performance

I posted my GoldenGate / Gnuplot lag time alerting shell scripts to a new GitHub repository: http://github.com/dmann99/ggtools

The scripts worked on v8 and I don't think the INFO ALL screen has changed much since then so hoping it will work on v11 with a minimum of changes. I will post a more comprehensive walkthrough after I verify the v11 compatibility...

I also have an Alerting perl script that I will be tidying up and including in the ggtools repository ... but that won't be available until later this month.

If you have a database where querying against v$ Performance views is painful, consider gathering Fixed Object Stats.

Fixed objects are the 700+ X$ tables that feed a lot of these system and performance views.

Fixed Object Stats are _not_ gathered by the default stats gathering procedures. They are not gathered at all unless you run the proc to gather them. In this case Oracle relies heavily on Dynamic Sampling… but that isn’t really a substitute for really gathering stats.

Here are two runs of the same SQL from different programs. The top is an automated process, the bottom is Toad. (SQL is the same, but SQL_ID is different because of white space issues):

So why is one finished in 7.2 minutes and one still at 4.7 hours and still counting up?

I recently had an issue where a query being submitted to the database had a large change in performance. I could attribute the change to a recent change in object statistics, but the query wasn't behaving because of a combination of a view, subquery, UNION, and LIKE operators. After spending too much time on this statement I decided to hint it to make it perform like it did before the most recent stats change.