There’s a new blog about Dimensional Data Warehouse Architecture & Design written by Nick Galemmo, and his recent entries caught the eye of one of my coworkers. Galemmo asked “What on Earth is a Data Mart?” and comes to the answer:

“The Great Truth in Dimensional Data Warehousing boils down to this: To achieve success in building an integrated dimensional data warehouse, build a series of event specific atomic data marts. Don’t design or even worry about aggregates, those will fall into place later. Just make sure you are collecting the right events.”

I can definitely vouch for that. During the initial design phase of an atomic mart, look closely at each dimension and make sure it’s the most low-level dimension possible. For example, when building a sales data mart, if the lowest level of salesperson dimension is Sales Team #, then it’s not atomic enough. The mart needs to include the exact salesperson, if it’s available in the source. Once the data mart is in place, it’s expensive and time-consuming to revise the atomicity of individual dimensions. You can go higher, but you can’t go lower.