Most of you reading this book will have access to your specific vendor's performance tools. For example, HP's tool is tusc (my guess is Trace Unix System Calls). With AIX and Solaris, you have truss (my guess is Trace Unix System Statistics). If you can find the bottleneck using standard tools, you will certainly be able to do the same thing using tools developed specifically for your platform. So, let's get started!

Of the four subsystems we will dig into, the CPU is probably the easiest to understand. Yet, identifying and understanding CPU contention breeds insights into every part of your work. When you look at the CPU subsystem with a response-time analysis perspective, everything seems to fall very neatly into place. Service time, queue time, and the run queue all make sense, and you can see how this translates into the Oracle system. And you'll also find that the basic concepts of queuing theory are blatantly exhibited. It's like a living science experiment!

Models are a wonderful way to distill something very complex into something we can quickly understand. The CPU subsystem model shown in Figure 4-1 provides surprising insights.