We are trying to create a Tabular model for a client. The client has created a single, de-normalized table in Hive. All of the measures and dimensions are in that one table. This table will be the source for our Tabular model.
Here is this question: should we have a single-table Tabular model like the source, or should we split the data out into a star schema and have several tables in our Tabular model? Which approach will yield better performance?
By the way, that single table has over 2 BILLION rows.