Here, I'll introduce queue time classification and service time classification, discussing how these classifications can be used and their data sources. You'll find more details about these topics in Chapter 5.

The objective with time classification is to improve our diagnosis and the ability to communicate the results of our analyses. Use just enough complexity to get the job done. In most cases, four or five queue time classifications are all that are needed. One level up from the wait events themselves, I typically use only three classifications: other, IO reads, and IO writes. If the issue is a combination of similar other category wait events, I'll collapse those into a single "sub-other" category. For example, in Figure 2-14, notice that I broke down the other category into latching and non-latching. With just a few categories, I am able to pinpoint the problem, perform an ORTA, and complete a 3-circle analysis, and I can communicate the results to both technical and nontechnical people. All queue time information was gathered from the Oracle wait event views.

Figures 2-15, 2-16, and 2-17 (shown below) are parts of an instance-level ORTA report based on a 120-second interval. Looking at the report in Figure 2-15, you can see the queue time is classified at different levels. While total queue time is 621 seconds, the next level is simply IO time (615 seconds) and non-IO time (6 seconds). If you work down the report and to the other figures, you'll notice the IO queue time classified as read time (586 seconds) and write time (30 seconds). Finally, all the queue time classifications result in the actual wait events with their associated time. Chapter 5 will give you a deeper understanding of queue time classification, including how being too focused on wait events can lead to a misdiagnosis.