It has been argued that the data warehouse should be implemented as a fully normalized
relational model, based directly on the enterprise data model, with no summary
data, postponing any denormalization and aggregation to the data marts loaded
from the warehouse (e.g., Moody and Kortink 2000). Current practice however usually
does incorporate denormalized and summarized data in the data warehouse (Silverston
et. al. 1997). A nice overview of alternative approaches is provided by Jukik (2006).