Archives of the TeradataForum
Message Posted: Fri, 25 Oct 2002 @ 13:54:59 GMT
We do this by means of a small system that logs the start time, executes the query, and then captures the end time along with the delta resusage information (cpu, io). The user running the query can't be doing multiple things at once, or it will distort the captured metrics. We can judge improvements to a query (or relevant parts of the db, indexes, etc) by evaluating what happens to the actual resource utilization.
We use a similar methodology at a system wide level also, to calculate average cpu and io usage for queries run by user profiles (i.e. Business Object users, Campaign Managmenet user, developers, etc.
This allows us to evaluate workloads against each other to determine which groups may be running inefficient queries. It also allows us to trend profile usage over time, to look for negative trends.
As others have mentioned, wall-clock time is just not an accurrate method due to mixed workloads during the timeframes. However, we do also use a canary query (runs every half-hour) to give us an impression of what the end users are experiencing due to workload trends.
David W. Roth
|Copyright 2016 - All Rights Reserved|
|Last Modified: 28 Jun 2020|