Database Reference
In-Depth Information
Figure 12-16. Partition storage design categories
The partition data leaf-level values are the row values within the data warehouse tables. They are normally
copied from the data warehouse into the SSAS database folder. Aggregated total and subtotal values are created
by SSAS, and are normally stored within the SSAS database folder as well, but are created when the cube is
processed. Partition metadata is always stored by SSAS in the database folder and is created when the cube is
deployed and then updated when the cube is processed.
By default all three data categories are stored in the SSAS database folder. This particular storage design
is referred to as MOLAP for short and stands for Multidimensional Online Analytical Processing. That is quite
a mouthful for a name that is somewhat meaningless. It simply indicates that you are storing all the data into
folders managed by a multidimensional OLAP cube server such as SSAS.
It is also possible to store some partition data elsewhere. For example, you can choose to keep the leaf-level
values within the tables of the data warehouse and store only the aggregate values in the SSAS database folder.
When this is done, the configuration is considered a hybrid approach, or HOLAP for short (Figure 12-16 ).
A third possibility is to store both the leaf-level values and the aggregate values within the relational
database. This is known as a ROLAP configuration (Figure 12-16 ).
As we mentioned, the SSAS database metadata will always be stored within the database folder regardless of
your configuration choices.
We recommend using a MOLAP design for most situations (which is the default). This design gives you the
best report performance at the cost of storing a copy of all the leaf-level values within the SSAS database folder.
In other words, if you have 6 million rows of data in your data warehouse tables, you are going to have to store the
equivalent of 6 million rows of data in the SSAS database folder.
ROLAP and HOLAP
SSAS is very good about compressing leaf-level data, so for most situations, this is still considered a good option.
In those rare cases where having two copies of the data is just too much of a strain on your environment, you can
choose to use either a hybrid OLAP or a relational OLAP design (HOLAP and ROLAP).
Another occasion where a ROLAP design may be convenient is when report data needs to be as real time as
possible. For example, consider a situation where a cube is built on top of an OLTP database that tracks patient
 
Search WWH ::




Custom Search