Archives of the TeradataForum

Message Posted: Tue, 13 Jul 2005 @ 00:24:16 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

With R5 I pretty much work off the DBQL rather than the explains, then I just eyeball the bad queries. Most of our queries are 12-15 tables without any nesting, usually you can see several alternative implementations. The ones with really bad performance usually have several fairly obvious issues. I'll put the changes in one at a time and see if it changes the join types (using my query summary tool). You can't control execution at the step level anyway. The most frequent fixes are mismatched data types on where criteria, and changing filters to use the foreign key in fact tables (ie, replace filter on customer.customer_name with sales_data.cust_id).

I'd thought about enhancing my explain summary tool to replace the spool file numbers with concatenated table names, but I never seem to get around to it. Sometimes I'll print out the explain summary report, then mark up the spool file references with different colored pens. Every table gets a different color.

	<Prev	Next>		<<First	<Prev	Next>	Last>>