ships in Fedora 18, is validated with fake injections and supports Intel hardware error injection when BIOS enables it, but vendors often disable that feature so end-to-end testing is limited.
AI-written summary · about
chapters (8)
- 0:00Start
- 3:46Improving RAS (1)
- 6:01RAS features on Linux (1)
- 8:45RAS features on Linux (2)
- 15:16Typical D-RAM implementation
- 18:24Memory Arrangements on PC
- 20:01Evolution of RAS on Kernel (2)
- 23:36Firmware First x Hardware First
related talks
43mRTLA: Real-time Linux Analysis Toolset
19mRISC-V RAS Error-Record Register Interface (RERI)
51mRTLA Timerlat - the next step in scheduling latency analysis
50mrtla timerlat: Debugging Real-Time Linux Scheduling Latency
19mAdvance the Performance Analysis on RISC-V
27mShipping Software with an SRE Mindset
49mDebugging Methodologies for Realtime Issues in Linux Systems
32mThe culture of Site Reliability Engineering