Session profiling has reached fanatical appeal. I spoke about this topic at a massive Oracle conference a few years ago. Afterward, a person meekly walked up to me. Looking over his shoulders, he quietly said, "Do you know what you just said? You are saying that profiling is not the best approach." He was clearly disturbed. Session profiling had become so personal to him it was like I was questioning this deeply held faith. My response was gentle, and I clearly restated my key point: profiling a session is not always the best approach. And, in fact, it is easy to be misled by reading more into the situation than is actually there. To bring clarity to this sometimes emotional topic, I will first define profiling and then present how to avoid being misled.

Besides all the political DBA industry talk about profiling, software engineers also use the term when analyzing performance. The term can take many different flavors and is applied to many disciplines. For example, the term profiling can be used for software analysis that is event-based, statistics-based, instrumentation-based, and simulation-based. So to limit the term profiling to one particular method of Oracle performance analysis breaks the actual definition of the word.

Unfortunately, when many DBAs hear profiling, they think of tracing an Oracle session and categorizing the time. However, based on the industry-accepted definition of profiling, analyses based on an Oracle trace, an operating system trace, Oracle's performance views, and sampling directly from the SGA are all profiling. So there are many ways to profile an Oracle system, an individual session, or a group of Oracle sessions.