Programming pearls: perspective on performance

TL;DR: This column is about the opportunities for efficiency at various design levels of a computer system: each technique is applicable to one of several design levels at which computer systems are built.

Abstract: There are many ways to make a program run faster. Previous columns have discussed such techniques as problem definition, algorithms, data structures, and code tuning. In this column we'll see how those parts fit together into a whole: each technique is applicable to one of several design levels at which computer systems are built. There are at least two good reasons for programmers to study run-time efficiency. The first is pragmatic: while a program's performance is often unimportant, every now and then it matters a great deal. The second reason for studying performance is educational. While many qualities of software are as important as efficiency (if not more so), they are harder to discuss. Anybody can measure that a program is five times faster than another or uses half as much space, but I have to resort to more subjective terms to describe functionality, maintainability, or robustness. Because performance can be quantified and is intellectually challenging, it's a fine training ground for software engineers. This column is about the opportunities for efficiency at various design levels of a computer system. We'll first study one particular program, and then turn to a more systematic view of design levels.