|
|
Archives of the TeradataForum
Message Posted: Thu, 11 Aug 2005 @ 17:12:20 GMT
Subj: | | Re: Primary Index vs Single Condition |
|
From: | | Stover, Terry |
My first guess is that the data is skewed, the join's not done until the last amp is done even if all the others are idle. I'm not sure why
the join approach is sorting the spool, seems like a PI join should be in rowhash order already.
I've always found moving the filters to the large transaction table is a much faster solution, other than when the join column is a highly
selective nusi (>99%). Heuristically that makes sense because the non-qualifying rows are discarded when the table is first scanned instead of
joining all the rows then discarding the non-qualifying rows. If the header TRANSACTION_NO is a UPI you might try using soft RI.
| |