On Failure Diagnosis of the Storage Stack
Talkat 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage '20) · 2020-07-13..07-14
summary
A cross-layer diagnosis platform called X-ray combines a host agent that captures kernel and application events with a dev agent that intercepts device commands inside a VM without kernel mods or special hardware, aligns events using ndp timestamp sync, builds a correlation tree of system calls, kernel functions and device commands, and applies rule-based pruning to highlight suspects. Tested on five real failures it reduces the search space to under 5% of nodes on average and isolates root causes such as a misattributed kernel bug.
AI-written summary · about
