When to Tune

---

When should you tune? Constantly and throughout the life cycle of your product, perhaps only when performance is a problem. On the other hand, the database should be configured and organized properly in the first place to avoid continuously putting out fires. If you are always putting out fires the chances are you will make little progress with anything else. It is much better to be proactive and preempt potential problems rather than to react when they occur. It is preferable to tune your database during development rather than after development in production.

Tuning in development will lengthen the life cycle of any system. Why is this the case? If different tasks can be compartmentalized during distinctly separate phases of a project, such that those phases do not overlap each other, a better product will result. For instance, any developer knows that changing code in order to tune SQL code, after completion of development, could change the integrity and neatness of that code. This may not seem too important. However, the more times a piece of code is altered the more it will deteriorate, not only because the code is changing but because different coders may be changing that code. Every coder has a different style and approach and quite often subsequent coding is confused by code written by other people and vice versa. Errors can be introduced into application code when it is changed. The more changes the more potential errors.

We do not live in a perfect world. Distinct lines cannot be drawn between development and production. Changing requirements often cause changes in different project phases making gray areas. There is often overlap between development and production.

It is best to tune in development, particularly data models and SQL code. When tuning after development in a post-development requirements change phase, or in production, tuning should be done carefully, especially when it comes to changing application code. If tuning cannot be done fully in the development cycle, which it probably cannot, take the following approach when tuning in production:

• Set performance targets.

• Use test environments.

• Use test environments matching production as closely as possible.

• Tune with care.

---