Toward Automated Diagnosis and Forecasting of Performance Problems in Enterprise IT Infrastructures: A Pattern Recognition Approach

Moises Goldszmidt (Hewlett-Packard Labs)

There are various offerings of commercial products to monitor and collect measurements in large-scale enterprise (IT) systems.  These products are capable of aggregating these measurements and presenting them graphically to the user.  Yet, in spite of these tools, it is widely recognized that the complexity of deployed systems surpasses the ability of humans to diagnose and respond to problems rapidly and correctly.  In this talk I will describe an approach based on pattern recognition and probabilistic modeling for filling this gap, by mining, analyzing, and transforming measurement data into information. The objective is to use this information to enable the efficient and (semi-)automated management of these systems.  I will also present results on the application of this approach to the diagnosis and forecasting of performance problems.  Finally, I will describe some of the research challenges that these approaches present in general, and specifically in the context of their incorporation into a commercial product for enterprise systems monitoring and performance management.