 |
 |
Archives of the TeradataForum
Message Posted: Wed, 18 Oct 2000 @ 14:11:24 GMT
| Subj: | | Re: Star Schema |
| |
| From: | | Alan Friel |
Steve Bracewell wrote:
| | There are a couple of methods we use. One is an old tried and true NCR solution, not pretty, but effective. By adding a
meaningless column to the dimension tables and using syntax like below, you can 'force' the product join of the smaller
tables. | |
| | AND (T.Meaningless_column = P.meaningless_column
OR T.meaningless_column <> P.meaningless_column)
AND (T.meaningless_column = G.meaningless_column
OR T.Meaningless_column <> G.Meaningless_column) | |
That is really old fashion and orginiated about Teradata Version 2 (probably about the late 1980's). The parser and
optimizer have made real improvements since that time and using such a trick should be unnecessary.
In the best case it doesn't do anything better than what the parser would do on its own. In the worst case, doesn't it prevent
the parser and optimizer from developing the best work plan? I guess it would be OK if you had a very narrow range of handcrafted
queries, but if we're going to open our databases to our users, shouldn't we expect more flexibility and that the parser should be
able to create a good plan without resorting to tricks?
| |
