Using TinyInt as keys in Analysis Services

Fact tables should be narrow and deep.  If you are recording every debit card transaction for every customer across the fifteen retail banks you own, you end up with 15 million rows a day, so to save space they need to be as thin as possible.  If one of your dimensions can only ever have a small number of values then use tinyint as the key to this dimension.  Actually I always use tinyint instead of boolean for yes/no male/female stuff because you often need to have three or four values i.e. to include ‘not applicable’, ‘unknown’ etc. and I also break with Kimball and use special keys for these such as 0.