Mining NANOG Mailing List∗

It had been shown that the misbehaviors by few malicious, compromised or misconfigured BGP routers could lead to serious outages in Internet. This failing becomes progressively crucial as the recent prosper of outage-sensitive applications such as Voice over IP, streaming media, and video conferencing. To address these misbehaviors, previous work mainly focus on distributedly detect or prevent outages using limited state of Internet. In this paper, we present a first step towards efficient troubleshooting by mining network operators mailing lists. Using Natural Language Processing (NLP) and Machine Learning, we develop a new approach to extract useful information from the mailing forum on North American Network Operators Group (NANOG). Our experimental results show that the proposed approach detects 94 out of 105 outages from NANOG with a false positive rate of only 7.3%. We validate the extracted outage using real network logs collected by Route Views project. While our approach is not perfectly accurate, we envision it to be a useful information to existing anomaly detection/prevension mechanisms.