Unveiling Heisenbugs with Diversified Execution
summary
Presents a concrete system and results: a WebAssembly-backed dynamic data-race detector called Beanstalk that uses trap delay injection and stochastic partial instrumentation across a 21‑device heterogeneous cluster to run over 2 million executions and find 109 data‑race bugs. Introduces a heisen factor metric to quantify nonuniform observability, shows diversity outperforms a homogeneous max‑instrumentation baseline (captures rare bugs up to 0.3 detection probability) and beats the baseline with only ~20% instrumentation.
AI-written summary · about
chapters (13)
- 0:00Heisenbugs in embedded systems
- 1:04Debugging non-deterministic systems
- 2:10Detecting data races with trap delay
- 3:34Bug detectability observations
- 4:17Hypothesis for embedded debugging
- 5:20Beantock with WebAssembly
- 6:21Metrics for bug analysis
- 7:24Heisenbug patterns detected
- 9:13Diversity for bug detection
- 10:58Beyond data races
- 11:40Future of diversity testing
- 12:15Q&A: Bug categorization
- 12:36Q&A: Reproducing bugs
related talks
17mPoster/Demo Previews
32mWebAssembly on the Factory Floor: Efficient and Secure Processing of High Velocity Machine Data
48mCapturing 0Day Exploits With Perfectly Placed Hardware Traps
11mBoost SoC Debug with Smart Monitors and Embedded Software
07mREHASH: A Flexible, Developer Focused, Heuristic Adaptation Platform for Intermittently Powered ...
24mInception: System-Wide Security Testing of Real-World Embedded Systems Software
09mRace to Charge: Concurrency Bugs and State Integrity Violations in EV...
14mFreedom from memory interference