Creating data marts from data warehouse : Architecture Design considerations

In a typical BI architecture, the regular layers that you find in an architecture diagram are source systems, ETLs, data warehouse and reporting layers. But whether to encompass data marts into your architecture is one such design decision that demands some convincing reasons.

Below are a few scenarios when you might want to consider creating data marts in your architecture design.

1) Customization for business units: Different business units of an organization can need their own version, shape and volume of data which would be originating from a set of common source systems. Customization to this degree is not possible at the data warehouse layer, so independent data marts can be created to cater this requirement.

2) Performance Optimization: A single cube / sets of cubes created out of a single data warehouse and sourcing data from the same data warehouse can be real challenge to performance. By creating data marts you can divide the load depending on the user base and corresponding volumes of data access.

3) Detailed What-If Analysis: Data needs to be manipulated for what-if analysis and for the same it might require a write-back to your underlying data. This is not possible when you have a lot of users who would be accessing the same data warehouse. This can be very well catered by creating a data mart for this requirement.

4) Isolating data discovery related initiatives: Lots of research and development related activities needs to be carried out on a OLAP system for intelligent data discovery like predictive analysis, adjusting the data model to use with advanced analytical visualizations, data mining, forecasting and budgeting activities by applying external data to your data in DW. Such RnD are safe to carry out on an isolated environment, and data mart can be one perfect solution for this requirement.

Creating a Data Mart is not free of efforts. It requires additional ETLs, additional space, additional maintenance overheads. But considering the business value it brings to the table, it is worth creating data marts in certain scenarios. Above list is not an isolated list of scenarios, but in my experience, these has been the prominent ones. Feel free to share your experience with me on the same lines.