The previous chapters have discussed what performance metrics may be useful for the performance analyst, how to summarize measured data, and how to understand and quantify the systematic and random errors that affect our measurements. Now that we know what to do with our measured values, this chapter presents several tools and techniques for actually measuring the values we desire.

The focus of this chapter is on fundamental measurement concepts. The goal is not to teach you how to use specific measurement tools, but, rather, to help you understand the strengths and limitations of the various measurement techniques. By the end of this chapter, you should be able to select an appropriate measurement technique to determine the value of a desired performance metric. You also should have developed some understanding of the trade-offs involved in using the various types of tools and techniques.

Events and measurement strategies

There are many different types of performance metrics that we may wish to measure. The different strategies for measuring the values of these metrics are typically based around the idea of an event, where an event is some predefined change in the system state. The precise definition of a specific event is up to the performance analyst and depends on the metric being measured. For instance, an event may be defined to be a memory reference, a disk access, a network communication operation, a change in a processor's internal state, or some pattern or combination of other subevents.