Archives of the TeradataForum
Message Posted: Mon, 15 Oct 2001 @ 17:32:33 GMT
All programmers who develop optimization software share a unique curse. Every change they make, no matter how beneficial it is to the vast majority of their users, is certain to severely penalize the performance at least one user. And while testing using a representative workload can validate the new algorithm, the real world is vastly more complex and you can't test for it.
That said, I would still open an incident with NCR to let them know that you have a 100:1 performance problem caused by collecting statistics. They'll want to see the DML, DDL & demographics on all the objects, but it may reveal some kind of end case where the optimization costs are not properly calculated.
|Copyright 2016 - All Rights Reserved|
|Last Modified: 28 Jun 2020|