Automated Fault Diagnosis with Event Traces
Two papers from Microsoft Research Asia (with Ni Lao as co-author) on automating system problem diagnosis using machine learning.
Papers Merged
- Automated Known Problem Diagnosis with Event Traces — Chun Yuan, Ni Lao, Ji-Rong Wen, Jiwei Li, Zheng Zhang, Yi-Min Wang, Wei-Ying Ma — EuroSys 2006
- Combining High Level Symptom Descriptions and Low Level State Information for Configuration Fault Diagnosis — Ni Lao, Ji-Rong Wen, Wei-Ying Ma, Yi-Min Wang
Key Contributions
Paper 1: Event Trace Classification
- Uses system call sequences (event traces) to correlate with solved problems
- Applies statistical learning (SVM) to classify system behavior
- Replaces vague text descriptions with structured behavioral data
- Achieves high accuracy on root cause recognition across four case studies (e.g., IE “page cannot be displayed” error)
Paradigm: Problem description → known problem → resolution
Paper 2: Hybrid Symptom-State Diagnosis
- Combines high-level symptom descriptions with low-level system state information
- Applied to Windows Registry configuration problems
- Bridges user-facing symptoms with system internals
Affiliations
- Microsoft Research Asia (Beijing)
- Microsoft Research (Redmond)
- Tsinghua University
- USTC
Connection to Vault
Early work by Ni Lao before his PRA research at CMU. Demonstrates ML applied to systems engineering — relates to the infrastructure monitoring concepts in Accelerate (Four Key Metrics) and the diagnostic automation theme in Automated Ticket Management.