As a society, we have come to rely upon a diverse assortment of computer systems to carry out essential tasks ranging from stock market transactions to Internet searches to FaceBook updates. Such systems are often expected to stay operational 24×7, come hell or high water. At the same time, software and systems have become increasingly large and complex, and therefore prone to failure. The task of debugging and diagnosing these failures can be as intricate as needlework and as Herculean as wrangling an angry rhinoceros. In this talk, I will give an overview of recent research projects that try to tame the systems beast using tools from machine learning and statistics. These methods analyze and mine the data contained in large volumes of automatically generated system logs, looking for clues as to why the system is failing. The results are meant to aid human operators in honing in on parts of the system that may require attention. I will describe successes and failures in this domain, and give some ideas of where we might go next.

Bio:
Alice Zheng is a researcher in the Machine Learning Group at Microsoft Research, Redmond. She spent her formative graduate student years at UC Berkeley, working on using statistical methods to automatically diagnose software failures. She spent two years at Carnegie Mellon University as a postdoc, working on analyzing social networks and diagnosing file systems. Since joining Microsoft, she has worked on lock contention problems in the Windows operating system. More recently, she is working on building tools that automate machine learning to enable easier and wider applications.
———-

As always, there will be dinner sponsored by Silicon Mechanics. Check them out at http://www.siliconmechanics.com/

There will also be several CACert assurers present.

The meeting will be at the Electrical Engineering building on the University of Washington Campus, aka EE1. Directions are linked to the EE Department’s web site above. Parking is $5 after 5pm.