Archives of the TeradataForum
Message Posted: Thu, 21 Mar 2002 @ 17:38:02 GMT
| Subj: || || Re: Large Table Join |
| From: || || Geoffrey Rommel |
| ||I need to join two large tables. Will I gain any speed if I break one or both tables into smaller tables. For example, instead of
join table A & B directly, I will do A(id <=50%) & B; then A(id > 50%) & B. Then combine the results.|| |
Since no one else has replied, I, though unworthy, shall give it a go. I think this would be a bad idea. Keep in mind that the
biggest time consumer in just about any query is the I/O operations. Most likely, this approach would result in two full-table scans of the
A table, which are going to take longer than one scan.
I can't say more without more information. For instance, are these tables hashed on the column(s) you are joining on? Can you redefine
them with different primary indexes in order to co-locate the rows to be joined, or would that be unacceptable to other users of these
tables? Just how large are they? What kind of response time are you getting now, and what would be acceptable? Do you have plenty of spool
space? You get the idea....