Using DTA against a "build" instance of the DB | SQL Server Performance Forums

SQL Server Performance Forum – Threads Archive

Using DTA against a "build" instance of the DB

Hey all, we’re looking at using the Database Tuning Advisor to suggest tuning improvements on a nightly basis against the latest database produced by the automated build process. It looks like you can run DTA against a test server and it will automagically get the production server’s metadata, statistics and hardware configuration. The problem is that, although we need the production server’s statistics and hardware configuration, we want to run the load against the latest development version of the database. I’m guessing that we won’t be able to "get there from here". I’m betting that statistics must map to a specific set of tables, indexes, etc. Any attempt to bring statistics into a different physical structure would be fruitless, probably foolish, if not downright impossible. Appreciate your input, -=michael=-
So, it just occurred to me that this might come under the heading of a "what-if" analysis, which appears to be supported by the tool. See:http://msdn2.microsoft.com/en-us/library/ms190997.aspx. In this case, the "what-if" analysis is simply: "If we deployed the current development database, would it break the current production workload?" The trick would be to update a tuning results .xml file produced via the GUI with the physical database configuration of the development database. -=michael=-
quote:Originally posted by mseankelly Hey all, we’re looking at using the Database Tuning Advisor to suggest tuning improvements on a nightly basis against the latest database produced by the automated build process. It looks like you can run DTA against a test server and it will automagically get the production server’s metadata, statistics and hardware configuration. The problem is that, although we need the production server’s statistics and hardware configuration, we want to run the load against the latest development version of the database. I’m guessing that we won’t be able to "get there from here". I’m betting that statistics must map to a specific set of tables, indexes, etc. Any attempt to bring statistics into a different physical structure would be fruitless, probably foolish, if not downright impossible. Appreciate your input, -=michael=-

]]>