I've been reading the Handbook of Massive Data Sets. The article on data cubes generally gave me the impression that one just sort of makes a great, big data cube of the data warehouse and you go to town, viewing various aggregates of data. Literature always emphasises great, big, and often great, big doesn't translate into useful.

The problem with great, big data cubes isn't so much that they exceed the capabilities of the tools for viewing them, but that they take too long to analyze. I tend to set myself up for a successful analysis by planning ahead about what data I intend to load into the cube.

Lately I've been using Data Beacon, and it doesn't do great, big cubes, but is very easy to use. Instead of focusing on size and spending all their time with elaborate caching mechanisms, they've focused on making it easy to turn reports into cubes and then export those reports into Excel or a statistics package.

When I build data cubes, I am looking at some specific portion of the data warehouse and I don't want to have to drill down through countless aggregate layers. I save a lot time by thinking carefully about what data to include in the cube.