Edit: But seriously, does anyone know if this column-oriented stuff is really revolutionary, and if the article is correct in praising the technology?
Edit: But seriously, does anyone know if this column-oriented stuff is really revolutionary, and if the article is correct in praising the technology?
http://en.wikipedia.org/wiki/Sybase_IQ has more details on Sybase IQ.
Since a single datamart might have 200 columns spread across dimension and fact tables, and since this query is only touching 5-6 of them, it pays to make it easy to get to individual columns, instead of reading entire rows. In a row-store database you'd have to read the main fact table rows and the Employee and Office dimension rows, and that might add up to 50 columns. If you can only touch the exact columns you need you're better off.
All that said, I've never done it. I really need to get an open source column store and whack at it for a bit.
What changes is the physical storage of the data. Instead of a disk read picking up all the columns including unused ones, a given disk read gets many values for a single column.