Dirty Tricks of Computer Hardware
summary
Built a dependency graph from Linux 3.10 RC2 module imports to show kernel complexity, and cites concrete failure rates: 8% of DIMMs per year show errors and a university study finding 70% of consumer SSDs misbehave under repeated power cycles (13% brick). Recommends concrete fixes that actually work: ECC memory, checksums on metadata, journaling plus fsync, and verified drive flushes; warns that device write caches, FTL power loss, torn writes and unstable page writes are real causes of data loss.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 2:24How computers work
- 6:17Cosmic rays and bit flips
- 15:46Cache coherency problems
- 22:38SSDs and power loss vulnerabilities
- 27:16Voltage thresholds
- 31:40Torn disk writes
- 37:04Unstable page writes
- 41:59Future of disk writes
- 42:42Q&A
related talks
28mComplexity Killed Security
27mProtecting Data In-Use from Firmware and Physical Attacks
50mSmashing the Stack: Let’s Make It Less Fun And Unprofitable!
10mMcSee: Evaluating Advanced Rowhammer Attacks and Defenses via Automated...
01h 06mIt's Time for Operating Systems to Rediscover Hardware
14mHow not to secure your embedded system, lessons from automotive testing
46mFail-Safe Embedded Linux: Designing for Power Resilience
52mSecuring The Unseen: Vulnerability Research In Confidential Computing