Authors: Peter Struss, Vikas Shivashankar, Mohamed Zahoor
The paper describes a model-based approach to developing a general tool for localizing faults in applications of data warehouse technology. A model of the application is configured from a library of generic models of standard (types of) modules and exploited by a consistency-based diagnosis algorithm, originally used for diagnosing physical devices. Observing intermediate results can require high efforts or even be impossible, which limits the discriminability between different faults in a sequence of data processing steps. To compensate for this, fault models are used. This becomes a feasible solution for standard modules of a data warehouse application along with a stratification of the data. Fault models capture the potential impact of faults of process steps and data transfer on the data strata as well as on sets of data. Reflecting the nature of the initial symptoms and of the potential checks, these descriptions are stated at a qualitative level. The solution has been validated in customer report generation of a provider of mobile phone services.