Gustav Brock
gustav at cactus.dk
Tue Feb 3 13:42:30 CST 2004
Hi Robert > What I think he is saying is that in 99% of the cases, the regular date > dimension table will work. In a few cases, where the date may be missing > and you want to get some kind of value for it, you cannot use the date > itself as the key, thus the surrogate. The surrogate would be stored in > the fact tables (another data warehousing concept, for those that do not > know about them). That way when you hit an invalid or null date value, the > date dimension table would be able to handle it in a predefined way. I noticed this too and wondered. How can one "hit an invalid date"? Do dataware house people not validate data before storing them? A missing date may, of course, be accepted, but an invalid? Where would that come from? /gustav >>I'm not even sure what Ralph is saying there, and it flies in the face >>of what he does in some of his books, so I have no answer for you, John. >>Keep in mind, though, that datawarehouses are indexed somewhat >>differently from transactional databases, so that may be where the >>confusion is coming in. If you were using surrogate keys and >>meaningless values as the PK, the dates in the main records would need >>to be inserted *from* the date table, and that isn't generally the way >>the thing is used. >> >>Charlotte Foust