Students and clients frequently ask me how I know when to stop improving performance. The answer, while not providing the warm, fuzzy feeling people may want, is straightforward. I keep working on improving performance until there is no time remaining, no budget remaining, or performance is acceptable. This is a very simple statement, but obviously there is much more to it.

At some point, it just doesn't make sense to keep chipping away at performance. Diminishing returns are always a factor. For example, the first tuning cycle may improve performance by 50%, the second by an additional 10%, and the next by 5%. Should we keep tuning? Perhaps other options should be considered, such as increasing system capacity or altering the workload. Regardless of how you disguise the situation, it always comes down to time, money, and performance.

Since you are periodically collecting key business process timing information, gathering the final baseline sample should be trivial. Still, write it down and publish it. Let people know how awesome a job you and your team have done! If performance went from a query taking 12 hours to that query running in 2 seconds, tell them. You don't need to be an egomaniac about it, but at the same time, people need to know they can count on you in the future. Use this opportunity to further instill confidence and hope.