In the world of database performance, normal is a highly prized goal. It means nothing extraordinary is happening that requires intervention or diagnosis. It implies no fire-fighting or frantic finger-pointing exercises across functions in the IT department. So, in this case, healthy is the goal. Typical performance means normal service levels and minimal drama from dissatisfied users and customers. However, to be measured and monitored, normal must first be defined and measured. That is sometimes more easily said than done.

If you have attended any best practice presentations or courses for database administrators over the years, the presented undoubtedly told that the first step in measuring and monitoring performance is to establish baselines. The trick is that different people throw that term around without definition while meaning very different things. So I want to discuss some of those meanings and then explain what it means in the context of monitoring tools. Moreover, even in that isolated context of monitoring tools, you need to understand how different vendors use the same term.

This whitepaper describes what baselines are, how to detect anomalies, baseline approaches, whether normal means Gaussian, and how to use calculated baselines.