Modeling junk dimensions

As we've already seen, junk dimensions are built from groups of attributes that don't belong on any other dimension, generally columns from fact tables that represent flags or status indicators. When designing an Analysis Services solution, it can be quite tempting to turn each of these columns into their own dimension, having just one attribute, but from a manageability and usability point of view, creating a single junk dimension is preferable to cluttering up your cube with lots of rarely-used dimensions. Creating a junk dimension can be important for query performance too. Typically, when creating a junk dimension, we create a dimension table containing only the combinations of attribute values that actually exist in the fact table—usually a much smaller number of combinations than the theoretical maximum, because there are often dependencies between these attributes, and knowing these combinations in advance can greatly improve the performance of MDX queries...