Archives of the TeradataForum

Message Posted: Wed, 01 Oct 2003 @ 15:23:27 GMT

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

Hi Tony,

Just to add to Clay's comments on query log in V2R5, our group set up some simple tests to measure the overhead on query throughput running a session of simple, short, all-AMP queries (2 second stand-alone times). We found no reduction in throughput, even with the most extreme level of logging defined for these queries, so I would expect query log to be invisible for the vast majority of data warehouse activity.

However, with single-AMP, primary-index access queries we found about 6% overhead with default logging enabled (log to DBQLogTbl only), 1.6% overhead with summary logging only, and about 15% with the ALL option (log to default, SQL, Step and Object tables). I am guessing that if you have an application composed of tuned single-AMP or group AMP queries, summary logging is likely to be all you need, once you move out of the test environment.

Even running TPump with logging enabled we saw no performance impact, because query log writes one row to the default log table per TPump buffer, NOT per TPump row inserted...the macro's multi-statement request is the "query".

Because query log was implemented to optimize the insert-to-disk processing, (primary indexes on the query log tables were selected based on quick, batched-up inserts), there is some effort involved in the access and joining end, as illustrated in the Clay's email. The more you can keep the tables maintained and reasonably-sized, the better-performing the analysis end should be.

Thanks, --Carrie

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