Measure what is measurable, and make measurable what is not so.
Galileo Galilei (1564–1642)
Measurement and experimentation are the basis of any serious attempt to improve a program’s performance. This chapter introduces two software tools that measure performance: the profiler and the software timer. I discuss how to design experiments to measure performance so that their results are meaningful, rather than misleading.
The most basic and most frequently performed measurement of software performance answers the question, “How long?” How long does it take to execute a function? How long to fetch a configuration from disk? How long to start up or shut down?
These questions can be attacked (clumsily) with ridiculously simple instruments. Isaac Newton measured the gravitational constant by timing falling objects against his heartbeat. I’m sure every developer has informally timed activities by counting out loud; in the United States, we count “one-Mississippi, two-Mississippi, three-Mississippi...” to get approximate seconds. A digital wristwatch with a stopwatch function was once de rigueur for computer geeks, and not just for the fashion statement. In the embedded world, hardware-savvy developers have nifty instruments at their disposal, including frequency counters and oscilloscopes that can time even short routines accurately. Software vendors sell a variety of specialized tools too numerous to review here.
This chapter focuses on two tools that ...