Skip to Main Content
Configuration management is a complex task, even for experienced system administrators, which makes self-managing systems a desirable solution. Self-management implies the need for a model based on which configuration changes may be decided. In previous work, we described a method for constructing a state-transition model of application behavior, by observing the application in simulation. This method relied on an expert to manage the (simulated) application in order to collect the necessary observations for constructing the model. However, that method was agnostic about (a) the size of the system space space as implied by the granularity of the observations, and (b) the sufficiency of the actual observations collected for understanding the application in a variety of configurations and environments. In this paper, we replace the (expensive) expert domain knowledge with automatic approaches to ensuring coverage of the application, and demonstrate the superiority of this approach. We present empirical data regarding state space and granularity to explore the use of state models for understanding applications.