Databases Reference
In-Depth Information
Both the bottom-up and top-down approaches utilize ETL data stores to efficiently
and accurately process the data. However, the bottom-up approach does not include the
data warehouse data storage as defined here. The primary data storage for a bottom-up
approach is in the next layer
.
Publish the Data: Data Marts
In order to more directly support each business area, a collection of the data
needed for that specific area can be loaded into a data mart. Each data mart
is intended to directly support the needs of a specific business group. It may
be possible to define a semantic layer on top of the data warehouse to provide
the data mart view. If so, the data is not physically moved from the data
warehouse. The data architecture would reflect this here. It is also common
for data to be physically moved from the data warehouse into data marts.
This requires another layer of extraction, transformation, and loading (ETL)
to move the data from the data warehouse to the data mart(s). The following
describes the characteristics of this physical movement of data. As noted in
Figure 9-4, additional ETL data stores may be needed to support ETL processes
to load data marts. In some cases, the structures of the data warehouse itself
may be modified to support the processes to load and maintain data marts.
Mastering Data Warehouse Design
, by Claudia Imhoff, Nicholas Galemmo, and
Jonathon Geiger (Wiley, 2003), is a good resource to learn more about these
techniques.
1. What data will be stored here (reference and/or transaction data)? The
data needed to support the specific business analysis is to be loaded. Theo-
retically, only summarized data is needed because the detailed transac-
tions are in the data warehouse itself. In reality, all frequently used
transaction data is loaded into the data mart. The reference data needed
to support analysis is also loaded into the data mart.
2. What is the primary purpose of keeping the data here? The data marts
are designed to support specific business groups or analyses.
3. How will the data be structured? The data is usually stored in dimensional
structures in the data mart. The dimensional structures may be physically
stored in a relational database or a multidimensional cube (this is a
characteristic of the technical architecture). Sometimes data is stored
in structures that are specifically tailored to the business application.
For example, a flat file may be created to support a specific insurance
catastrophe modeling software package. If a semantic layer is used, rather
than physically moving the data, then that should be noted here.
4. What is the persistence of the data or how much history will be stored?
The historical data requirement varies according to the specific business
