Figure 4-6. Graphically, on a two-CPU core system, queuing doesn't look like an issue until around 65% busy. It's more difficult to make a risk assessment because the response-time curve does not have such a dramatic elbow as systems with more CPU cores.

DBAs who have control over the workload entering their system have been known to run the CPUs as busy right up to the beginning of the elbow of the curve. Now, not many administrators have this kind of workload control. But if your system runs a combination of online transaction processing (OLTP) and batch processes, then it's likely you can play this game yourself. Let's look at how.

If your system is more of an OLTP system or more batch-centric, you have some creative choices to maximize your user's experience. For an OLTP system, users want snappy response time. They really don't care about how much work the system is processing, but just want the system to be responsive to their requests. On the other hand, if the focus is on batch processing, responsiveness takes a backseat to throughput. For example, if there are 12 CPU cores along with a long batch process queue, having a CPU idle is something we don't want to see. Let's take this a step further.