I'm currently working to build a cube from our dimensional data warehouse.The way the data has been modelled in some cases is as a fact with a one-to-one relationship to the dimension. For instance you have a Transaction fact then a Transaction dimension with a corresponding record for each fact. This results in a dimension with as many rows as the fact.Also these dimension records can be very wide and contain multiple attributes and natural hierarchies, so in an SSAS OLAP cube I would like to break these into separate dimensions, e.g. Transaction Status & Transaction Type etc.This problem is similar to modelling from a single denormalised table.I'm not able to do anything much about the DW architecture.My question is: Is it OK to create multiple dimensions off the same dimension table, just using a subset of the columns for each one and how is performance affected in this scenario. There may be only 5 Transaction Statuses but they are joined to the fact by the Transaction Key so the dimension has a Transaction Key attribute which may have millions of values and then a Status attribute with only five. I'm only interested in the status and would hide the Key.Where will the above issues hit most, processing time or performance? Is there a better way to model this through the DSV or should I be using multiple hierarchies on the same Cube dimension?Many thanks.
↧