If there’s one big religious divide within the BI world, greater than the differences between vendors, then the question of whether to use an OLAP database or to query the relational database directly is it. Standing in the pro-OLAP camp as I do I probably have more in common with an Essbase guy than someone who wants to do BI with the SQL Server 2005 relational database exclusively. Anyway, here’s an article by Ralph Kimball in Intelligent Enterprise that sums up the arguments on both sides pretty well:
I’m pleased to see that he makes much of MDX as a plus-point on the side of OLAP; even Oracle’s supposedly all-SQL approach begins to look a bit MDXy when you see the details (see for example Mark Rittmann’s post, also from today: http://www.rittmanmead.com/2007/04/30/obi-ee-time-dimensions-and-time-series-calculations/) although from what I’ve seen MDX still has some important advantages. It’s just a pity that as a language it seems to be fragmenting into different vendor-specific implementations so quickly – for example, see this post from the Panorama blog on SAP BW MDX: http://www.panorama.com/blog/?p=44.