Reproduce “Silent Data Corruptions at Scale” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Harish Dattatraya Dixit, Sneha Pendharkar, Matt Beadon, Chris Mason, Tejasvi Chakravarthy, Bharath Muthiah, Sriram Sankar
Researchers from Facebook Inc. empirically found that Silent Data Corruptions (SDCs) are systemic and repeatable CPU hardware defects, not transient soft errors, occurring at rates far exceeding soft-error predictions. They presented a robust debugging methodology and demonstrated how SDCs cause critical application data loss, advocating for integrated hardware and software mitigation strategies.
Key insights
- SDCs are primarily caused by repeatable device characteristics and hardware defects within CPUs, rather than being limited to probabilistic soft errors.
- CPU SDC occurrence rates are orders of magnitude higher than traditional soft-error based FIT simulations, highlighting minimal error correction within many CPU functional blocks.
- SDCs are often data-dependent and do not generate system or kernel logs, making their detection and diagnosis profoundly challenging.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.