Abstract
Effective diagnosis and early identification of system problems are crucial to the reliable operation of today’s computer systems. In this article we describe the Melody project, where we apply machine learning in several tools that aim to aid in these tasks in IBM System X servers. Several inherent problems present themselves when developing machine learning tools in this domain. These include the difficulty of obtaining labeled examples of system problems, continuous changes in the features describing a system, and the requirement for a low false positive ratio. Our work puts forward several directions for dealing with these problems.
Original language | English GB |
---|---|
Title of host publication | Workshop on Statistical Learning Techniques for Solving Systems Problems at NIPS |
State | Published - 2007 |